{"articles": {"2026-09-01-ig-s-normal-tier-is-not-a-clean-all-clear-while-funding-reads-strain": {"body_md": "## The read\n\nThe most defensible reading of Undertow’s board for 2026-09-01 is deliberately narrow: investment-grade, or IG, liquidity is classified NORMAL, but that label is not a clean all-clear because the qualifying measures within the cell disagree while the separate Seiche funding overlay reads STRAIN.\n\nThat distinction is the point of the day’s evidence. The IG cell has a within-board score of 0.493, with 3 available measures and 3 qualifying measures. It is therefore a scored cell, not an absence rendered as calm. Yet the board also says the cell’s qualifying measures disagree. A tier is an aggregation rule; disagreement is information that the aggregation does not fully display. The appropriate conclusion is not that IG is stressed, nor that the funding regime overrides the IG classification. It is that the NORMAL label describes the board result, while the internal dispersion limits its use as a broad statement about market capacity.\n\nThe daily dispatch headline is instead about CN: coverage moves from 10/19 to 11/19. That is a fresh coverage change, not evidence that CN has reached a particular liquidity tier. UST, HY, EQUITY, ETF, FX, CN, CRYPTO, and BSTOCK are still accruing history. Only 1 of 9 segments scores today. That sparse scoring surface is a methodological restraint: incomplete histories do not support a complete cross-market map.\n\nThe funding observation belongs beside, not inside, the IG result. Seiche declares STRAIN for its funding overlay. Its disclosed composite index is 45.6/100, with 100% coverage, but it is explicitly a calibrated weighted blend of engine sub-scores rather than a percentile of its own history. Its stress percentile is withheld rather than inferred. Undertow’s IG tier and Seiche’s funding regime therefore cannot be combined into a blended number or treated as competing grades on a single scale. They are separate readings of separate mechanisms.\n\n## The mechanism\n\nLiquidity is not simply the presence of a market price. In credit, risk transfer can depend on intermediaries’ capacity to hold inventory, finance positions, hedge exposures, and bridge the interval between one client’s sale and another client’s purchase. A funding regime can matter to that process without dictating every cash-credit measure at the same time.\n\nThat is why a NORMAL IG label can coexist with a STRAIN funding overlay without establishing a contradiction. The IG board is a deterministic derivation from qualifying measures. The funding overlay is declared separately and is contextual only; it is not used in the tier calculation. One reading indicates that the IG measures meeting the board’s conditions aggregate to NORMAL. The other indicates strain in the funding framework. Neither statement, on the supplied evidence, establishes the direction of future spreads, returns, issuance, or defaults.\n\nThe disclosed IG evidence illustrates why the cell requires qualification. IG OAS, identified as BAMLC0A0CM, has a stress percentile of 0.2672 as of 2026-08-31. Its stated orientation is that a wider investment-grade spread means costlier credit liquidity and more stress. But the measure’s caveat is material: the FRED series is capped to a trailing approximately 3-year window for keyless ICE BofA downloads, omitting earlier crisis regimes including 2008 and 2020. The percentile may therefore understate rarity in calm tapes.\n\nThe supplied IG measure list also displays dealer corporate net positions, described as a congestion proxy, as of 2026-08-19. The stated logic is that high net dealer inventory can signal balance-sheet congestion and less capacity to absorb customer selling. But this is a proxy for IG, not IG alone: the corporate-securities series spans commercial paper, investment-grade bonds, and below-investment-grade bonds. Its levels are spliced across New York Fed series breaks, and its stress percentile is withheld. Those caveats do not negate the measure; they define its evidentiary boundary.\n\nThe mechanism is therefore not a claim of an imminent break. It is a conditional interpretation: funding strain could matter more for IG liquidity if it constrains intermediary balance sheets or the financing of risk transfer, particularly if qualifying IG measures cease to disagree and begin to align. The present board does not say that has happened.\n\n## The countercase\n\nThe strongest countercase is straightforward. The NORMAL IG tier may be the more informative result, while the funding overlay may describe pressure that is contained, temporary, or not transmitting into the qualifying IG liquidity measures. The supplied evidence does not identify which explanation, if any, is operating.\n\nA second countercase concerns construction rather than markets. The board states that qualifying measures disagree, and it withholds spans where publication would hand back a percentile for a row the free board withholds. The public record supports the existence of disagreement, but not reconstruction of every underlying contribution. Converting withheld detail into a claimed stress reading would manufacture precision.\n\nThere is also a timing boundary. The disclosed IG OAS observation is dated 2026-08-31, the dealer-position observation is dated 2026-08-19, and the funding measure is as of 2026-09-01. These inputs have different stated as-of dates. The dossier does not establish a same-clock causal sequence between them. The comparison is useful as a comparison, not as proof that one reading caused another.\n\n## What would falsify this\n\nThis article’s thesis would be weakened by later board evidence showing that IG’s qualifying measures no longer disagree while the tier remains NORMAL and coverage remains sufficient under the published rules. That would make the present qualification less central by showing a more internally coherent NORMAL result.\n\nIt would also be weakened if a subsequent disclosed funding regime no longer reads STRAIN while IG remains NORMAL. The cross-desk contrast motivating this analysis would then have narrowed. Conversely, a change in the IG tier alone would not prove a funding-transmission mechanism, because the dossier supplies no causal test tying the two products together.\n\nThe thesis would be most directly falsified if the stated internal disagreement were shown to be erroneous in the published point-in-time board record. The relevant evidence is the Undertow daily dispatch and board record, not a retrospective narrative. The daily reading is not attached to a sealed call and must not be promoted into a forward prediction.\n\n## Method and limits\n\nThis analysis uses the supplied public Undertow dispatch and disclosed evidence envelope for 2026-09-01. It does not calculate a new score, infer withheld percentiles, or merge Undertow and Seiche into a composite. Undertow’s board derivation is not an observed price and not a trade signal.\n\nPARTIAL means history is still accruing. A scored tier requires at least 2 measures with a stress percentile at 60+ observations each; an accruing or absent measure is disclosed rather than rendered as calm. The dispatch further states that every scored cell with qualifying-measure disagreement has its span at Agent/Desk tier because publication of that span would reveal a withheld percentile. This analysis preserves that withholding.\n\nCN’s movement from 10/19 to 11/19 is a coverage development, not a CN market verdict. The dossier records replay generation but supplies no historical verdict. Its adjudication is explicitly not attached to this daily reading. Research and market data only; not investment advice.\n\n## Sources\n\n- [Undertow daily dispatch record](https://liquilens-undertow.com/dispatch/2026-09-01.json)\n- [Undertow](https://liquilens-undertow.com)\n- [Undertow data interface](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-01-ig-s-normal-tier-is-not-a-clean-all-clear-while-funding-reads-strain/", "date": "2026-09-01", "dek": "The board’s narrow message is not that investment-grade liquidity has broken. It is that a NORMAL IG tier, built on qualifying but internally disagreeing measures, should remain separate from a STRAIN funding overlay—and from the fresh CN coverage expansion.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:b7d58fa7c7130067e800cf046490a554697c06b1ff5f0a636df73e9db9c6b0e1", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:3b1bb8b903f61835cfc4d62fb3086f216570eb9c6afc53b1a94718d7c26888ae", "status": "applied"}, "evidence_asof": "2026-09-01", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "d0832c6b74ec83b85576bb48e667311ec6cf75a3a759aed2d761693f6d10b088", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:b7d58fa7c7130067e800cf046490a554697c06b1ff5f0a636df73e9db9c6b0e1", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:3b1bb8b903f61835cfc4d62fb3086f216570eb9c6afc53b1a94718d7c26888ae", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The Sources section reaches six links only by repeating the Undertow homepage. That is padded sourcing, not six distinct approved references.", "The statement that “the replay material is not attached to this daily reading” is not supported by the dossier. The dossier says the adjudication is not attached to this daily reading; it separately records replay generation and provides no historical verdict.", "Repairs below replace the duplicate source with distinct dossier-grounded approved-domain records and correct the adjudication/replay boundary."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "IG’s NORMAL tier is not a clean all-clear while funding reads STRAIN", "published_at": "2026-09-01T21:53:49Z", "quality": {"failures": [], "numeric_tokens_checked": 20, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1203}, "slug": "2026-09-01-ig-s-normal-tier-is-not-a-clean-all-clear-while-funding-reads-strain", "topic": "IG", "type": "current_analysis", "word_count": 1203}, "2026-09-02-funding-shifts-to-erosion-while-ig-remains-normal-amid-disclosed-disagreement": {"body_md": "## The read\n\nUndertow’s 2026-09-02 board produces a narrow, qualified reading: IG is classified **NORMAL**, while the separate Seiche funding overlay is declared **EROSION**, after **STRAIN** in the prior reading. The board also discloses disagreement among qualifying measures in at least one scored cell. Because IG is the only scored segment today, that disclosure applies to the IG reading.\n\nIG has a within-board score of **0.5145**, with **3** available measures and **3** qualifying measures. Those inputs support a scored tier under the board’s method, and the published tier is NORMAL. That is not a finding that all IG liquidity measures are calm, nor an observed market price or trade signal. Undertow describes a tier as a deterministic board derivation.\n\nThe disagreement disclosure matters because the public board does not provide the underlying span for such a cell. The dispatch says every scored cell with qualifying-measure disagreement is shown at an Agent/Desk-tier span, because publishing the span would reveal a percentile from a row that the free board withholds. The appropriate reading is therefore bounded: the published method produces NORMAL for IG, but the public evidence does not support a claim of uniform conditions across the qualifying measures.\n\nSeiche’s EROSION designation adds funding context without changing that conclusion. Undertow’s evidence braid identifies Seiche as a funding-context overlay only and states that it is not used in the tier. There is no cross-product score, and none is inferred here.\n\n## The mechanism\n\nUndertow’s stated contribution is to compare market measures against each measure’s own accrued point-in-time history. That design avoids treating unlike raw series as interchangeable, but it also means a tier is not a single underlying price or a direct measure of every instrument in a segment.\n\nThe disclosed IG detail illustrates the limit. IG OAS, identified as BAMLC0A0CM, has a stress percentile of **0.3318** as of **2026-09-01**. Its stated direction is that a wider investment-grade spread indicates costlier credit liquidity and more stress. Its caveat is material: the available FRED history is limited to an approximately trailing **3-year** window, omitting earlier crisis regimes including 2008 and 2020 and potentially understating rarity in calm periods.\n\nA dealer corporate net-positions measure is also disclosed, but its stress percentile is withheld. The measure is explicitly a proxy for IG because it covers the broader corporate-credit complex, including CP, IG and below-IG bonds, rather than IG alone. The board also notes series breaks. Its stated interpretation is that high net dealer inventory represents balance-sheet congestion and less capacity to absorb customer selling. The missing percentile cannot be inferred from the NORMAL tier or assigned as the source of the disagreement.\n\nSeiche uses a different construction. Its disclosed composite index is **44.8/100**, with **100%** coverage. It is a calibrated weighted blend of engine sub-scores, with higher readings indicating stress; it is explicitly not a percentile of its own history. Its withheld stress percentile must not be invented or compared numerically with Undertow’s IG score.\n\n## The countercase\n\nThe strongest countercase is that NORMAL is the best available summary of the qualifying IG evidence. The board does not classify IG as stressed. It classifies it as NORMAL, based on **3** qualifying measures. The Seiche overlay is not an input to that calculation, so its shift to EROSION does not mechanically overturn the IG tier.\n\nThe public detail also cannot support a stronger directional claim. One disclosed IG percentile is available, while another disclosed measure has no public percentile. The board’s withholding policy prevents reconstruction of the dispersion among qualifying measures. Neither “IG is deteriorating” nor “IG is uniformly resilient” follows from the disclosed cell.\n\nCoverage elsewhere is also limited. Only **1 of 9** segments scores today; UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. Their unscored status is not evidence of either calm or stress.\n\n## What would falsify this\n\nThis reading would be falsified by a corrected or contradictory public record for the same 2026-09-02 daily board showing that IG was not NORMAL, that the funding overlay was not EROSION, or that the IG cell did not carry the disclosed qualifying-measure disagreement.\n\nIt would also fail if the point-in-time evidence braid showed Seiche was used in the Undertow IG tier rather than as funding context only. The dossier says the opposite: Seiche is an overlay only and is not used in the tier.\n\nLater market moves do not falsify what this daily board reported. Nor does a later board reading convert this point-in-time classification into a forward call. Undertow states that sealed calls are separate verifiable records and that a daily board reading is not silently promoted into one.\n\n## Method and limits\n\nThis article is confined to the supplied public pack for the Undertow dispatch dated **2026-09-02**. That is the research boundary. No LiquiLens result, historical replay conclusion, undisclosed component percentile or cross-product calculation is added.\n\nUndertow states that a scored tier requires at least **2** measures with a stress percentile at **60+** observations each. PARTIAL means still accruing, not calm; accruing or absent measures are disclosed rather than rendered as reassuring. IG is presented as scored with **3** qualifying measures, but the package does not disclose every component percentile.\n\nThe funding overlay and the IG tier remain separate analytical products. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow data interface](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow Telegram board access](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public site reference](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-02-funding-shifts-to-erosion-while-ig-remains-normal-amid-disclosed-disagreement/", "date": "2026-09-02", "dek": "Undertow’s only scored segment is IG, which holds a NORMAL tier despite disagreement among qualifying measures. Seiche’s separate funding overlay has changed from STRAIN to EROSION, but it is context only and is not part of the IG tier.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1380533100d6a0364dba51ca56ecf7e3575212e52f311bed0f49cb8484196b4a", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:ce67002a631f83bfc60a25c34e57fe36067cd189c7c071c4635efc1d1da99602", "status": "applied"}, "evidence_asof": "2026-09-02", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "5406577614f8a765d6d2f5027af0696ef2b16e201c4422d9c9eb644b87ef0f72", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1380533100d6a0364dba51ca56ecf7e3575212e52f311bed0f49cb8484196b4a", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:ce67002a631f83bfc60a25c34e57fe36067cd189c7c071c4635efc1d1da99602", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft adds an unsupported characterization when it says the withheld row is “subscriber-only.” The dossier says only that the free board withholds the row; it does not establish a subscriber product or access arrangement.", "The falsification section is internally inconsistent with the article’s stated point-in-time boundary. Later board evidence cannot weaken or falsify a claim about what the 2026-09-02 public record disclosed; only a correction, revision, or contradictory point-in-time record for that reading could do so.", "Several passages repeat the same methodological caution without adding evidence. The repair tightens the dispatch while retaining the disclosed countercase, falsifier, research boundary, and all required headings."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "Funding shifts to EROSION while IG remains NORMAL amid disclosed disagreement", "published_at": "2026-09-02T21:50:20Z", "quality": {"failures": [], "numeric_tokens_checked": 16, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 919}, "slug": "2026-09-02-funding-shifts-to-erosion-while-ig-remains-normal-amid-disclosed-disagreement", "topic": "IG", "type": "current_analysis", "word_count": 919}, "2026-09-03-ig-remains-normal-as-its-measures-disagree-funding-overlay-reads-erosion": {"body_md": "## The read\n\nUndertow’s 2026-09-03 reading is narrow. Investment-grade credit, or IG, is classified **NORMAL** on the board, but the qualifying measures within that scored cell disagree. Separately, Seiche’s funding overlay is declared **EROSION**. Those are adjacent observations, not a single combined market verdict.\n\nIG’s within-board score is 0.4336. The dossier reports 3 available measures and 3 qualifying measures, with the tier set at NORMAL. The score changed by -0.08 from the prior reading, which Undertow classifies as a fresh within-tier delta. The board also reports disagreement among qualifying measures. NORMAL therefore describes the board’s deterministic tier, not uniformity across every measure used inside the cell.\n\nCoverage is limited. Only 1 of 9 segments scores on this date. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. That prevents any claim of broad cross-market confirmation. The dossier also provides no cross-product score, and its editorial boundary prohibits creating one.\n\nThe funding comparison has an equally important limit. Seiche’s declared regime is EROSION. Its accompanying note reports a composite index of 44.5/100 and 100% coverage, while explicitly stating that the index is a calibrated weighted blend of engine sub-scores, not a percentile of its own history. Undertow’s IG score is instead a within-board derivation based on measures assessed against their own accrued histories. Seiche is funding context only and is not used in the IG tier.\n\nThe evidence-led thesis is therefore limited: the published NORMAL IG tier contains disclosed internal disagreement, while the separate funding overlay reads EROSION. Neither fact permits a blended score, a causal claim, or an investment conclusion.\n\n## The mechanism\n\nUndertow describes a tier as a deterministic board derivation, not an observed price and not a trade signal. Its framework orients segment measures toward stress and evaluates them against their own accrued point-in-time histories. The resulting tier communicates the board classification; the disagreement flag preserves the fact that qualifying measures do not all align within that classification.\n\nThat distinction is material here. A NORMAL tier can coexist with disagreement among qualifying measures without either observation cancelling the other. The tier does not authorize a claim that every component is benign, while the disagreement flag does not authorize reconstruction of component values that the public board withholds.\n\nThe visible IG detail shows why the limits matter. IG OAS, identified as BAMLC0A0CM, has a published stress percentile of 0.3321. Its note defines wider investment-grade spreads as costlier credit liquidity and greater stress. But that same note warns that the FRED download is capped to a trailing roughly 3-year window, excluding earlier crisis regimes including 2008 and 2020. The published percentile is consequently not a universal measure of rarity across all historical regimes.\n\nA second displayed IG measure is dealer corporate net positions, labelled congestion. Its stress percentile is withheld. The dossier says this is a proxy for IG because the New York Fed corporate-securities series spans commercial paper, IG and below-IG bonds rather than IG alone. It also states that high net dealer inventory is treated as balance-sheet congestion and stress, and that levels are spliced across series breaks.\n\nThe dossier reports 3 available and 3 qualifying IG measures, but names and provides measure-level detail for only 2 of them. That disclosure gap is a boundary, not a license to infer an unnamed measure or its contribution. The dispatch further says that when qualifying measures disagree, the span is restricted to Agent/Desk tier because publishing it would reveal a percentile from a row withheld on the free board. This analysis does not estimate that withheld value or reverse-engineer it from the visible 0.3321 percentile.\n\nThe Seiche comparison operates at another layer. The evidence braid calls it a funding-context overlay only and says it is not used in the tier. NORMAL and EROSION therefore cannot be averaged, netted or treated as opposing readings on one shared scale.\n\n## The countercase\n\nThe strongest countercase is that the disagreement carries limited weight. IG remains NORMAL, all 3 available measures qualify, and the score moved easier by 0.08. On that reading, internal dispersion may be ordinary variation inside a stable board category rather than evidence of an emerging liquidity problem.\n\nThe disclosed limits support caution. The OAS percentile is based on a history constrained by the stated trailing-window limitation. Dealer corporate net positions are a broad corporate-credit proxy rather than an IG-only series, include spliced series levels and have a withheld percentile. The dossier establishes that qualifying measures disagree, but it does not disclose the numerical span or provide detailed measure-level information for every reported qualifying measure.\n\nThere is also a countercase to treating the funding overlay as an IG contradiction. Seiche’s EROSION regime is declared under its own methodology, and its 44.5/100 composite is not an own-history percentile. It may be relevant funding context without mechanically describing the construction or state of Undertow’s IG tier. The published record supports keeping the two systems distinct.\n\n## What would falsify this\n\nThe thesis would be falsified by a correction to the dated point-in-time record showing that IG was not NORMAL, did not have a 0.4336 score, did not have 3 available and 3 qualifying measures, or was not marked as disagreeing on 2026-09-03. It would also fail if the published relationship were corrected to show that Seiche’s EROSION overlay was an input to Undertow’s IG tier rather than funding context only.\n\nA subsequent change in tier, score or funding regime would not itself falsify this dated account. It would describe a later board state, not revise the content of this board. Nor could later market outcomes validate a forward call in this article: the dossier says no adjudication is attached to this daily reading, and the sealed-call record is separate and verifiable.\n\nThe appropriate future test is procedural. Later boards can show convergence or further divergence between their separately published readings. They should be compared without retroactively filling withheld values or blending products into a score the dossier does not provide.\n\n## Method and limits\n\nThis article uses only the supplied point-in-time Undertow dispatch dated 2026-09-03 and the limits stated in that record. The board says a scored tier requires at least 2 measures with a stress percentile at 60 or more observations each. Accruing or absent measures are disclosed rather than rendered as calm.\n\nThe analysis does not claim an observed market price, an executable trading condition, a causal transmission from funding to IG, or a cross-market composite. It does not identify the unnamed measure, invent withheld IG percentiles, or convert Seiche’s calibrated composite into a historical percentile. It is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com)\n- [Dated Undertow dispatch](https://liquilens-undertow.com/dispatch/2026-09-03.json)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-03-ig-remains-normal-as-its-measures-disagree-funding-overlay-reads-erosion/", "date": "2026-09-03", "dek": "IG’s within-board score moved easier by 0.08 while remaining NORMAL on 2026-09-03. That tier is separate from, and does not incorporate, Seiche’s declared EROSION funding overlay.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:0ca3e44e0ce8510919eb5ce47e06a8002349e4cc236dacb22c3c9b224efae525", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:53149087ecbb89ae5b1471429e83925f8ede471bafb5e034f65dcfe113d7ad68", "status": "applied"}, "evidence_asof": "2026-09-03", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "a1ff9ce6d41d3270083943bcf258d92f9b5fb0e9f877ec2f6586f765951c915d", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:0ca3e44e0ce8510919eb5ce47e06a8002349e4cc236dacb22c3c9b224efae525", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:53149087ecbb89ae5b1471429e83925f8ede471bafb5e034f65dcfe113d7ad68", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft states that an unnamed third qualifying measure “contributes to the tier.” The dossier reports 3 available and 3 qualifying IG measures but supplies detail for only 2 named measures; it does not identify or describe a third measure. The repaired copy states only the documented disclosure gap.", "The source list repeats the same Undertow API URL to reach six links. The repair supplies six dossier-grounded approved-source links, including the dated canonical dispatch URL supplied in the dossier.", "All other material numerals and central claims in the repaired version are grounded in the dossier: 1 of 9 scoring segments; IG score 0.4336; 3 available and qualifying measures; -0.08 delta; Seiche 44.5/100 and 100% coverage; visible OAS percentile 0.3321; and the 2-measure, 60+-observation qualification rule."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "IG remains NORMAL as its measures disagree; funding overlay reads EROSION", "published_at": "2026-09-03T21:49:31Z", "quality": {"failures": [], "numeric_tokens_checked": 16, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1140}, "slug": "2026-09-03-ig-remains-normal-as-its-measures-disagree-funding-overlay-reads-erosion", "topic": "IG", "type": "current_analysis", "word_count": 1140}, "2026-09-04-autumn-2022-treasury-replay-passed-a-narrow-ust-test": {"body_md": "## The read\n\nUndertow’s historical replay of Autumn 2022 Treasury illiquidity passed its registered test. The expectation was specific: the UST cell had to reach STRAINED or DEFICIENT on at least one weekly as-of within the window from 2022-09-15 through 2022-11-30. The replay reports a peak UST score of 0.8193, classified as STRAINED, on 2022-10-05. Its verdict is PASS.\n\nThat is a narrow finding. It says this UST construction cleared its pre-stated bar in one registered historical episode. It does not establish that Undertow detects every Treasury-liquidity disruption, measures every channel of market functioning, or supplies a forecast or trade signal. The dossier describes a tier as a deterministic board derivation, not an observed price.\n\nThe replay recorded 52 as-of dates and scored all 52. The relevant peak reading was reported as knowable on 2022-10-13. That timing matters because the replay truncates the stored pack to observations dated on or before each historical as-of. It is an attempt to prevent later-dated observations from entering an earlier reading.\n\nThe current daily board does not alter that historical conclusion. On 2026-09-04, the dispatch reports that one of nine segments scored, with IG classified NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK were still accruing history. Separately, Seiche declared its funding overlay EROSION. That overlay is funding context only: the evidence braid says it was not used in Undertow tiering, and there is no blended cross-product score.\n\n## The mechanism\n\nThe peak UST reading drew on three qualifying measures: dealer UST bills net positions, dealer UST coupon net positions and dealer UST settlement fails. Their reported stress percentiles at the 2022-10-05 peak were 0.7857, 0.6932 and 0.9789, respectively. The UST cell therefore had three qualifying measures and a score of 0.8193.\n\nThe construction uses weekly levels ranked against expanding histories of the respective measures. It frames dealer position measures as congestion measures and settlement fails as a measure of strain. That makes the output a defined instrument reading, not a comprehensive reconstruction of Treasury-market liquidity.\n\nThe record also discloses what was unavailable. In the peak measure set, the 2s10s slope and dealer UST net positions were null. The pack also lacked pre-window observations for HY and IG OAS series. Those OAS absences did not stop the UST cell from scoring because it retained three qualifying measures, but they bar any claim that this was a complete cross-market reconstruction.\n\n## The countercase\n\nA pass in a historical stress episode can be easier to obtain than a robust validation result. The principal concern is not whether the 2022 episode occurred; the replay tests the instrument’s registered threshold, not the historical existence or full anatomy of the episode.\n\nThe dossier identifies a material limitation in the dealer-position measures. They are weekly levels ranked against expanding own histories beginning in 2013. Dealer balance sheets grew structurally over that period, so a high level percentile can contain structural trend as well as stress. A percentile in this construction is therefore not automatically a direct measure of executable liquidity.\n\nThe product’s own cited negative control is Q2-2017. The replay documentation says that control fails for the reason that the level-based position measures can carry structural trend. That does not reverse the Autumn 2022 PASS. It does, however, show why a successful registered episode alone cannot justify broad claims about reliability.\n\nThe vintage limitation is equally important. The replay is explicitly “truncated-latest”: it uses today’s stored pack, limited to observations dated on or before each as-of. FRED and New York Fed series may be revised, and the product has no ALFRED-style vintage archive. The result is therefore a construction replay on final data, not a demonstration of exactly what users could have calculated from data vintages available during Autumn 2022.\n\n## What would falsify this\n\nThe article’s central claim would fail if the published replay could not reproduce a PASS against the registered expectation: a UST tier of STRAINED or worse at least once in the stated window. It would also fail if the displayed 2022-10-05 peak did not have the reported score of 0.8193 and STRAINED classification, or if the registration had been changed after results were known without a separately visible record.\n\nThe broader interpretation would weaken if a correction to the dated-observation treatment, measure availability or final-data revision treatment removed the qualifying STRAINED reading. It would weaken, too, if appropriate calm-period controls repeatedly produced stress classifications. The Q2-2017 negative control is relevant precisely because the dossier identifies structural trend in level-based position measures as a known limitation.\n\nNothing in this replay becomes a forward call by passing. The dossier states that a daily board reading is not silently promoted into a forward call, and its sealed-call record is separate from the daily reading.\n\n## Method and limits\n\nThis is a historical replay of a registered UST expectation, not investment instruction. The episode window runs from 2022-09-15 to 2022-11-30. The result reports 52 scored as-of dates, a peak score of 0.8193 on 2022-10-05, three qualifying measures at that peak and a PASS verdict.\n\nThe research boundary is strict. Undertow provides the historical cross-market-liquidity board and replay; Seiche provides a declared funding-context overlay only. No composite or blended score is claimed. Accruing or absent measures are disclosed rather than rendered as calm, and withheld percentiles are not invented. The replay is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow public research surface](https://liquilens-undertow.com)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche funding research surface](https://seiche.info)\n- [LiquiLens research surface](https://liquilens.in)\n- [Undertow LiquiLens Telegram bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public research surface: board and replay context](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-04-autumn-2022-treasury-replay-passed-a-narrow-ust-test/", "date": "2026-09-04", "dek": "Undertow’s registered Autumn 2022 test reached STRAINED in the UST cell. The result supports a bounded historical replay, not a live-vintage simulation, cross-market score or forecast.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:96f0a44f14db3064e5082866c61ce7b2d80ddd709d233ab8de573bad93448e13", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:f5a9337c50301d9503b30a5ec8908528ec9fc81fe7d18c664b1031938e0272d9", "status": "applied"}, "evidence_asof": "2026-09-04", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "e9bf8514df37c94190f7a45dde805f4f298ad75512ba08c011f4729cb08281c4", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:96f0a44f14db3064e5082866c61ce7b2d80ddd709d233ab8de573bad93448e13", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:f5a9337c50301d9503b30a5ec8908528ec9fc81fe7d18c664b1031938e0272d9", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft contains unsupported methodological wording, including that the replay applies “orientation and coverage rules.” Those rules are not stated in the dossier.", "The source list is padded with a duplicate Undertow link presented as a separate replay reference. It should be replaced by substantive sourcing or removed.", "The discussion of current-board status is accurate but overextended for a historical-replay thesis; it should be limited to the research boundary and kept distinct from the replay result.", "The countercase should identify the dossier’s specific Q2-2017 negative control rather than refer vaguely to a later period.", "Repairs are needed to tighten conditional claims, preserve the final-data-vintage limitation, and retain six approved Markdown links without adding facts."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "Autumn 2022 Treasury replay passed a narrow UST test", "published_at": "2026-09-04T21:47:22Z", "quality": {"failures": [], "numeric_tokens_checked": 19, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 941}, "slug": "2026-09-04-autumn-2022-treasury-replay-passed-a-narrow-ust-test", "topic": "oct2022_ust_illiquidity", "type": "historical_replay", "word_count": 941}, "2026-09-05-the-march-2020-ig-replay-passed-but-only-as-a-test-of-this-instrument": {"body_md": "## The read\n\nThe narrow conclusion from Undertow’s historical replay is that its investment-grade credit instrument registered severe stress during the March 2020 dash for cash. It is not a broader claim that every part of the episode was measured, that the instrument would identify every future dislocation, or that a current board reading carries a directional market implication.\n\nThe current dispatch offers little reason to manufacture a larger present-tense story. It records one scored segment out of nine, with IG reading NORMAL on a within-board score of 0.4338. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. Separately, Seiche’s declared funding overlay reads EROSION. These are adjacent observations, not ingredients in a combined score: the funding overlay was explicitly not used in Undertow’s tier derivation.\n\nThat quiet and incomplete current board makes the registered replay more useful as a question of instrument design. The episode covers 2020-03-01 through 2020-04-30. Its preregistered expectation was limited: the IG cell would reach STRAINED or DEFICIENT on at least one weekly as-of within that window. The [replay record](https://api.seiche.info/undertow) reports a PASS. Its peak was DEFICIENT, with a score of 0.9824, on 2020-04-15. All 9 scored as-of dates in the window qualified.\n\nThe verdict describes the instrument, not whether history happened. It says that the registered construction, using the evidence available in its replay pack and its stated transformations, crossed its defined threshold in a named period. It does not turn a historical episode into an investment signal. Research and market data, not investment advice.\n\n## The mechanism\n\nThe instrument’s IG cell rests on two New York Fed primary-dealer measures: dealer corporate net positions, treated as congestion, and dealer corporate settlement fails. Both histories are held back to 2013. In the replay, the two measures qualified on every scored as-of date, allowing the cell to receive a tier.\n\nThe sequence is revealing without requiring a claim that the measures alone explain the event. On 2020-03-04, the replay classified the cell as NORMAL, with dealer corporate net positions at 0.4268 and settlement fails at 0.9876. On 2020-03-11 and 2020-03-18, the cell remained NORMAL despite settlement fails near the top of the measure’s own accrued history, because the net-position measure was much lower. On 2020-03-25, the combined score reached 0.8479 and the tier became STRAINED. It moved to DEFICIENT on 2020-04-01 and remained there through the end of the replay window.\n\nThat progression is the mechanism this replay can support: the tier is a deterministic board derivation from the qualifying measures, rather than an observed price, a claim about trading conditions at every moment, or a trade signal. The score rose as the registered inputs jointly met the construction’s stress conditions. The replay’s result is therefore stronger than a retrospective narrative that merely identifies a famous date and weaker than a full reconstruction of market liquidity.\n\nThe historical rationale identifies why IG was registered as an independent segment for this episode. Corporate-bond market functioning broke down alongside Treasuries in March 2020; the rationale also records dealers stopping bids, bond ETFs trading at historic discounts to NAV, and the Federal Reserve’s announcement of the PMCCF and SMCCF credit facilities on 23 March. But the replay should not be credited with measuring every one of those facts. Its result comes from the stated primary-dealer measures, not from a blended narrative of prices, ETFs, policy and funding.\n\nThere is an important timing distinction. The as-of date for the peak was 2020-04-15, while the replay says that reading was knowable at 2020-04-23. A signal reconstructed from data dated by an as-of is not automatically a signal available contemporaneously. Undertow records this gap rather than concealing it.\n\n## The countercase\n\nA PASS in a crisis window is necessary evidence for an instrument intended to register stress, but it is not sufficient evidence of broad reliability. The available inputs may be sensitive to structural balance-sheet trends as well as dislocation. The replay itself flags that dealer positions and fails are percentiles of weekly levels against expanding own histories beginning in 2013. Dealer balance sheets grew structurally over that span, so a level percentile can contain a trend as well as a stress signal.\n\nThe omitted IG and HY option-adjusted-spread legs are another constraint, not a minor footnote. The pack has no IG OAS or HY OAS observation on or before the replay dates because those series begin in 2023-09-05. They could not contribute to the 2020 result. The PASS is consequently a pass for the measures that existed in the pack, not validation of a richer credit-liquidity composite.\n\nNor should today’s funding designation be used to repair that historical gap. Seiche’s EROSION regime is a declared funding-context overlay. It has no qualifying measures in the supplied current funding record, no own-history stress percentile supplied there, and no role in the Undertow tier. A cross-desk reading may frame comparison; it may not be fused into an apparently more authoritative number.\n\nA further countercase is that the instrument may be too slow, too narrow or too dependent on its sampling cadence for some market breaks. A genuine disruption can occur through channels absent from dealer net positions and settlement fails. Conversely, a construction can pass a dramatic window too easily if it is prone to classify elevated readings as stress. The existence of a PASS does not decide between those risks.\n\n## What would falsify this\n\nThis editorial’s claim would fail if the stated replay could not be reproduced from the registered episode, its recorded evidence and its disclosed construction boundary. It would also fail if the peak DEFICIENT result, the 0.9824 score, the 2020-04-15 as-of date, or the 9 scored dates were absent from the replay record.\n\nThe more substantive claim—that the instrument detected stress in this particular episode—would weaken if a defensible reconstruction using the same stated eligible information did not reach STRAINED or DEFICIENT within the registered window. It would also weaken if the qualifying inputs were shown not to be the inputs represented in the record.\n\nThe method should be allowed to fail outside this episode. Confidence would be reduced if calm-period controls generated comparable stress without a convincing reason, or if later replay work showed that small, defensible changes in timing or proxy choice erased the result. None of those possibilities establishes a contrary result here; they define the test the instrument must continue to face.\n\n## Method and limits\n\nUndertow’s replay is a construction replay on final stored data. For each historical as-of, readings are recomputed from today’s stored pack truncated to observations dated on or before that as-of. This is described as what was knowable-by-date, not the vintage published at the time. FRED and New York Fed series can be revised, and the product has no ALFRED-style vintage archive. The result must therefore never be presented as a live simulation.\n\nThe registered expectation predates the result record, and the result is bounded to a single IG episode. The replay reports two qualifying measures at its peak and identifies absent OAS inputs. It does not reconstruct withheld evidence, infer a current IG tier from historical performance, or issue an investment instruction.\n\nCurrent-board evidence is also limited. A scored tier requires at least two measures, each with a stress percentile based on at least 60 observations. Accruing or absent measures are disclosed rather than represented as calm. The current IG reading is a deterministic board output. The other segments that are still accruing history should not be read as a clean all-market confirmation of normality.\n\nFinally, the replay and the current funding overlay answer different questions. Undertow’s historical tier is an instrument-specific replay. Seiche’s EROSION designation is funding context. They can be compared, but they are not a blended cross-product score.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com) — public Undertow product surface and the published dispatch context.\n- [Undertow API](https://api.seiche.info/undertow) — replay, board and separate-record access point.\n- [Seiche](https://seiche.info) — funding-context product referenced as an overlay only.\n- [LiquiLens](https://liquilens.in) — related liquidity product surface; not part of the replay score.\n- [Undertow LiquiLens Telegram bot](https://t.me/Undertow_LiquiLens_bot) — live product access point.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-05-the-march-2020-ig-replay-passed-but-only-as-a-test-of-this-instrument/", "date": "2026-09-05", "dek": "Undertow’s registered investment-grade replay reached DEFICIENT during the March 2020 window. That establishes that the construction detected stress in its own inputs, not that it recreated the market in real time or validates a forecast.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:c749f532dd8ebf14b3817f3f45d01535793dcff2a05242546a07fe47cc34fa59", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:cbd3812151590566e748684e7046c991ee47c2a68f58031e72c68ad8a335e109", "status": "applied"}, "evidence_asof": "2026-09-05", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "05f28b9bf1cb648f27106cf322901cf8740d83a0e60fa9a14454cec80630956c", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:c749f532dd8ebf14b3817f3f45d01535793dcff2a05242546a07fe47cc34fa59", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:cbd3812151590566e748684e7046c991ee47c2a68f58031e72c68ad8a335e109", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The current-board methodology is misstated: a scored tier requires at least two measures with a stress percentile and at least 60 observations each; it does not require a “stress percentile at 60 or more observations.”", "The sixth Markdown link duplicates the Undertow API entry in the Sources list and is padded rather than a distinct, necessary citation. Retain five source-list links and place the sixth approved link where it substantively supports the replay discussion."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The March 2020 IG replay passed—but only as a test of this instrument", "published_at": "2026-09-05T21:49:19Z", "quality": {"failures": [], "numeric_tokens_checked": 21, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1358}, "slug": "2026-09-05-the-march-2020-ig-replay-passed-but-only-as-a-test-of-this-instrument", "topic": "mar2020_dash_for_cash_ig", "type": "historical_replay", "word_count": 1358}, "2026-09-06-the-treasury-replay-passed-its-hard-test-its-limits-matter-as-much-as-the-verdict": {"body_md": "## The read\n\nThe current board offers little basis for a broader market drama: only IG scores, reading NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. The funding overlay separately reads EROSION. Those are comparisons across distinct desks, not ingredients in a blended cross-market score. They are also not the subject of this piece.\n\nThe sharper question is historical and narrower: did Undertow’s registered Treasury instrument recognize the March 2020 dash-for-cash episode under its own rules? The answer in the replay ledger is PASS. The test covered 2020-03-01 through 2020-04-30 and required the UST cell to reach STRAINED or DEFICIENT on at least one weekly as-of within that window. Its peak result was DEFICIENT, with a within-board score of 0.9151 on 2020-04-08. The ledger records 43/43 scored as-of dates.\n\nThat is a meaningful result, but only if it is described with discipline. The verdict concerns the instrument: its registered construction, its available measures and its historical window. It does not certify that every aspect of the event was captured. It does not establish that the instrument will identify every future disruption. And it does not convert a historical replay into a trading signal or a claim about today’s market.\n\nThe narrow thesis is therefore straightforward: the UST construction cleared a deliberately hard historical bar because its qualifying inputs registered Treasury-market congestion and settlement strain during the named episode; its usefulness is bounded by the quality, availability and vintage of those inputs.\n\n## The mechanism\n\nThe registered episode rationale identifies the relevant channel. Customer Treasury selling overwhelmed dealer intermediation capacity in the first half of March 2020; primary-dealer net Treasury inventories rose as dealers absorbed paper they could not distribute; and settlement fails spiked before the Federal Reserve’s March purchases relieved the market. The replay does not claim ownership of those underlying historical facts. It tests whether the board’s specified measures could register conditions consistent with that transmission mechanism.\n\nThe UST construction used dealer net positions in bills and coupons and settlement fails, defined as fails to deliver plus fails to receive. These are not price observations and do not ask whether an investor’s return was positive or negative. They are proxies for congestion in intermediation and settlement. That distinction is central. A liquidity instrument should be judged on whether it recognizes impaired capacity to move risk through the market, not on whether it retells a familiar price chart.\n\nThe sequence in the replay shows why the PASS has analytical content. Earlier readings in the window were NORMAL. The tier became STRAINED on 2020-03-11, when coupon-position congestion and settlement-fails readings were elevated in the construction. It later reached DEFICIENT on 2020-04-08. The peak score combined qualifying measure readings rather than relying on a single headline proxy. The result thus reflects agreement among the qualifying UST measures available to the replay, not a cross-market composite and not the separate Seiche funding regime.\n\nYet the chronology also supplies a caution. The ledger records knowledge dates after several as-of dates, and the New York Fed primary-dealer source has a stated publication lag. This is why the replay is framed as construction point-in-time: it truncates the stored pack to information dated on or before each as-of and records what the framework could calculate. It is not a claim to possess an untouched contemporaneous archive for every source release.\n\n## The countercase\n\nThe most important countercase is that a crisis-window pass can be too easy. An instrument may appear successful in a famous stress event for reasons that do not generalize. A percentile can be high because the series has a structural trend, because its own available history is limited, or because a proxy reflects a related market condition rather than executable liquidity itself.\n\nThe replay documentation identifies precisely that risk for the dealer-position and settlement-fails measures. Their percentiles are calculated from weekly levels against an expanding own history beginning 2013-04-03. Dealer balance sheets grew structurally over that period, so a level percentile may contain both a trend and a stress signal. The dossier explicitly notes a negative control that fails for this reason. That is not an incidental caveat; it limits how much confidence should be extracted from this PASS.\n\nThere is a second countercase: the instrument may have missed important channels even while passing. The registered UST replay lists a slope measure as unavailable in the cited rows, and the pack lacks pre-2023 observations for the IG and HY OAS legs. Those absences do not make the Treasury episode disappear, but they prevent a claim that the broader board supplied a complete contemporaneous map of all markets or all transmission channels.\n\nA third countercase is interpretive. A DEFICIENT tier says that this deterministic board derivation found severe conditions relative to its own construction. It is not an observed market price, an assertion of causality, or an instruction to transact. The current IG NORMAL reading and the separate funding EROSION overlay should remain separate readings, not evidence retroactively folded into the Treasury replay.\n\n## What would falsify this\n\nThe editorial claim is falsifiable. It would fail if the publicly described replay could not reproduce the registered PASS from its recorded pack and rules; if the registered expectation had been altered after results were inspected without a distinct visible registration; or if the cited UST measures did not produce the stated peak DEFICIENT outcome in the specified window.\n\nThe broader proposition—that the construction can identify meaningful Treasury liquidity stress—would weaken if defensible changes to timing, proxy treatment or data vintage removed the result, or if calm-period controls repeatedly registered comparable stress. It would gain credibility if independently available, point-in-time evidence supported the same mechanism while relevant negative controls remained calm. The method has to be allowed to fail, because a framework that cannot fail cannot be validated.\n\nThe test does not need to be falsified by history itself. The historical episode occurred regardless of the board. What can be falsified is the instrument’s claimed ability to recognize that episode under transparent conditions. That is the appropriate standard for a replay verdict.\n\n## Method and limits\n\nThis analysis uses the published Undertow dispatch and its attached historical replay record. The replay was generated on 2026-09-06. Its registered episode is March 2020 dash for cash — Treasuries, with the window running from 2020-03-01 through 2020-04-30. The registered threshold was STRAINED or worse on at least one weekly as-of. The result was PASS.\n\nThe evidence boundary is strict. The replay uses stored data truncated to observations dated on or before each historical as-of. But its own vintage note says this is a truncated-latest reconstruction, not the vintage published at the time. FRED and New York Fed series may be revised, and the product does not hold an ALFRED-style vintage archive. Accordingly, this is a reconstruction of what the construction can show with final stored data, not a live simulation of every information set available during the event.\n\nThe public pack also withholds some evidence and keeps some segments in accruing status. Those gaps are disclosed rather than translated into calm. The Seiche funding overlay is context only and was not used in the UST tier. No cross-product score is calculated here. Research and market data only; not investment advice.\n\n## Sources\n\n- [Undertow public research surface](https://liquilens-undertow.com)\n- [Undertow API and replay records](https://api.seiche.info/undertow)\n- [Seiche funding-liquidity desk](https://seiche.info)\n- [LiquiLens institution-liquidity desk](https://liquilens.in)\n- [Undertow LiquiLens Telegram access](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public research surface](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-06-the-treasury-replay-passed-its-hard-test-its-limits-matter-as-much-as-the-verdict/", "date": "2026-09-06", "dek": "A quiet current board shifts the focus to the March 2020 Treasury replay: the registered UST instrument reached DEFICIENT, clearing its stated bar without turning a historical reconstruction into a forecast.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:bccb5a0d1e3fd80dd336c821384b34c25f2cdb0987a0a8d4b767b9be982bc3f8", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:9352a44460553f97688b2e911d1b6e68cbcc91c3e5a2e2de053573e741e6ed4e", "status": "applied"}, "evidence_asof": "2026-09-06", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "d51ce97eabbcc1709d96a2f52dc0a01afa05db869e987c188084957870395530", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:bccb5a0d1e3fd80dd336c821384b34c25f2cdb0987a0a8d4b767b9be982bc3f8", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:9352a44460553f97688b2e911d1b6e68cbcc91c3e5a2e2de053573e741e6ed4e", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft’s current-board claims match the dossier: IG is the sole scored segment and reads NORMAL; the separately reported funding overlay is EROSION; remaining named segments are accruing history.", "The historical replay claims match the registered record: the window is 2020-03-01 through 2020-04-30, the expectation is STRAINED or DEFICIENT on at least one weekly as-of, the peak is DEFICIENT at 0.9151 on 2020-04-08, and the result is PASS with 43/43 scored as-of dates.", "The draft states the required research boundary: the replay is truncated-latest final-data reconstruction rather than a contemporaneous vintage archive or live simulation; it discloses revisions, missing/withheld evidence, and the absence of a blended cross-product score.", "The countercase and falsifier are substantive and dossier-consistent, including the structural-trend risk in dealer-position and settlement-fails level percentiles, unavailable measures, and reproducibility/timing/vintage failure conditions.", "All six required Markdown headings are present. The Sources section contains six Markdown links, each to an approved source domain. The prose does not introduce investment instruction or unsupported market facts.】【。"], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The Treasury replay passed its hard test. Its limits matter as much as the verdict.", "published_at": "2026-09-06T21:48:25Z", "quality": {"failures": [], "numeric_tokens_checked": 14, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1250}, "slug": "2026-09-06-the-treasury-replay-passed-its-hard-test-its-limits-matter-as-much-as-the-verdict", "topic": "mar2020_dash_for_cash_ust", "type": "historical_replay", "word_count": 1250}, "2026-09-07-a-passed-replay-shows-the-equity-instrument-s-boundary-not-a-market-wide-verdict": {"body_md": "## The read\n\nUndertow’s historical replay of the August 2024 yen-carry unwind passes its preregistered EQUITY test, but the proper conclusion is deliberately narrow: the instrument detected an extreme in its VIX measure during the registered window. It did not produce an EQUITY tier, and it does not establish that the full equity-liquidity cell captured the event.\n\nThe replay covers 2024-07-25 through 2024-08-16. Before the result was assessed, the registered expectation required the VIX close measure to reach a point-in-time own-history stress percentile of at least 0.95 on at least one day in that interval. The recorded result was 0.973 on 2024-08-05, above the registered floor, and the ledger therefore marks the episode PASS. Every as-of date in the window carried a VIX percentile.\n\nThat is a useful result because the claim is testable and limited. It says the registered measure cleared its stated hurdle under the replay construction. It does not say that August 2024 was fully represented by the board, that all equity-liquidity measures confirmed the signal, or that a passed historical test supplies a forward market call.\n\nThe distinction matters especially because the replay’s EQUITY cell is PARTIAL throughout the window. Of the measures listed for that cell, only VIX qualified. The other measures were unavailable for a score under the stated history and coverage rules. A tier needs at least 2 qualifying measures, each with sufficient accrued observations. The correct editorial reading is therefore measure-level success and cell-level incompleteness—not a retroactive declaration that the entire EQUITY segment had been scored.\n\n## The mechanism\n\nThe mechanism under examination is the board’s own historical-percentile construction. At each as-of date, the replay truncates the stored series to observations dated on or before that date, then evaluates the measure against its own accrued history. The episode was registered at measure level because the EQUITY cell could not meet its coverage gate at a 2024 as-of.\n\nThat design makes VIX the decisive observable in this test. The registered rationale states that VIX closed at 38.57 on 2024-08-05, its highest close since 2020. The expectation was not that an editorial could identify every pathway through the event after the fact. It was that, against the full VIXCLS daily history beginning in 1990, the point-in-time percentile machinery should recognize that close as an upper-tail observation. The recorded percentile of 0.973 clears the registered threshold of 0.95.\n\nThe value of that construction is discipline rather than drama. The threshold, window, measure and expected source were set down before the replay result. The resulting PASS thus tests whether the named instrument behaved as its registration said it should in one historical setting. It does not convert the percentile into an observed price, a liquidity estimate, or a universal stress detector.\n\nThe current Undertow dispatch is a separate record. It reports IG as NORMAL on its within-board score and records the Seiche funding overlay as EROSION. Those are cross-desk readings, not inputs to the replay and not components of a blended score. The public dispatch also says most segments are still accruing history. That reinforces the lesson of the replay: a disclosed absence or partial cell is not evidence of calm.\n\n## The countercase\n\nThe strongest countercase is that a passed VIX test may be too narrow to substantiate a broader equity-liquidity interpretation. The replay shows that one qualifying measure reached an extreme own-history percentile. It does not show whether the unavailable Amihud and lambda measures would have agreed, disagreed, or changed the segment-level result had they possessed enough accrued history.\n\nThere is also a material historical-data limitation. The equity-liquidity source is described as backfilled for a basket committed later, with survivorship-conditioned membership. Time truncation addresses what dates enter the calculation; it does not remove that membership limitation. The replay must therefore not be treated as a clean out-of-sample simulation of the information set that a market participant held at the time.\n\nNor is the replay an archival vintage reconstruction. The recorded vintage is truncated-latest: readings are recomputed from the currently stored pack, restricted to observations dated on or before each historical as-of. The dossier expressly warns that FRED and New York Fed series can be revised and that no ALFRED-style vintage archive is held. A PASS on final-data construction is evidence about the construction; it is not proof that the exact contemporaneous published dataset would have generated the same result.\n\nFinally, a crisis-window pass alone is not a complete validation program. A measure can pass a dramatic episode yet remain uninformative elsewhere. This replay’s verdict describes the instrument, not whether the historical event happened. It should not be read as an investment recommendation or as confirmation of a general causal account of that episode.\n\n## What would falsify this\n\nThis editorial’s central claim would fail if the public replay record could not reproduce the registered outcome: a VIX close percentile of 0.973 on 2024-08-05 against a floor of 0.95 within the stated window. It would also fail if the expectation had not been registered before assessment, if the documented coverage rules did not leave the EQUITY cell PARTIAL, or if the article treated unavailable measures as supportive evidence.\n\nThe interpretation would weaken if a defensible reconstruction using a genuine contemporaneous vintage materially changed the VIX percentile or if the stated point-in-time truncation could not be verified. It would strengthen only through additional separately recorded evidence—not by reinterpreting this PASS as a full-cell score.\n\nA future replay with sufficient qualifying measures could produce a segment tier. That would be a new result with its own evidence boundary. It cannot be inferred from the lone qualifying VIX measure in this record.\n\n## Method and limits\n\nThis article relies only on the supplied Undertow historical replay and current dispatch evidence. The episode window, preregistered expectation, threshold, measure result, tier status, source caveats and vintage description are drawn from that record. The article does not reconstruct withheld measures, calculate a new percentile, infer missing values, or add a composite across Undertow and Seiche.\n\nThe funding overlay remains context only. Seiche reports a declared EROSION regime, while its composite is explicitly not a percentile of its own history and its stress percentile is withheld. It is therefore compared with, rather than merged into, the Undertow reading. Likewise, the current board’s IG result is a separate within-board derivation, not evidence for the historical EQUITY replay.\n\nUndertow states that a tier is a deterministic board derivation rather than an observed price or trade signal. Historical replay is a construction replay on final data, not a live simulation. Research and market data only; not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com) — product context and public board access.\n- [Undertow API](https://api.seiche.info/undertow) — public dispatch and replay records.\n- [Undertow dispatch record](https://liquilens-undertow.com) — current-board context referenced separately from the replay.\n- [Seiche](https://seiche.info) — funding-overlay comparison only, never a blended score.\n- [LiquiLens](https://liquilens.in) — adjacent institutional-liquidity product context.\n- [Undertow on Telegram](https://t.me/Undertow_LiquiLens_bot) — access point for the live board.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-07-a-passed-replay-shows-the-equity-instrument-s-boundary-not-a-market-wide-verdict/", "date": "2026-09-07", "dek": "The preregistered August 2024 EQUITY replay passed because its VIX measure reached the registered own-history threshold. The result validates a narrow measurement test while the cell itself remains PARTIAL.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:6f95929f34a2d1f0777fe05d32f6bf349788e6cda1a295a7d4754fdd2ac2f948", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:7dee6b23dfd7a7e5c3c57b0cf91dc65a8fca678add2d710e11e2207a5f2cb3a4", "status": "applied"}, "evidence_asof": "2026-09-07", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "7ada0a47033ae2eed0304237bf0a90093f0a397aa5c467d61a2f6b53c91639fd", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:6f95929f34a2d1f0777fe05d32f6bf349788e6cda1a295a7d4754fdd2ac2f948", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:7dee6b23dfd7a7e5c3c57b0cf91dc65a8fca678add2d710e11e2207a5f2cb3a4", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["All required headings are present, including a substantive countercase and explicit falsifiers.", "The article contains six approved Markdown links, all drawn from the approved-source list.", "The stated research boundary is maintained: it relies on the supplied replay and dispatch record, does not create a blended Undertow-Seiche score, and identifies Seiche as context only.", "Dossier-grounded numerals and dates are accurate: the 2024-07-25 to 2024-08-16 window, 0.95 registered threshold, 0.973 result on 2024-08-05, 38.57 VIX close, 17/17 percentile coverage, and the two-measure tier requirement all align with the record.", "The article correctly confines the PASS to the measure-level VIX expectation and does not misrepresent the EQUITY cell as scored; it accurately identifies the cell as PARTIAL with one qualifying measure.", "It fairly states the principal falsification conditions and material limitations, including survivorship-conditioned membership, truncated-latest rather than contemporaneous-vintage reconstruction, and the absence of a full-cell tier.", "The prose is evidence-led and avoids investment instruction, forward calls, and claims beyond the public pack."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "A passed replay shows the EQUITY instrument’s boundary, not a market-wide verdict", "published_at": "2026-09-07T21:47:07Z", "quality": {"failures": [], "numeric_tokens_checked": 12, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1167}, "slug": "2026-09-07-a-passed-replay-shows-the-equity-instrument-s-boundary-not-a-market-wide-verdict", "topic": "aug2024_yen_carry_vix", "type": "historical_replay", "word_count": 1167}, "2026-09-08-fx-replay-of-the-august-2024-yen-carry-episode-remains-inconclusive": {"body_md": "## The read\n\nUndertow’s historical replay of its registered August 2024 yen-carry FX episode is **INCONCLUSIVE**. That is the full result. It is not a finding that FX liquidity was calm, that the episode did not occur, or that the registered expectation failed on its merits. The FX cell did not obtain a scored tier at any as-of in the stated window.\n\nThe registered test asks whether the FX cell reached **STRAINED** or **DEFICIENT** on at least one as-of between 2024-07-25 and 2024-08-16. The replay reports 1 as-of date: 2024-07-31. On that date, the FX cell was **PARTIAL**, not scored. Its two measures, amihud (fx) and lambda_hat (fx), were present but withheld. The replay records 0 qualifying measures out of 2 and 0 scored as-ofs.\n\nThe coverage rule is decisive. A scored tier requires at least 2 qualifying measures, with each stress percentile requiring at least 60 observations. The replay says the best as-of had no qualifying measures. It consequently supplies no FX tier verdict for the episode. The record is explicit that a withheld cell is an absence of verdict, never a pass.\n\nThe narrow thesis is therefore falsifiable: this registered construction could neither validate nor reject its tier expectation in the episode window because its FX cell did not meet the stated coverage requirement. That conclusion concerns the construction and its disclosed evidence boundary. It is not a broader adjudication of the August 2024 event.\n\nThe current 2026-09-08 board is separate context. It reports that 1 of 9 segments scores, with IG at NORMAL. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. The dispatch describes bounded persistence relative to 2026-09-07, while its headline states that no segment changed tier. Those current readings do not retrospectively create a score for the 2024 FX window.\n\n## The mechanism\n\nThe replay gives a direct reason for the inconclusive result. The FX cell uses Amihud and lambda-hat on an FX ETF-proxy bucket. Each measure requires 60 monthly observations before it can qualify. The registered episode rationale says the monthly history begins in 2020-08 and places the earliest scoreable FX as-of in mid-2025.\n\nThe historical window ended in 2024-08, before that stated earliest scoreable period. At the replay’s only as-of, 2024-07-31, both FX measures were withheld. The board therefore rendered PARTIAL rather than NORMAL, STRAINED or DEFICIENT. A tier is a deterministic board derivation, not an observed price and not a trade signal.\n\nThe episode registration identifies the intended condition precisely: FX was to reach STRAINED or DEFICIENT at least once in the window. The registered rationale references the fall in USD/JPY from its July 2024 peak into 5 August and identifies the unwind of yen-funded carry positions as the proximate cause of that day’s global equity and volatility shock. But the replay result tests the registered FX cell under its own rules; it does not add an adjudication of those broader historical claims.\n\nThe data construction also limits what “historical replay” means. The result says readings are recomputed from today’s stored pack after truncation to observations dated on or before each as-of. The record calls this knowable-by-date, not the publication vintage that was available at the time. It says FRED and New York Fed series can be revised, and that the product has no ALFRED-style vintage archive.\n\nThe source caveat adds a membership limitation. The monthly history is backfilled for a basket committed on 2026-07-17. The replay is therefore point-in-time in dates but survivorship-conditioned in membership. The dossier directs readers not to treat a verdict from that source as a clean out-of-sample result. In this instance there is no tier verdict in any event, because qualification was absent.\n\n## The countercase\n\nThe countercase is contained in the result itself: the replay does not demonstrate that the registered FX expectation was wrong. It records insufficient qualifying evidence to render the tier the expectation requires. Calling the episode a pass, a failure, or calm would all go beyond the PARTIAL cell.\n\nA second constraint is the proxy and coverage boundary disclosed by the registration. The test concerns Amihud and lambda-hat on an FX ETF-proxy bucket, rather than an unrestricted claim about all FX market conditions. The result names the two measures and says both were present but withheld. It does not provide their withheld values, and this article does not reconstruct them.\n\nThe current funding context does not resolve that absence. On 2026-09-08, Seiche’s declared funding regime is EROSION. Its underlying note reports a Seiche composite index of 44.9/100 and 100% coverage, while also stating that the scale is a calibrated weighted blend of engine sub-scores, not a percentile of its own history. Undertow classifies Seiche as a funding-context overlay only and says it was not used in the tier. There is no cross-product score to infer.\n\n## What would falsify this\n\nThis article would be falsified by a public replay record for this registered episode showing that the FX cell had a scored tier on an as-of within 2024-07-25 to 2024-08-16. In particular, a record showing 2 qualifying FX measures and a scored cell would contradict the statement that the reported replay had no scored as-of.\n\nIt would also be falsified if the record did not show the reported sole as-of of 2024-07-31, if it did not classify that cell as PARTIAL, or if it did not report the two named FX measures as present but withheld. A record that changed the registered window or expectation would be a different test from the one described here.\n\nThe article does not claim that no alternative construction could produce a result. It claims only that this disclosed replay, with its registered expectation and reported coverage rule, returned INCONCLUSIVE. The falsifier is thus the replay record itself, not an interpretation of the event outside the stated instrument.\n\n## Method and limits\n\nThis article uses only the public Undertow dispatch and the attached historical replay record. The dispatch is dated 2026-09-08; the replay was generated at 2026-09-08T21:46:55+00:00. The episode was registered on 2026-07-23, before the reported replay result, with a stated segment, window and expectation.\n\nThe research boundary is strict. The article makes no investment instruction, no claim beyond the public pack, and no blended Undertow-Seiche score. It states missing and withheld evidence rather than assigning a tier the board withheld. The historical result uses truncated latest data rather than contemporaneous publication vintages and is survivorship-conditioned in membership. Research and market data are not investment advice.\n\n## Sources\n\n- [Undertow public desk](https://liquilens-undertow.com)\n- [Undertow public records](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow records access](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-08-fx-replay-of-the-august-2024-yen-carry-episode-remains-inconclusive/", "date": "2026-09-08", "dek": "Undertow’s registered FX test produced no scored as-of in its 2024-07-25 to 2024-08-16 window. The result is not evidence of calm: both measures were withheld and the FX cell remained PARTIAL.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:b6fe5f2b4afc6aed7b61b8503261741048fe5ba9b0d4883b162884a397085887", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:fcbda250adb9c5b377482270e094216158776a6e7010fa5e50532368bcdd5814", "status": "applied"}, "evidence_asof": "2026-09-08", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "48355a44c96d24032a4142b2bc909118357961f713da07ac5011d6571a63aaae", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:b6fe5f2b4afc6aed7b61b8503261741048fe5ba9b0d4883b162884a397085887", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:fcbda250adb9c5b377482270e094216158776a6e7010fa5e50532368bcdd5814", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft adds unsupported countercase claims about timing intervals, venues, currency pairs, funding, volatility, collateral and intermediary behavior. The dossier does not establish those as explanations for this result.", "The draft adds an unsupported assertion that a future scored result could reflect executable liquidity in relevant channels. The dossier states methodological limits, but does not make that claim.", "Several passages are padded with general methodological rhetoric rather than necessary dossier-grounded dispatch prose. A repair should retain the registered test, the withheld evidence, the countercase supplied by the record, and explicit falsifiers."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "FX replay of the August 2024 yen-carry episode remains inconclusive", "published_at": "2026-09-08T21:50:01Z", "quality": {"failures": [], "numeric_tokens_checked": 24, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1114}, "slug": "2026-09-08-fx-replay-of-the-august-2024-yen-carry-episode-remains-inconclusive", "topic": "aug2024_yen_carry_fx", "type": "historical_replay", "word_count": 1114}, "2026-09-09-june-2013-china-basin-replay-remains-not-runnable-under-undertow-s-disclosed-coverage": {"body_md": "## The read\n\nUndertow’s registered replay for the **June 2013 Shibor shock — China basin** returned **NOT_RUNNABLE**. That is not a calm reading, a failed stress signal, or a finding about whether the historical disruption occurred. It is a disclosed boundary on what this replay construction can test.\n\nThe registered expectation was specific: the CN cell would reach **DEFICIENT** on at least one as-of within the window from **2013-06-01** through **2013-07-15**. The result contains no scored as-ofs and no evidence trace. Its stated reason is equally specific: no wired replay source feeds the CN segment, while the packs that carry CN have no history reaching the registered window.\n\nThe appropriate conclusion is narrow. Undertow cannot, on this record, establish whether its CN construction would have reached the registered tier during that episode. It also cannot treat the absent result as evidence of normal conditions.\n\nThe current **2026-09-09** board is separate context. It reports that **1 of 9** segments scores, with IG reading NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK continue accruing history. Its funding overlay is declared EROSION. That overlay is context only: it is not used in an Undertow tier and does not supply missing CN replay history.\n\n## The mechanism\n\nA replay requires historical inputs for the registered segment and window before it can produce a tier outcome. Here, the replay record identifies CN as the segment, the June-to-July 2013 interval as the test window, and DEFICIENT as the threshold to be evaluated. It then identifies the missing prerequisite: no available wired replay source declares CN for the test.\n\nThe record also says that the packs carrying CN do not extend to the window. Without a qualifying historical input surface, there is no basis for calculating the CN cell at an as-of in that interval. The result is therefore NOT_RUNNABLE rather than a scored tier.\n\nThat distinction follows Undertow’s disclosed treatment of incomplete coverage. Its limitations state that PARTIAL means a segment is still accruing: a scored tier requires at least **2** measures with a stress percentile at **60+** observations each. An accruing or absent measure is disclosed and is not rendered as calm. The CN replay result applies the same principle to a historical test with no run-ready source coverage.\n\nThe replay’s vintage statement imposes a second limit. Readings are recomputed from stored packs truncated to observations dated on or before each historical as-of. The record says this is what was knowable by date in the construction, not the vintage published at the time. It further warns that FRED and New York Fed series can be revised and that Undertow has no ALFRED-style vintage archive. Even a runnable result would therefore be a construction replay on final stored data, not a live simulation of contemporaneous market information.\n\n## The countercase\n\nA reasonable countercase is that the named episode is historically significant enough that a coverage failure should not prevent a conclusion. But knowledge that an event occurred is different from validation of a particular measurement construction.\n\nThe registered replay does not test history in the abstract. It tests whether Undertow’s CN cell, using its disclosed sources and rules, would have reached DEFICIENT during the specified window. On the published record, the needed CN source coverage is absent. Substituting another series, expanding the window, or changing the segment would create a different test rather than complete this one.\n\nThe current Seiche EROSION regime does not cure that gap. The funding measure is a declared regime from a calibrated weighted blend of engine sub-scores, with high values denoting stress. It is explicitly not a percentile of its own history, and its stress percentile is withheld. More importantly, the evidence braid says the Seiche relationship is a funding-context overlay only and is not used in an Undertow tier. No cross-product score is supplied.\n\n## What would falsify this\n\nThis article would be wrong if the published replay record contained a wired source that feeds CN and reaches the registered **2013-06-01** to **2013-07-15** window. It would also be wrong if the registered expectation were not a CN result reaching DEFICIENT on at least one as-of, or if the recorded verdict were not NOT_RUNNABLE.\n\nA disclosed replay rerun that produced scored CN as-ofs for the same registered episode would supersede the central finding. So would a correction to the published source-coverage record showing that qualifying CN history had been available for the window.\n\nWhat does not falsify the finding is evidence about the historical episode from another provider or a separate market measure. Such evidence may bear on the episode itself, but the claim here concerns only whether the disclosed Undertow construction was runnable for its registered CN test.\n\n## Method and limits\n\nThis is a historical-replay article based only on the published Undertow record: the registered episode, its expectation, the result, the disclosed missing-source reason, and the stated replay-vintage limitation. It does not reconstruct missing CN inputs, infer withheld percentiles, or add external evidence about the historical episode.\n\nUndertow describes a tier as a deterministic board derivation, not an observed price or a trade signal. The current funding overlay remains separate context rather than a CN tier, a replay result, or a forward call. The sealed-call record is expressly separate from this daily reading.\n\nResearch and market data only; not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow API record access](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-09-june-2013-china-basin-replay-remains-not-runnable-under-undertow-s-disclosed-coverage/", "date": "2026-09-09", "dek": "The registered test for the June 2013 Shibor shock cannot establish whether Undertow’s CN cell reached DEFICIENT: no wired replay source feeds CN for the window, and the packs carrying CN do not reach it.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:e091f7bd084ddb3b467209d048004e865305cd6d5653790c9f9ae5569c2bdb05", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:b89f1c05ab80f53c96a40775ef406ac46bfda0540b706b5427446e2cd896e53d", "status": "applied"}, "evidence_asof": "2026-09-09", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "5f70557497ca8c14febf922ce382bd2c6205bf89371fa7f2da95ae1b7259e788", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:e091f7bd084ddb3b467209d048004e865305cd6d5653790c9f9ae5569c2bdb05", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:b89f1c05ab80f53c96a40775ef406ac46bfda0540b706b5427446e2cd896e53d", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft is substantially padded for a narrow, single-record finding. It repeats the distinction between an unavailable replay and a historical-event judgment across multiple sections without adding dossier-grounded evidence.", "The statement that the current dispatch explains “why the desk returned to a replay” asserts an editorial motive not supplied by the dossier. The dossier identifies the current board as a daily reading and separately supplies the replay record; it does not state a causal reason for publishing this replay article.", "Several broad formulations—such as a replay being “harder and more useful” and claims about what a replay “should” do—are unnecessary editorial generalizations. The article should stay with the registered expectation, the disclosed source gap, the NOT_RUNNABLE result, and the stated vintage limit."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "June 2013 China-basin replay remains NOT_RUNNABLE under Undertow’s disclosed coverage", "published_at": "2026-09-09T21:49:12Z", "quality": {"failures": [], "numeric_tokens_checked": 11, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 917}, "slug": "2026-09-09-june-2013-china-basin-replay-remains-not-runnable-under-undertow-s-disclosed-coverage", "topic": "shibor_shock_jun2013_cn", "type": "historical_replay", "word_count": 917}, "2026-09-10-march-2020-hy-replay-is-not-runnable-under-undertow-s-documented-coverage-boundary": {"body_md": "## The read\n\nUndertow’s historical replay for the March 2020 high-yield episode is **NOT_RUNNABLE**. The finding is narrow: no series feeding the HY segment in the stored pack holds an observation dated within the registered window of 2020-03-01 through 2020-04-30. The replay therefore has no as-ofs and no scored as-ofs from which to test its stated condition.\n\nThat condition is specific. The registered expectation says the HY cell reaches STRAINED or DEFICIENT on at least one as-of inside the window. The result neither passes nor fails that expectation. It says the product cannot adjudicate it using the stored coverage identified in the replay record.\n\nThis is an evidence boundary, not a claim that the historical episode was calm. The episode registration itself describes ICE BofA US High Yield OAS as moving from roughly 360bp in mid-February 2020 to above 1000bp on 23 March, its widest since 2009. But that historical description cannot supply the missing observations needed for this product’s own registered board calculation.\n\nThe current daily reading does not repair the gap. The 2026-09-10 dispatch says IG reads NORMAL and that the separate Seiche funding overlay is declared EROSION. It also says 1 of 9 segments scores today while other named segments continue to accrue history. Those are current, bounded readings; they are not a blended cross-product score and are not evidence for a March 2020 HY replay.\n\nThe article’s claim is therefore falsifiable and limited: the published NOT_RUNNABLE verdict accurately reports that Undertow’s stored HY coverage cannot test its registered March 2020 expectation.\n\n## The mechanism\n\nA replay requires observations dated within its registered historical window. Undertow’s result records that no series feeding HY holds a single observation in the March 2020 window. Its missing-coverage record identifies the relevant FRED-served series as BAMLH0A0HYM2, with stored coverage from 2023-09-11 through 2026-09-09.\n\nThe episode registration anticipated this limitation. It says the FRED-served ICE BofA series is available to the product only as a trailing window and that the pack begins after March 2020. The result consequently reports zero as-ofs and zero scored as-ofs. With no eligible as-of reading, the registered cell-tier condition cannot be evaluated.\n\nNOT_RUNNABLE preserves an important distinction. It does not mean that the test found no stress. It means the test cannot be run for this window from this stored pack. Recasting that result as either a successful test or a calm historical reading would exceed the record.\n\nThe replay has a second, separate boundary. It is recomputed from today’s stored pack after observations are truncated to dates on or before each historical as-of. The record calls this a construction replay on final data, not a live simulation. It warns that FRED and NY Fed series can be revised and that Undertow holds no ALFRED-style vintage archive.\n\nThe Seiche overlay is also bounded. The funding record declares EROSION, but its composite is null, it has zero qualifying measures, and its stress percentile is withheld rather than invented. The dispatch identifies it as a funding-context overlay only and says it was not used in Undertow tiering. It cannot be added to the historical HY result to create a score that the replay did not produce.\n\n## The countercase\n\nThe strongest countercase is straightforward: the documented coverage gap does not settle the historical episode. The replay record says only that this product lacks eligible stored observations for the registered HY window. It does not purport to measure every historical fact associated with March 2020.\n\nThat countercase supports, rather than defeats, the NOT_RUNNABLE classification. If a different stored coverage set could support an HY reading inside the window, it would need to be identified and assessed under an explicit replay construction. It would be a new evidentiary record, not an unstated replacement for the published one.\n\nThere is an opposite limit as well. A future runnable result would answer the registered cell-tier condition only. Undertow’s stated limitations say a tier is a deterministic board derivation, not an observed price or a trade signal. A result on one registered condition should not be expanded into a broader claim than the construction supports.\n\nCurrent readings offer no shortcut. IG’s NORMAL within-board score and the funding overlay’s declared EROSION regime can coexist because the dossier presents them as separate readings. Their coexistence cannot fill the missing March 2020 HY observations or transform the historical verdict.\n\n## What would falsify this\n\nThis article would be falsified if the published stored coverage could produce one or more scored HY as-ofs dated between 2020-03-01 and 2020-04-30 under the replay’s stated rules. It would also be falsified if a series feeding HY were shown to hold an eligible observation in that window, contrary to the recorded missing-coverage result.\n\nThe claim would also require revision if the registered segment, window, or expectation were shown not to match the published replay record. The record identifies HY, the March 2020 window, and an expectation of STRAINED or DEFICIENT on at least one as-of.\n\nEvidence that the March 2020 episode was severe would not itself falsify the claim. The claim is not that the episode lacked stress. It is that this stored-pack replay cannot adjudicate its specified condition for that period. New coverage may support a separately documented replay, but it should not silently alter the existing NOT_RUNNABLE record.\n\n## Method and limits\n\nThis article uses the published Undertow dispatch and historical replay record only. It does not reconstruct missing observations, infer a withheld funding percentile, or merge the Seiche overlay into an Undertow tier.\n\nThe replay result is limited to the registered HY expectation and window. It is not a contemporaneous vintage simulation, because the stored pack is truncated by observation date while the record acknowledges that source data can later be revised. It is not a forecast, investment instruction, or general certification of market conditions.\n\nThe daily board has its own stated limit: PARTIAL means history is still accruing, while a scored tier requires at least 2 measures with a stress percentile at 60+ observations each. An accruing or absent measure is disclosed rather than rendered as calm. The dispatch also says daily board readings are not silently promoted into forward calls; sealed calls are held in a separate verifiable record.\n\nResearch and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow source surface](https://liquilens-undertow.com)\n- [Replay-record source surface](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-10-march-2020-hy-replay-is-not-runnable-under-undertow-s-documented-coverage-boundary/", "date": "2026-09-10", "dek": "Undertow’s registered high-yield replay cannot test its March 2020 tier expectation because the stored HY coverage contains no observation inside the event window. The verdict describes the replay’s evidence boundary, not the history of the episode or a signal for today.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:e0239e0a5637b6a3f5499cffb7f292da715f2b4046fea72208d3ff19b0011392", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:4599fa4a0d2821240a16c60fa93f7fd774f1af2cd5d76b47b91ab261fa3761a9", "status": "applied"}, "evidence_asof": "2026-09-10", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "180a8b6554b581fa915d5177548bd4454d5351d22c7cbc632d10a8f63f35abe5", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:e0239e0a5637b6a3f5499cffb7f292da715f2b4046fea72208d3ff19b0011392", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:4599fa4a0d2821240a16c60fa93f7fd774f1af2cd5d76b47b91ab261fa3761a9", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft is largely dossier-grounded and includes all required headings, a clear countercase and falsifier, the research boundary, and at least six Markdown links using approved source URLs.", "Repairs are required because the countercase introduces unsupported generalized assertions: that unheld measures or market structure may reflect the episode, that proxies can respond to volatility or trend, and that the available series cannot resolve event timing. Those statements are not established by the dossier.", "The repaired version confines the countercase to the documented coverage gap, preserves the distinction between a historical episode and this product’s replay verdict, and avoids treating current-board or funding information as evidence for the historical HY result."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "March 2020 HY replay is NOT_RUNNABLE under Undertow’s documented coverage boundary.", "published_at": "2026-09-10T21:49:35Z", "quality": {"failures": [], "numeric_tokens_checked": 18, "required_headings": 6, "source_links": 7, "status": "PASS", "word_count": 1075}, "slug": "2026-09-10-march-2020-hy-replay-is-not-runnable-under-undertow-s-documented-coverage-boundary", "topic": "mar2020_dash_for_cash_hy", "type": "historical_replay", "word_count": 1075}, "2026-09-11-undertow-s-registered-india-options-replay-cannot-test-the-episode": {"body_md": "## The read\n\nUndertow’s historical replay reaches a decisive but deliberately narrow conclusion: it cannot test its registered INDIA_OPTIONS hypothesis for the July 2025 SEBI order against Jane Street — single-firm withdrawal. The verdict is **NOT_RUNNABLE** because `INDIA_OPTIONS` is not a segment in the Undertow board and the product observes no market of that name. The replay therefore has no construction to run, no as-of observations, no scored as-of observations, no trace and no evidence from which to infer a tier.\n\nThat limitation is the story. The registered expectation was specific: an `INDIA_OPTIONS` cell would reach STRAINED or DEFICIENT at least once during the episode window. But Undertow’s board vocabulary is UST, IG, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK — not `INDIA_OPTIONS`. Replacing the absent row with a nearby market would change a registered test into an editorial substitute.\n\nThe current board does not resolve the historical gap. On the 2026-09-11 daily reading, 1 of 9 segments scores: IG reads NORMAL. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK continue to accrue history. Separately, the Seiche funding overlay is declared EROSION. Those are current and distinct desk readings, not evidence about the historical India-options episode, and they are not blended into a cross-product score.\n\nThe thesis is therefore limited: the replay verdict describes Undertow’s coverage and construction, not whether the historical episode occurred, how severe it was, or whether another dataset could document it. The product supplies an explicit no-result boundary rather than a retrospective tier.\n\n## The mechanism\n\nThe registered episode asks whether liquidity supplied by a single firm leaves with that firm. To test that question, the instrument needs a defined market row, an eligible set of measures, observations during the selected window and rules that can convert those observations into a tier. Undertow has none of those ingredients for `INDIA_OPTIONS`.\n\nThis is not merely a sparse-data result. The replay record says `INDIA_OPTIONS` is not in the board’s vocabulary and that this repository collects no market of that name. It also identifies a different product, LiquiLens, as having a different data plane that includes an FBIL, CCIL and RBI daily snapshot. That does not fill Undertow’s gap. Material from a separate product cannot be moved into this construction, and a neighboring segment cannot be substituted for the named market.\n\nThe separation matters because Undertow’s liquidity tiers are deterministic board derivations, not observed prices or trade signals. A scored tier requires at least 2 measures with a stress percentile and at least 60 observations for each qualifying measure. Measures that are still accruing or absent are disclosed rather than rendered as calm. The free board also withholds percentile detail for a row in the stated circumstance; the missing value is not available for editorial reconstruction.\n\nSeiche’s EROSION reading has a separate role. It is a declared funding-context overlay only, not an input that creates or changes an Undertow liquidity tier. Its displayed index is described as a calibrated weighted blend of engine sub-scores, with high values indicating stress, rather than a percentile of its own history. Undertow does not supply a blended Undertow-Seiche score.\n\nHistorical replay adds a further limit. Readings are recomputed from today’s stored pack, truncated to observations dated on or before each historical as-of. The dossier calls this a construction replay on final data, not a simulation of the information vintage published at the time. It warns that FRED and New York Fed series can be revised and that Undertow has no ALFRED-style vintage archive. Here, the prior question arises even earlier: no INDIA_OPTIONS construction exists to replay.\n\n## The countercase\n\nA meaningful event can be real and consequential while falling outside an instrument’s coverage. The July 2025 episode may have affected markets, participants, execution conditions or risk transfer in ways that an India-options row could have captured had such a row and inputs existed. It may also have appeared through channels not represented by a tiering construction. NOT_RUNNABLE does not mean calm, immaterial, disproved or unworthy of further reporting.\n\nNor would a neighboring row solve the problem. Different segments can have different participants, instruments, trading conventions, liquidity measures, observation cadence and historical distributions. A substitute could produce an apparently complete replay while answering a different question. The registered expectation names INDIA_OPTIONS; the board does not contain it.\n\nCurrent readings cannot be used as a shortcut. Seiche’s current EROSION regime is not a result for the historical India-options window. IG’s current NORMAL reading likewise says nothing about an unmeasured market in a prior period. The dossier specifically keeps the products and their constructions separate.\n\n## What would falsify this\n\nThis editorial’s limited claim would fail if the public replay record showed an eligible INDIA_OPTIONS construction for the registered window and a calculated replay result. It would also fail if Undertow’s board vocabulary included `INDIA_OPTIONS`, if the repository contained a named Indian-market input suitable for the test, or if the published replay record contradicted its stated reason for the missing result.\n\nA future, explicitly defined product extension could make a different test possible: it would need India-options coverage, recorded inputs, a stated construction and a separately dated replay result. That would be new evidence. It would not turn the current NOT_RUNNABLE verdict into a historical pass or fail by assertion.\n\nThe broader proposition that a single provider’s withdrawal can matter for liquidity requires evidence from an instrument that observes the relevant market and can distinguish that proposed channel from alternatives. Undertow’s present replay neither confirms nor rejects that proposition.\n\n## Method and limits\n\nThis is a historical-replay article based only on Undertow’s published evidence envelope. The registered window runs from 2025-07-03 through 2025-08-15. The replay reports 0 as-of observations and 0 scored as-of observations because the named segment is absent. Its missing-input record, stated reason and NOT_RUNNABLE verdict control this account.\n\nThis article does not reconstruct withheld measures, infer a hidden percentile, substitute a neighboring segment, claim an observed price move or turn a daily board reading into a forward call. It does not use LiquiLens material to populate Undertow and does not blend the Seiche overlay into an Undertow tier. The dossier says sealed calls are maintained as a separate verifiable record and are not attached silently to a daily board reading.\n\nCurrent-board context appears only to clarify the separation of products and readings. It is not a verdict on the historical event. Research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com) — public product context.\n- [Undertow API](https://api.seiche.info/undertow) — published Undertow records.\n- [Undertow’s 2026-09-11 dispatch record](https://liquilens-undertow.com/dispatch/2026-09-11.json) — the daily reading cited for current-board context.\n- [Seiche](https://seiche.info) — separate funding-context product.\n- [LiquiLens](https://liquilens.in) — separate product referenced in the replay limitation.\n- [Undertow on Telegram](https://t.me/Undertow_LiquiLens_bot) — public Undertow access point.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-11-undertow-s-registered-india-options-replay-cannot-test-the-episode/", "date": "2026-09-11", "dek": "The registered replay for the July 2025 SEBI order against Jane Street — single-firm withdrawal is NOT_RUNNABLE. That is a boundary on what Undertow’s instrument can establish, not a finding on the event’s severity.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:48263e0eb07abadb8b1ad80a24c65deb44047a25f83adc8b75c2b2964af491a2", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:eb3e242bed06c19bc90df07ea7b68959e1874384633b8b48799a53f725bb4bac", "status": "applied"}, "evidence_asof": "2026-09-11", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "c0255eecb4e0ac65cf1ce20a6b8ea1c947b0cea46f768dd3fea56aaa01841bc3", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:48263e0eb07abadb8b1ad80a24c65deb44047a25f83adc8b75c2b2964af491a2", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:eb3e242bed06c19bc90df07ea7b68959e1874384633b8b48799a53f725bb4bac", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The headline overstates the record by saying the replay was “built to examine” this episode. The dossier says the episode was registered as an on-thesis case, but it also establishes that Undertow has no INDIA_OPTIONS row or Indian-market construction.", "The draft omits the dossier’s precise 60+-observation condition when describing the threshold for a scored tier. The repaired copy states the condition exactly.", "The sixth source link duplicates the Undertow homepage and is padded. The repaired source list replaces it with the dossier’s specific published daily-dispatch record, yielding six substantive, dossier-grounded Markdown links."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "Undertow’s registered India-options replay cannot test the episode", "published_at": "2026-09-11T21:49:28Z", "quality": {"failures": [], "numeric_tokens_checked": 13, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1124}, "slug": "2026-09-11-undertow-s-registered-india-options-replay-cannot-test-the-episode", "topic": "sebi_janestreet_jul2025", "type": "historical_replay", "word_count": 1124}, "2026-09-12-autumn-2022-treasury-replay-passed-its-registered-test-but-only-for-the-instrument-tested": {"body_md": "## The read\n\nUndertow’s current board is quiet, so the more useful question is historical and narrow: could the registered UST instrument recognize the Treasury-liquidity episode it was built to test? For the replay titled *Autumn 2022 Treasury illiquidity*, the answer is yes, within the construction’s stated limits.\n\nThe registered expectation was that the UST cell would reach STRAINED or DEFICIENT at least once during the window from 2022-09-15 through 2022-11-30. The recorded result is PASS. The peak reading was STRAINED, with a score of 0.8193 on 2022-10-05. All 52 as-of dates in the window scored. That result is the thesis: the specified UST construction detected stress during this named episode under its registered bar.\n\nThat is a more limited claim than it may first appear. A replay verdict describes the instrument, not whether history happened. The episode’s occurrence is not being put to a model vote. Nor does the PASS establish that the board would identify every liquidity event, capture every channel of this event, or provide a tradable signal. It says that, when the stored data pack is truncated to each historical as-of and the fixed rules are applied, the UST cell reached its registered threshold.\n\nThe distinction matters because the current dispatch itself does not offer a basis for a broad market narrative. The public board says that one of nine segments scores today, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK continue accruing history. IG reads NORMAL on three qualifying measures, with a within-board score of 0.4172. Separately, Seiche’s declared funding overlay reads EROSION. Those are adjacent readings, not ingredients in a common score. Undertow does not blend the funding overlay into the UST replay, and the replay does not validate or invalidate the current funding designation.\n\n## The mechanism\n\nThe replay’s mechanism is measurement discipline rather than retrospective drama. Treasury-market liquidity can be discussed through prices, depth, volatility, financing, settlement and intermediary capacity. This construction does not claim to observe all of them. Its scored UST evidence at the peak consists of dealer UST bills net positions, dealer UST coupon net positions and dealer UST settlement fails. The recorded stress percentiles at the peak were 0.7857, 0.6932 and 0.9789 respectively.\n\nThe board uses an expanding history for these weekly level measures. At the peak, the three qualifying measures produced the STRAINED cell. Other listed measures were not substituted in to make the story cleaner: the dealer UST net-position measure and the 2s10s slope were null in the replay evidence. HY and IG OAS inputs were absent because their history in this pack begins after the episode. Their absence is disclosed, not converted into an inference about either credit market.\n\nThis is why a replay needs an explicit, pre-existing threshold. The registered bar was STRAINED or worse, not a threshold selected after the high point was visible. The episode rationale itself acknowledged that the period was a weaker prior than a rapid seizure because it was described as a slow grind and the measures are weekly levels ranked against expanding histories. A genuine failure was therefore plausible before the result was inspected. The reported PASS has value because it cleared a test that allowed for the possibility of not clearing.\n\nThe point-in-time construction also matters. The ledger recomputes readings from the stored pack using observations dated on or before each as-of date. It reports what the current pack can construct as knowable by date, not an untouched archive of every data vintage published during the event. The replay’s knowledge dates follow the stated publication lag for the relevant source. That procedure is materially more restrictive than applying a modern full-history chart to an old episode, but it is not equivalent to possessing an archival vintage database.\n\nThe historical pattern was not a single isolated print. The UST cell was NORMAL earlier in the window, then reached STRAINED on the peak date and remained STRAINED across subsequent recorded as-of readings before returning to NORMAL on later dates and then reaching STRAINED again near the end of the window. That persistence is evidence that the instrument was responsive within its own rules. It is not evidence that every underlying market condition moved in lockstep, because the instrument is a derivation from selected measures rather than a complete observation of execution quality.\n\n## The countercase\n\nThe strongest countercase is that a PASS may still overstate the construction’s practical breadth. Settlement fails and dealer-position congestion can be historically elevated for reasons that do not map one-for-one to executable liquidity. The replay documentation explicitly cautions that dealer position and fails measures are percentiles of weekly levels against an expanding history beginning in 2013. Dealer balance sheets changed structurally over that history, so a level percentile can carry trend as well as stress.\n\nThat caveat is not cosmetic. A measure with a strong historical percentile can describe an unusual reading without isolating its cause. It may reflect stress, structural change, proxy behavior, or some mixture that this public evidence cannot disentangle. The replay therefore supports a statement about the UST cell’s classification, not a precise causal account of Treasury-market conditions.\n\nThere is a second countercase: the test’s coverage is narrow. The UST replay used three qualifying measures at the peak, while several inputs were null or unavailable. The absent HY and IG OAS observations do not weaken the UST result mechanically, but they do rule out an account that presents the replay as a comprehensive cross-asset diagnosis. Likewise, the current Seiche EROSION overlay is a funding-context comparison only. Its underlying composite is calibrated as a weighted blend of engine sub-scores and is explicitly not a percentile of its own history; Undertow’s replay has no license to convert it into corroborating UST evidence.\n\nA final countercase is timing. Weekly level data can recognize a persistent deterioration while missing an intraday rupture, or remain elevated after a faster-moving condition has changed. The episode window scored on every recorded as-of date, but frequency is part of what the instrument is. The result should not be read as a claim about every moment inside the episode.\n\n## What would falsify this\n\nThe central editorial claim would be falsified if the stated replay could not reproduce its PASS from the recorded construction: the registered window, the threshold of STRAINED or worse, the stored data pack, the as-of truncation and the disclosed qualifying-measure rules. It would also fail if the threshold had been revised after the result without a separately visible registration, or if the article treated absent measures as supportive observations.\n\nThe broader interpretation would weaken if a defensible correction to source timing, measure orientation, historical availability or proxy construction changed the peak cell below STRAINED. It would also weaken if comparable calm-period controls showed the same pattern frequently enough to reveal that the construction was chiefly tracking structural trend rather than episode-specific stress. The method must be allowed to fail; otherwise a historical replay is only an after-the-fact illustration.\n\nConversely, a new source or a richer vintage archive could improve a future replay, but it would create a new recorded result rather than silently rewrite this one. The present verdict remains bounded by the evidence package that produced it.\n\n## Method and limits\n\nThis article relies only on the supplied public Undertow and Seiche materials. The replay was generated from today’s stored pack truncated to observations dated on or before each historical as-of. The ledger states plainly that FRED and the New York Fed can revise series and that the product does not maintain an ALFRED-style vintage archive. Accordingly, this is a construction replay on final data, never a live simulation.\n\nA scored tier requires at least two measures with stress percentiles at 60 or more observations each; accumulating or absent measures are disclosed rather than rendered calm. The UST replay had three qualifying measures at its peak. The current IG cell also has three qualifying measures, but its NORMAL reading is not evidence about the historical UST test. Cross-desk readings are comparisons, never one blended score.\n\nNo sealed prospective call is attached to this story. The sealed-call record is separate and verifiable, while the daily board is not silently promoted into a forward call. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com) — public product and historical-replay context.\n- [Undertow API](https://api.seiche.info/undertow) — public analytical-record surface.\n- [Seiche](https://seiche.info) — funding overlay used only as adjacent context.\n- [LiquiLens](https://liquilens.in) — separate institution-liquidity product; not blended with this replay.\n- [Undertow on Telegram](https://t.me/Undertow_LiquiLens_bot) — public access point for the live board.\n- [Undertow public product record](https://liquilens-undertow.com) — source boundary for the UST construction and current board.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-12-autumn-2022-treasury-replay-passed-its-registered-test-but-only-for-the-instrument-tested/", "date": "2026-09-12", "dek": "The UST replay reached STRAINED during the registered window, showing that Undertow’s construction detected stress in its chosen Treasury measures. It does not prove a live forecast, a universal liquidity gauge, or a blended cross-market verdict.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:f6adab8c815d217a4e42fdfa4665af1fdb36fde718d602c40adff09f54865b6c", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:5c8ef944a56413c96156cb4363ae2e62c1faccba06e3811b78b1784359816847", "status": "applied"}, "evidence_asof": "2026-09-12", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "a64f9fd29a5703b4a20fd8b0570ba952c5f8b7e8ecd0151543e25eb7cba856c1", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:f6adab8c815d217a4e42fdfa4665af1fdb36fde718d602c40adff09f54865b6c", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:5c8ef944a56413c96156cb4363ae2e62c1faccba06e3811b78b1784359816847", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["Approved. The draft stays within the supplied public-pack boundary and makes no investment instruction, forward call, or blended cross-product score.", "The historical-replay thesis is narrow and supported: the registered UST threshold was STRAINED or worse within 2022-09-15 to 2022-11-30; the replay records PASS, a 0.8193 peak score on 2022-10-05, and 52 scored as-of dates.", "The article correctly separates the current board’s single scored IG cell (NORMAL, 0.4172, three qualifying measures) from the historical UST replay and from Seiche’s declared EROSION funding overlay.", "The countercase and falsifiers are substantive and dossier-grounded, including trend contamination in expanding-history weekly level percentiles, incomplete inputs, timing limits, source revisions, and the lack of an archival vintage database.", "All required headings are present. The source section contains six Markdown links, each using an approved URL."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "Autumn 2022 Treasury replay passed its registered test—but only for the instrument tested", "published_at": "2026-09-12T21:50:30Z", "quality": {"failures": [], "numeric_tokens_checked": 17, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1443}, "slug": "2026-09-12-autumn-2022-treasury-replay-passed-its-registered-test-but-only-for-the-instrument-tested", "topic": "oct2022_ust_illiquidity", "type": "historical_replay", "word_count": 1443}, "2026-09-13-march-2020-ig-replay-passed-a-narrow-test-it-does-not-validate-a-forecast": {"body_md": "## The read\n\nUndertow’s March 2020 investment-grade, or IG, replay passed its registered test. That is a narrow result: the IG cell reached the required stress threshold during the specified historical window. It is not a certification of the entire crisis narrative, a demonstration that the construction will catch every disruption, or a forecast from the current board.\n\nThe episode window ran from 2020-03-01 through 2020-04-30. Its registered expectation was that the IG cell would reach STRAINED or DEFICIENT on at least one weekly as-of date. The replay verdict was PASS. Its peak reading was DEFICIENT, with a score of 0.9824, on 2020-04-15. All 9 weekly as-of dates in the window scored.\n\nThat result matters because the hurdle was specified before the reported result: STRAINED or worse, rather than a threshold selected after inspecting the trace. But the finding stops at the instrument’s historical behavior. The replay does not establish a general causal account of March 2020, nor does it establish that the same measures will give timely or complete warnings in a later event.\n\nThe current daily reading is separate. On 2026-09-13, 1 of 9 segments scores: IG reads NORMAL with 3 available and qualifying measures and a within-board score of 0.4172. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. Seiche’s declared funding regime is EROSION, but it is a funding-context overlay only, not an input to the IG tier and not part of a blended cross-product score.\n\nThe defensible thesis is therefore limited: the registered IG construction recorded severe conditions in its designated March–April 2020 episode, while the sparse current board and separate EROSION overlay do not justify a claim about what markets will do next.\n\n## The mechanism\n\nThe historical test concerns a defined cell, not a universal measure of liquidity. In the replay record, the IG cell rests on 2 New York Fed primary-dealer measures: dealer corporate net positions, described as congestion, and dealer corporate settlement fails. The two measures were qualifying in the replay’s reported peak reading.\n\nThe construction compares measures with their own accrued histories. In the trace, settlement fails were at elevated percentiles across the reported dates, while dealer corporate net positions rose sharply later in the window. The combined score reached STRAINED on 2020-03-25 and DEFICIENT on 2020-04-01, remaining DEFICIENT through the final reported weekly date of 2020-04-29.\n\nThat is not a price-only account, and it is not an all-in credit-liquidity composite. It asks whether the selected dealer-position and settlement-fails measures were extreme under the board’s own historical comparison method. It does not claim to represent every transaction, quote, fund flow, dealer decision or market dislocation.\n\nThe missing evidence is material. The historical pack had no qualifying IG or HY option-adjusted-spread observations for the episode: the stated available history for those series begins on 2023-09-12. The replay disclosed those absences rather than filling them with substituted data. Its IG result consequently rests on the 2 qualifying New York Fed measures, not on an implied combination with unavailable spread data.\n\nThe Seiche overlay cannot serve as confirmation of the historical replay. The dossier describes Seiche’s index as a calibrated weighted blend of engine sub-scores, with high values indicating stress; it is not a percentile of its own history. Its stress percentile is withheld rather than invented. EROSION is a declared regime, and the Undertow record explicitly marks it as context only.\n\n## The countercase\n\nThe strongest countercase is methodological. A crisis replay can pass without making the instrument complete, robust across regimes, or predictive. The replay cautions that its New York Fed measures are percentiles of weekly levels against expanding histories that begin in 2013. Structural growth in dealer balance sheets can therefore place trend alongside stress in a level percentile.\n\nThe record provides a concrete warning: a Q2-2017 negative control fails for that reason. This does not undo the March 2020 PASS, but it prevents a high percentile from being treated as a pure or self-explanatory measure of acute dysfunction.\n\nTiming is another limitation. The trace is weekly, and the source record assigns the New York Fed data an 8-day publication lag. A dislocation could peak between weekly dates, emerge first in a measure not represented in the cell, or be incompletely described by the two measures that were available for this episode.\n\nThe present board also cannot settle the question. IG’s NORMAL tier is not evidence that every segment is calm, because 8 segments are still accruing history and do not score. Conversely, Seiche’s EROSION regime does not prove current IG stress. These are separate products with separate methods, and the dossier prohibits blending them into one score.\n\n## What would falsify this\n\nThis article’s central claim would be false if the documented replay did not show a PASS against its registered expectation, or if the stated peak of DEFICIENT at 0.9824 on 2020-04-15 were absent from the replay record. It would also fail if the episode did not contain 9 scored weekly as-of dates, as reported.\n\nThe narrower mechanism claim would be weakened if the disclosed New York Fed measures were not qualifying at the peak, or if a corrected historical pack changed the reported replay outcome. A correction would need to appear in a visible revised record; it cannot be assumed from a narrative revision.\n\nA broader claim of robustness would require more than this single PASS. It would require further registered episodes and meaningful controls, including evidence that the construction is not simply tracking structural level trends. This article makes no such broader claim and makes no forecast for IG, funding or any other segment.\n\n## Method and limits\n\nThis is a historical replay, not a live simulation. The replay recomputes readings from today’s stored pack after truncating observations to dates on or before each historical as-of. The result is therefore a construction replay using final stored data, not proof of the exact data vintage available to market participants at the time.\n\nThe record specifically cautions that FRED and New York Fed series can be revised and that the product has no ALFRED-style vintage archive. Its reported knowledge dates should not be mistaken for preserved contemporaneous publication vintages.\n\nA scored tier requires at least 2 measures with stress percentiles at 60 or more observations each. Measures that are accruing or absent are to be disclosed, not rendered as calm. A tier is a deterministic board derivation, not an observed price or a trade signal. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com)\n- [Undertow dispatch record for 2026-09-13](https://liquilens-undertow.com/dispatch/2026-09-13.json)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Undertow sealed-calls record](https://api.seiche.info/undertow/sealed_calls.json)\n- [Seiche](https://seiche.info)\n- [Seiche dispatch source cited for the funding overlay](https://seiche.info/dispatches/news.json)\n- [LiquiLens](https://liquilens.in)\n- [Undertow Telegram channel](https://t.me/Undertow_LiquiLens_bot)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-13-march-2020-ig-replay-passed-a-narrow-test-it-does-not-validate-a-forecast/", "date": "2026-09-13", "dek": "Undertow’s registered investment-grade replay reached DEFICIENT during the March 2020 window. The result supports a limited claim about this construction on final stored data, not a claim about today’s quiet board or future market conditions.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:a0fec44d486fcf2e6b3b665f04d62b0107e8f0612530672dcbc024ef551e6fd6", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:ff6bf48ff3297c9ee5cfa160c269ee5bd9f98f43f286def0f6f210747f5df6bd", "status": "applied"}, "evidence_asof": "2026-09-13", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "32dc4037df08c84c4911d17abd85d47454fab35e8bf1eab51360824f5b9f5e18", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:a0fec44d486fcf2e6b3b665f04d62b0107e8f0612530672dcbc024ef551e6fd6", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:ff6bf48ff3297c9ee5cfa160c269ee5bd9f98f43f286def0f6f210747f5df6bd", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft is substantively close, but its Sources section pads the required link count by repeating the same Undertow API URL.", "It describes LiquiLens material as an adjacent comparison and Telegram as an access point for the live board without support for those characterizations in the dossier.", "The repair retains the narrow, falsifiable historical-replay thesis; states the current-board and funding-overlay boundary; and supplies distinct dossier-listed or dossier-cited source URLs."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "March 2020 IG replay passed a narrow test; it does not validate a forecast", "published_at": "2026-09-13T21:49:05Z", "quality": {"failures": [], "numeric_tokens_checked": 23, "required_headings": 6, "source_links": 8, "status": "PASS", "word_count": 1141}, "slug": "2026-09-13-march-2020-ig-replay-passed-a-narrow-test-it-does-not-validate-a-forecast", "topic": "mar2020_dash_for_cash_ig", "type": "historical_replay", "word_count": 1141}, "2026-09-14-march-2020-treasury-replay-cleared-its-registered-bar-with-material-limits": {"body_md": "## The read\n\nUndertow’s registered replay of the March 2020 Treasury episode passed its stated bar. That is the claim, and it should remain narrow.\n\nThe registered expectation was that the UST cell would reach STRAINED or DEFICIENT on at least one as-of within the episode window. The replay’s peak was DEFICIENT, with a score of 0.9151, on 2020-04-08. The result records 43 scored as-of dates in the window, and the verdict is PASS.\n\nThis does not mean the instrument forecast the episode, that it would identify every future liquidity event, or that it measures all forms of market stress. It says that, on the product’s stated construction and historical replay convention, the UST cell cleared a pre-registered test in an episode selected as a hard bar for Treasury-market stress.\n\nThe current board is separate from that historical result. On 2026-09-14, 1 of 9 segments scores. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history, while IG reads NORMAL. Seiche’s funding overlay is declared EROSION. It is funding context only: the dossier explicitly says it is not used in an Undertow tier and provides no blended cross-product score.\n\nThat separation is essential. A daily board reading is not silently promoted into a forward call, and the current EROSION overlay does not imply a Treasury outcome.\n\n## The mechanism\n\nThe registered episode rationale identifies March 2020 as a relevant test for the UST construction. It describes customer Treasury selling overwhelming dealer intermediation capacity, dealer inventories rising as paper was absorbed, and settlement fails spiking before the Federal Reserve’s 23 March purchases relieved the market.\n\nThe replay’s UST evidence includes dealer net positions in bills and coupons and settlement fails, defined as fails to deliver plus fails to receive. At the 2020-04-08 peak, the reported qualifying measures were dealer UST bills net positions, dealer UST coupon net positions and dealer UST settlement fails. The reported peak set also contains a null two-year-to-ten-year slope field and a null aggregate dealer UST net-positions field.\n\nThe path was not uniformly stressed. Earlier March as-of dates were NORMAL. The replay reached STRAINED on 2020-03-11 and DEFICIENT on 2020-04-08. That sequence supports the limited finding that the construction registered sufficiently elevated dealer-position congestion and settlement-fails evidence, against each measure’s accrued own history, to clear the episode bar.\n\nIt is not a causal account of the whole event. Nor is it a price observation relabeled as liquidity. The dossier describes a tier as a deterministic board derivation rather than an observed price or a trade signal.\n\n## The countercase\n\nA pass in a famous crisis window can be less persuasive than it first appears. A construction can rise in a severe episode because its inputs are sensitive to structural trends as well as stress, or because a proxy captures conditions associated with liquidity rather than executable liquidity itself. Passing this test therefore does not certify Undertow as a universal crisis detector.\n\nThe replay documentation supplies a specific concern. Dealer-position and fails measures are percentiles of weekly levels against expanding own histories beginning on 2013-04-03. The documentation warns that dealer balance sheets grew structurally over that period, so a level percentile can reflect trend as well as stress. It further says that the documented second-quarter 2017 negative control fails for that reason.\n\nThat limitation matters beside the March pass. A construction that flags a severe event but can also be affected by structural growth requires cautious interpretation. The March result establishes that the registered UST bar was met; it does not resolve this trend-sensitivity concern.\n\nCoverage narrows the inference further. The replay records that HY OAS and IG OAS were absent for the March 2020 test because their series histories in this pack begin in 2023. Those absent legs cannot validate, explain or broaden the UST result. This is a Treasury replay, not evidence that credit measures confirmed the same episode.\n\nThe timing record is also a limit. The registered condition required at least one qualifying as-of in the window, not immediate or continuous detection. Earlier readings were NORMAL before later readings reached STRAINED and DEFICIENT. The PASS therefore should not be presented as a complete chronology of market conditions.\n\n## What would falsify this\n\nThe narrow editorial claim would fail if the disclosed record did not show the registered UST expectation, the reported PASS, or the stated peak of DEFICIENT with a 0.9151 score on 2020-04-08. It would also fail if a reproduction from the disclosed stored pack under the reported truncation convention produced a materially different result.\n\nThe more limited inference—that this construction showed sensitivity to this Treasury-stress episode—would weaken if a defensible correction to source coverage, timing, measure orientation or historical treatment removed the PASS. It would also weaken if calm-period controls repeatedly produced stressed readings for the structural reasons the documentation already identifies.\n\nAnother successful historical episode would be additional episode-specific evidence, not proof of robustness. Conversely, disclosed missing inputs, failures and negative controls are not incidental: they are part of the evidence boundary.\n\n## Method and limits\n\nThis is a construction replay on final stored data, not a live simulation of what participants saw at the time. For each historical as-of, the product recomputes a reading from the current stored pack truncated to observations dated on or before that as-of. The dossier describes this as knowable-by-date, not as the contemporaneous published vintage.\n\nThat distinction is material because FRED and New York Fed series may be revised, while the product does not hold an ALFRED-style vintage archive. The replay verdict should therefore be treated as a construction replay on final data, not as proof of a real-time signal available in March 2020.\n\nA scored tier requires at least 2 measures with stress-percentile histories of 60 or more observations. Accruing and absent measures are disclosed rather than rendered as calm. The free board withholds some percentile detail, and this article does not reconstruct withheld values or claim precision beyond the public pack.\n\nSeiche’s declared EROSION regime is also bounded. Its composite index is described as a calibrated weighted blend of engine sub-scores, with high values indicating stress; its own-history stress percentile is withheld rather than invented. It remains a separate funding overlay, not an Undertow percentile or a synthetic cross-market signal.\n\nThis article is limited to the public pack and the sources below. It offers research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com)\n- [Undertow dispatch record](https://liquilens-undertow.com/dispatch/2026-09-14.json)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow Telegram bot](https://t.me/Undertow_LiquiLens_bot)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-14-march-2020-treasury-replay-cleared-its-registered-bar-with-material-limits/", "date": "2026-09-14", "dek": "Undertow’s UST replay reached DEFICIENT in its registered March 2020 window. The result supports a narrow construction test, not a forecast, a universal detector or a cross-product score.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:80129be1610ea2c421aefd266db52a53de9db69b21ff3b97ee5f1168e31e5cfd", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:9fa0ef24046d875f36373acf7a323b5afc7777776a609d013ab1bb8bf04b2405", "status": "applied"}, "evidence_asof": "2026-09-14", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "447818951827217051d1bf74c1b20233618fc092bc3796cac7787f8b5b6d0980", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:80129be1610ea2c421aefd266db52a53de9db69b21ff3b97ee5f1168e31e5cfd", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:9fa0ef24046d875f36373acf7a323b5afc7777776a609d013ab1bb8bf04b2405", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft calls the replay path a sequence of “weekly as-of” readings while also stating that all 43 scored as-of dates received scores. The dossier’s trace is dated at daily intervals; “weekly” is only the registered expectation’s wording. The article should not characterize the full trace as weekly.", "The draft says the two-year-to-ten-year slope was “unavailable” at the peak. The reported peak measure set shows that field as null, but the dossier does not establish the stronger explanation used in the draft. It should be described as null in the reported peak set.", "The draft’s mechanism section risks implying that the listed dealer-position and settlement-fails measures exhaust the UST construction, although the replay reports five measures, including a null aggregate dealer-position field and a null slope field. The repaired copy distinguishes the measures supporting the peak from the entire reported set.", "Repairs are needed to tighten the historical-replay framing, retain the countercase and falsifiers, and state the public-pack and final-data research boundary without overclaiming."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "March 2020 Treasury replay cleared its registered bar, with material limits", "published_at": "2026-09-14T21:49:57Z", "quality": {"failures": [], "numeric_tokens_checked": 18, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1095}, "slug": "2026-09-14-march-2020-treasury-replay-cleared-its-registered-bar-with-material-limits", "topic": "mar2020_dash_for_cash_ust", "type": "historical_replay", "word_count": 1095}, "2026-09-15-august-2024-replay-passed-its-vix-test-not-an-equity-tier": {"body_md": "## The read\n\nUndertow’s historical replay of the August 2024 yen-carry unwind produced a narrow result: the registered VIX test passed, but the EQUITY cell did not score. The distinction defines the finding.\n\nThe registered window ran from 2024-07-25 to 2024-08-16. Its expectation was measure-specific: VIX close had to reach a point-in-time own-history stress percentile of at least 0.95 on at least one day in that window. The replay recorded a peak percentile of 0.973 on 2024-08-05 and returned a PASS verdict.\n\nThat result supports only the stated proposition. In this construction replay, VIX became extreme against its accrued own history during the registered episode. It does not establish a scored EQUITY tier, a complete reading of equity liquidity, a causal account of the episode, or a forward-looking market call.\n\nAt the peak as-of date, the EQUITY evidence had one qualifying measure: VIX close. The Amihud and lambda measures were null for the historical score, leaving the cell PARTIAL with no score and no tier. Undertow’s policy requires at least 2 qualifying measures, each with stress-percentile history of 60+ observations, before it renders a scored tier. A high reading in one measure does not satisfy that gate.\n\n## The mechanism\n\nThe replay was registered at measure level because the EQUITY cell could not meet its coverage requirement at a 2024 as-of. The monthly Amihud pack begins in 2020-08, while the relevant gate requires 60 months of history. The registered question therefore concerned the VIX series rather than a cell-level composite.\n\nThe replay’s rationale notes that VIX closed at 38.57 on 2024-08-05, its highest close since 2020. The result measured that observation against VIXCLS daily history dating from 1990. The registered threshold was 0.95; the reported percentile was 0.973.\n\nFor each replay date, the construction uses observations dated on or before that as-of date. All 17 as-of dates in the window carried a VIX percentile, while none produced a scored EQUITY as-of. The published result therefore distinguishes a working measure-level percentile calculation from a coverage-complete segment verdict.\n\nThe current funding overlay does not change that result. The 2026-09-15 board declares the Seiche funding regime EROSION, but the dossier defines it as a separate funding-context overlay, not an input to Undertow’s liquidity tier. Its composite is a calibrated weighted blend of engine sub-scores, not a percentile of its own history; its stress percentile is withheld rather than inferred.\n\n## The countercase\n\nThe principal countercase is straightforward: a VIX result can be extreme in a volatility episode without demonstrating broad equity-market liquidity conditions. The replay itself preserves that objection. Its other EQUITY measures did not qualify, and the cell remained PARTIAL throughout the window.\n\nThat limitation is material rather than cosmetic. The available record does not permit the PASS to be converted into a retroactive full-segment stress tier. Nor does it show that VIX alone captured trading friction, depth, market impact, or every other potential liquidity channel.\n\nThe equity-liquidity pack also carries an explicit survivorship limitation. Its monthly historical cross-section is backfilled for a basket committed on 2026-07-17, with knowledge of which names survived. Truncating observations by date does not remove that membership issue. The replay is point-in-time in the dates it admits, but the dossier says it is survivorship-conditioned in membership.\n\nThere is also no contemporaneous vintage archive. The replay uses today’s stored pack truncated to data dated on or before each historical as-of. FRED and New York Fed series can be revised, and Undertow does not hold an ALFRED-style vintage archive. The verdict is therefore a construction replay on final data, not a claim that a user in August 2024 saw the same data vintage.\n\n## What would falsify this\n\nThis article’s claim is testable: the registered VIX measure passed its 0.95 floor in the stated window, while the EQUITY cell had one qualifying measure and remained PARTIAL without a score or tier.\n\nIt would be falsified if the public replay record did not show a 0.973 VIX percentile on 2024-08-05, if the registered expectation did not specify a 0.95 floor before assessment, or if the trace showed at least 2 qualifying EQUITY measures and a scored cell. It would also be falsified if the funding overlay were part of the Undertow replay calculation rather than the separately disclosed context overlay.\n\nA correction to the stored data, percentile calculation, dating, or registration record that removed the threshold breach would overturn the measure-level PASS. Additional qualifying evidence could support a later cell-level result, but it would be a new recorded result, not a revision of this replay’s stated coverage.\n\n## Method and limits\n\nThis article is confined to the dossier and the approved public product surfaces listed below. It makes no claim beyond that public pack, gives no investment instruction, and does not construct a blended cross-product score.\n\nThe historical replay’s window is 2024-07-25 through 2024-08-16. Its final verdict is PASS at measure level; its EQUITY evidence is PARTIAL at cell level, with one qualifying measure, no score, and no tier. Missing or still-accruing evidence is disclosed rather than treated as calm.\n\nThe 2026-09-15 dispatch is context only. It reports that 1 of 9 segments scored, identifies IG as NORMAL, and declares the separate funding overlay EROSION. Those current-board observations neither revise the August 2024 replay nor create a causal or predictive link to it.\n\nUndertow states that a tier is a deterministic board derivation, not an observed price or trade signal. Research and market data are not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com)\n- [Undertow public API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow API replay surface](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-15-august-2024-replay-passed-its-vix-test-not-an-equity-tier/", "date": "2026-09-15", "dek": "A registered replay of the August 2024 yen-carry unwind found a 0.973 point-in-time VIX stress percentile on 2024-08-05, clearing its 0.95 floor. The EQUITY cell remained PARTIAL, with one qualifying measure and no score or tier.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1ce665d1d5a4b0918333c3a31dbdffc357354324839c5c7869c2a93002651473", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:da2474bf2651375c641b400fd9119647494cc6bf1efea8d920ad761e3b7d9e91", "status": "applied"}, "evidence_asof": "2026-09-15", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "077321e4a5ceb10b9c7120fe5c3f87d60043dc8f751ff727842d8e645c6b4bbe", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1ce665d1d5a4b0918333c3a31dbdffc357354324839c5c7869c2a93002651473", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:da2474bf2651375c641b400fd9119647494cc6bf1efea8d920ad761e3b7d9e91", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft calls the Seiche overlay a “cross-desk comparison.” The dossier supports a separate cross-product, funding-context overlay only; it does not support that characterization.", "The draft says the 0.95 threshold was “intentionally lower than certainty.” That rationale is not in the dossier and is editorial inference.", "The draft describes the non-VIX measures as unavailable. The record establishes that they were non-qualifying/null for the historical score; the repair uses that narrower, documented formulation.", "The research boundary should expressly limit the article to the approved public product surfaces. The repaired version does so.", "The draft supplies six Markdown link instances, but only five distinct approved source URLs. The repair retains at least six approved Markdown links while making the source boundary explicit and removing unsupported framing."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "August 2024 replay passed its VIX test, not an EQUITY tier", "published_at": "2026-09-15T21:49:30Z", "quality": {"failures": [], "numeric_tokens_checked": 19, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 948}, "slug": "2026-09-15-august-2024-replay-passed-its-vix-test-not-an-equity-tier", "topic": "aug2024_yen_carry_vix", "type": "historical_replay", "word_count": 948}, "2026-09-16-the-august-fx-replay-is-inconclusive-because-the-instrument-had-not-yet-earned-a-verdict": {"body_md": "## The read\n\nThe useful conclusion from Undertow’s historical replay is not that the August yen-carry unwind was uneventful, nor that an FX liquidity instrument failed a known episode. It is narrower: the registered FX test could not issue a verdict for that episode because the cell had not accumulated the qualifying history required for a scored tier.\n\nThe replay covers the registered August yen-carry unwind — FX episode, from 2024-07-25 through 2024-08-16. Its expectation was explicit: the FX cell would reach STRAINED or DEFICIENT on at least one as-of inside that window. The result is INCONCLUSIVE. There was one as-of in the replay trace, 2024-07-31, and it was PARTIAL rather than scored. Neither of the two FX measures qualified. Amihud and lambda-hat were present but withheld; the replay therefore recorded no score and no tier beyond PARTIAL.\n\nThat distinction is the thesis. A withheld cell is an absence of verdict, never a pass. The replay describes the instrument’s historical capability, not whether the underlying episode occurred and not whether history was calm. It also does not supply a retrospective forecast record for the current board.\n\nThis is particularly important because the current Undertow dispatch is itself a quiet-board watch note rather than a full market story. Its public reading says one of nine segments scores today, identifies IG liquidity as NORMAL, and records the Seiche funding overlay as EROSION. Those are separate current observations. The funding regime is a comparison context only; it is not blended into an Undertow tier or into the FX replay verdict. The historical conclusion stands on the replay’s own coverage record.\n\n## The mechanism\n\nThe registered episode was selected to test a specific measurement construction, rather than to create a general narrative around a prominent market event. Its frozen rationale identifies an unwind of yen-funded carry positions as the proximate cause of the global equity and volatility shock associated with the episode. The relevant registered question was whether the FX cell could register stress in the stated window.\n\nThe instrument could not answer that question at the required standard. The FX cell relies on Amihud and lambda-hat in an FX ETF-proxy bucket. Each measure requires 60 monthly observations before it can qualify for a scored tier. The monthly history begins in 2020-08, placing the earliest scoreable FX as-of in mid-2025. The episode window predates that point. On the sole traced as-of, both measures were present but withheld, leaving zero qualifying measures out of two.\n\nUndertow’s policy requires at least two qualifying measures for a scored tier. That rule matters because an accruing measure is disclosed rather than rendered as calm. The replay did not infer a percentile from insufficient history, substitute a different measure, or convert the absence of qualifying evidence into a reassuring classification. The result is not a diluted STRAINED reading. It is no reading at all under the instrument’s stated tier rules.\n\nThe episode was registered before the replay result was inspected. That ordering makes the finding more useful than an after-the-fact threshold choice. The recorded expectation, segment, window, and test condition existed before the outcome. The eventual INCONCLUSIVE verdict was anticipated as a possible result because the FX history was known to be too short for the specified coverage requirement. A test can be appropriately designed to reveal that its own historical reach is limited.\n\nThe replay also has a second, material boundary. It uses a truncated-latest construction: readings are recomputed from the stored pack using observations dated on or before each as-of. The ledger says this is what was knowable by date, but not the vintage published at the time. It does not hold an ALFRED-style vintage archive, and the cited source series can be revised. In addition, the equity-liquidity pack declares look-ahead in the cross-section because its monthly history is backfilled for a basket committed later, making membership survivorship-conditioned. Those limits do not turn INCONCLUSIVE into a positive signal; they constrain what any eventual historical conclusion could mean.\n\n## The countercase\n\nThe strongest countercase is not that the replay should have passed. It is that the registered construction was too narrow, too early in its own history, or too proxy-dependent to observe the channel that mattered during the episode. FX stress may be expressed in information outside this cell, outside its ETF-proxy bucket, or between the available as-of dates. The dossier does not establish those possibilities as facts, but it explicitly leaves room for them by refusing to treat a withheld cell as calm.\n\nA second countercase concerns the event-to-instrument comparison itself. The frozen rationale says FX is one of the segments carrying a scored tier on the live board, but the historical test concerns an earlier period when the FX measures had not yet met the same scoring standard. Present-day scoreability cannot be projected backward. A live instrument can be eligible now while remaining historically unevaluable for an older window.\n\nNor should the current funding overlay rescue the result. Seiche’s EROSION regime is declared on its own funding surface. Its composite is not a percentile of its own history, and the stress percentile is withheld rather than invented. Undertow marks that relation as a funding-context overlay only and says it was not used in the tier. Reading EROSION as retroactive proof of an FX tier would violate both products’ stated boundaries.\n\n## What would falsify this\n\nThis editorial makes a limited claim: the published INCONCLUSIVE verdict accurately reports what the registered FX replay could establish for its stated window. That claim would be falsified if the public replay record did not show the registered expectation, the one traced as-of, PARTIAL status, zero qualifying measures, or the withheld Amihud and lambda-hat measures. It would also be falsified if the result had been silently substituted for a different registration or window.\n\nA later dataset with sufficient eligible history could support a new replay, but it would need a new recorded vintage and result. It could not convert this historical non-verdict into a pass, fail, or prediction by relabeling the old evidence. Likewise, independent point-in-time evidence might illuminate the episode, but it would be separate evidence rather than a repair to this replay.\n\n## Method and limits\n\nThis article uses only the supplied public Undertow and adjacent-product record. It does not reconstruct withheld measures, assign a score to a PARTIAL cell, calculate a cross-product score, or infer a missing percentile. The FX replay verdict is INCONCLUSIVE because the instrument did not satisfy its own coverage policy in the episode window.\n\nThe current board and historical replay answer different questions. The board is a deterministic derivation, not an observed price or trade signal. The replay is a construction replay on final stored data, not a live simulation or a claim about the untouched contemporaneous vintage. The Seiche overlay remains a comparison, never a blended score. Research and market data; not investment advice.\n\n## Sources\n\n- [Undertow public surface](https://liquilens-undertow.com)\n- [Undertow API and replay records](https://api.seiche.info/undertow)\n- [Seiche funding context](https://seiche.info)\n- [LiquiLens market context](https://liquilens.in)\n- [Undertow Telegram board access](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow replay reference](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-16-the-august-fx-replay-is-inconclusive-because-the-instrument-had-not-yet-earned-a-verdict/", "date": "2026-09-16", "dek": "The registered test for the August yen-carry unwind did not show a scored FX tier inside its window. That does not revise the event; it defines the historical boundary of an instrument whose required measures were withheld.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1186134605e216792bf4315bbcedfddddfb7f8990b52cb67eef70bb17dfe34a4", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:a10dca9b7ee2799c9b0b96a0c4529e6622e0b06efdc6a36d8e7cf280ed94c430", "status": "applied"}, "evidence_asof": "2026-09-16", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "7b8186d3224e6ac89e245f19f00fd66f33dd4914e1f4061ae20140be6b5f9b80", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1186134605e216792bf4315bbcedfddddfb7f8990b52cb67eef70bb17dfe34a4", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:a10dca9b7ee2799c9b0b96a0c4529e6622e0b06efdc6a36d8e7cf280ed94c430", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": [], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The August FX replay is inconclusive because the instrument had not yet earned a verdict", "published_at": "2026-09-16T21:49:43Z", "quality": {"failures": [], "numeric_tokens_checked": 9, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1180}, "slug": "2026-09-16-the-august-fx-replay-is-inconclusive-because-the-instrument-had-not-yet-earned-a-verdict", "topic": "aug2024_yen_carry_fx", "type": "historical_replay", "word_count": 1180}, "2026-09-17-ig-moves-easier-by-0-05-inside-normal-funding-reads-erosion": {"body_md": "## The read\n\nThe narrow conclusion from Undertow’s current reading is not that investment-grade credit liquidity is calm, and not that the funding overlay supplies a unified cross-market verdict. It is that IG remains classified NORMAL after moving easier by 0.05, while qualifying measures disagree within the scored cell and the separate Seiche funding overlay is declared EROSION.\n\nIG is the only scored segment on the board: 1 of 9 segments scores today. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. That limits the evidence boundary. The packet does not support a blended cross-market liquidity score, and it does not permit an inference that the unscored segments are calm.\n\nIG’s within-board score is 0.368. The board reports 3 available measures and 3 qualifying measures for the segment. The dispatch also states that qualifying measures disagree in at least 1 scored cell and that the span is withheld for every such cell, because publishing it would reveal the percentile of a row withheld from the free board.\n\nNORMAL is therefore a board classification, not a claim of unanimity among the qualifying IG inputs. EROSION is a separate declared funding regime. The evidence braid labels Seiche as a funding-context overlay only and says it was not used in the Undertow tier. The two outputs can be compared as distinct published readings, but they cannot be converted into one common score.\n\n## The mechanism\n\nUndertow describes a tier as a deterministic board derivation, not an observed price and not a trade signal. Its measures are evaluated against their own accrued point-in-time histories, with higher stress percentiles oriented toward more stress. The framework can therefore report an IG tier while also disclosing disagreement among qualifying inputs.\n\nOne disclosed IG measure is IG OAS, identified as BAMLC0A0CM, with a stress percentile of 0.1552 as of 2026-09-16. The dossier’s limitation is consequential: keyless ICE BofA downloads through FRED are capped to a trailing roughly 3-year window. The published percentile omits earlier crisis regimes identified in the dossier and can understate rarity in calm tapes. It should not be treated as a full-cycle rarity estimate.\n\nDealer corporate net positions are also disclosed as an IG measure, but their stress percentile is withheld. The series is explicitly a proxy rather than an IG-only measure because it covers the corporate-credit complex, including commercial paper, IG and below-IG bonds. Its published interpretation is that high net dealer inventory indicates balance-sheet congestion and less capacity to absorb customer selling. The dossier also says the levels are spliced across NY Fed series breaks. Those constraints rule out assigning an undisclosed percentile or treating the measure as a pure IG reading.\n\nThe withheld span matters. It means the public evidence establishes that qualifying measures disagree, but does not establish the size of that disagreement. The score cannot be used to reconstruct withheld component percentiles.\n\nSeiche has a different construction. Its composite is 42.8 out of 100, its coverage is 100%, and its declared regime is EROSION. The dossier says this is a calibrated weighted blend of engine sub-scores, where higher is more stressed, rather than a percentile of its own history. Its stress percentile is withheld rather than invented. The 42.8 out of 100 composite and the 0.368 IG within-board score are not interchangeable quantities.\n\n## The countercase\n\nThe strongest countercase is that NORMAL is the appropriate summary of IG under the published board rules, despite the disclosed disagreement. The score moved easier by 0.05, remained NORMAL, and the packet reports 3 qualifying measures. The disclosed OAS stress percentile is low on its available history. Nothing in this dossier supports reclassifying the IG tier as stressed.\n\nA methodological countercase is also available. The OAS percentile is constrained by the trailing roughly 3-year download window, while dealer corporate net positions are a broad corporate-credit proxy with spliced history. Disagreement can reflect differing series scope and history rather than one common deterioration in IG liquidity. Since the span is withheld, the public pack does not establish the magnitude or source of the divergence.\n\nThe funding comparison has an independent limitation. Seiche is context only and was not used in the Undertow tier. EROSION can coexist with a NORMAL IG tier without contradiction because the products publish different outputs under different constructions. It would exceed the evidence to say that funding conditions caused the IG tier, that the IG move disproves EROSION, or that the two readings constitute a common market signal.\n\n## What would falsify this\n\nFor the dated claim, the relevant falsifier is a correction or revision to the point-in-time Undertow record showing that IG was not NORMAL, did not move easier by 0.05, did not have 3 qualifying measures, or did not contain disclosed qualifying-measure disagreement on 2026-09-17. A correction to the Seiche record showing that its declared regime was not EROSION would likewise remove the stated funding context.\n\nA later board reading without disagreement would not falsify the historical description of this reading. It would, however, end any attempt to extend this article’s narrow present-tense interpretation beyond the dated record. Likewise, a later change in Seiche’s regime would describe a new overlay reading, not revise the 2026-09-17 IG classification.\n\nNo replay verdict is attached to this daily reading. The dispatch labels adjudication as NOT_ATTACHED_TO_THIS_DAILY_READING and says a daily board reading is not silently promoted into a forward call. Subsequent market outcomes therefore cannot validate or invalidate the article as a forecast, because this article makes no forecast.\n\n## Method and limits\n\nThis article uses the published 2026-09-17 Undertow dispatch and its point-in-time board evidence only. The event time is 2026-09-17. The dispatch publication time is 2026-09-17T21:48:18+00:00. No withheld percentile is inferred, no cross-product composite is calculated, and no accruing segment is described as calm.\n\nThe board’s stated qualification rule requires at least 2 measures with a stress percentile at 60+ observations each for a scored tier. PARTIAL means still accruing; an accruing or absent measure is disclosed rather than rendered as calm. The board reports IG as scored, while the other 8 segments remain in history accrual.\n\nThis is research and market data, not investment advice. It offers no investment instruction, price forecast, or claim beyond the public pack.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow Telegram board access](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow canonical dispatch, 2026-09-17](https://liquilens-undertow.com/dispatch/2026-09-17.json)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-17-ig-moves-easier-by-0-05-inside-normal-funding-reads-erosion/", "date": "2026-09-17", "dek": "IG is the only scored segment in the current board, but its NORMAL tier does not imply agreement among qualifying measures. The separate Seiche funding overlay is declared EROSION and is context only, not an input to the IG tier.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:d88bce42d3ceb852132ed34177971ee3bb51538dfb8cc7766f049b28bbe53ae6", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:18e5ca73328d785f053addb058fdde1698f0e0b742266324fb6b412bdac6f33b", "status": "applied"}, "evidence_asof": "2026-09-17", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "006b1ca27c7d6bc489ae5c31e617b00425bfe3ce078e5ceafdc09d2e5cce79e4", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:d88bce42d3ceb852132ed34177971ee3bb51538dfb8cc7766f049b28bbe53ae6", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:18e5ca73328d785f053addb058fdde1698f0e0b742266324fb6b412bdac6f33b", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft says the tier “averages qualifying readings.” The dossier calls a tier a deterministic board derivation but does not specify averaging; that mechanism must not be asserted.", "The falsifier is not logically sound for a dated reading: a later IG reading without disagreement would not falsify the documented disagreement on 2026-09-17. Falsification must concern a correction, revision, or contradiction of the point-in-time record, rather than subsequent conditions.", "The final “canonical dispatch” link points to the general Undertow site rather than the dossier’s canonical 2026-09-17 dispatch URL.", "Repairs below retain the required structure, distinguish the funding overlay from the IG tier, state the research boundary, and provide six dossier-approved or dossier-canonical Markdown links."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "IG moves easier by 0.05 inside NORMAL; funding reads EROSION.", "published_at": "2026-09-17T21:49:14Z", "quality": {"failures": [], "numeric_tokens_checked": 19, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1079}, "slug": "2026-09-17-ig-moves-easier-by-0-05-inside-normal-funding-reads-erosion", "topic": "IG", "type": "current_analysis", "word_count": 1079}, "2026-09-18-the-march-high-yield-replay-is-not-a-failed-crisis-test-it-is-an-exposed-coverage-gap": {"body_md": "## The read\n\nThe narrow conclusion from Undertow’s historical replay is not that March high-yield stress was absent, muted, or somehow contradicted by the record. It is that Undertow’s registered HY instrument cannot be evaluated over its specified March event window with the data pack it currently holds. The verdict is **NOT_RUNNABLE**. That is the finding.\n\nThe registered episode is titled “March dash for cash — high yield.” Its expectation was explicit: the HY cell would reach STRAINED or DEFICIENT on at least one as-of in the stated window. The replay did not produce a pass or a fail because it had no eligible HY observation dated inside that window. The ledger states that the window predates every series the product holds for the segment.\n\nThat distinction matters because a historical replay has only one legitimate job: assess a named instrument against a preregistered expectation and a defined historical interval. It does not adjudicate whether history occurred. It does not reduce a known dislocation to a missing-data artifact. And it does not permit an editor to replace unavailable evidence with a persuasive retrospective narrative.\n\nThe current Undertow dispatch offers a useful contrast, but not a blended signal. Its public board has one scored segment, IG, which reads NORMAL on its within-board score. The remaining listed segments are still accruing history. Separately, Seiche’s funding overlay is declared EROSION. That funding context is an overlay only; it is not included in Undertow’s tier derivation, and it cannot repair the unavailable historical HY inputs. [Undertow](https://liquilens-undertow.com) and [Seiche](https://seiche.info) are adjacent desks with distinct measurement surfaces, not ingredients in a single score.\n\nThe thesis, then, is deliberately narrow and falsifiable: the published NOT_RUNNABLE verdict accurately identifies a coverage boundary in Undertow’s HY replay. Its value is not that it offers a clean historical call. Its value is that it refuses to manufacture one.\n\n## The mechanism\n\nA replay needs more than an event that appears important in hindsight. It needs observations for the relevant instrument, dated inside the registered window, and a method capable of applying the instrument’s rules to those observations. Without that chain, a pass would be invented and a fail would be misleading.\n\nUndertow’s replay record makes the break explicit. The HY window falls before the stored coverage for the relevant series. The ledger identifies the absence directly: no series feeding HY contains a single observation inside the episode’s date range. The result is therefore not an ambiguous weak reading. It is an inability to run the stated test.\n\nThe episode rationale records why the gap exists. The relevant FRED-served ICE BofA high-yield series is available to this product only as a trailing window, leaving the stored pack without the historical observations needed for the registered episode. The rationale also notes that the episode was registered despite the expected limitation so that the gap would be visible and countable rather than silently omitted.\n\nThat is a stronger editorial posture than retrofitting a substitute series after seeing the outcome. A replacement input might be informative in a new study, but it would not be the same registered replay. Changing the input, source availability, or construction after the fact changes the instrument being judged. It cannot silently convert NOT_RUNNABLE into a historical success or failure.\n\nThe replay’s vintage boundary is equally important. Undertow describes its historical readings as recomputed from today’s stored pack, truncated to observations dated on or before each historical as-of. That is a construction replay using final stored data, not a claim that the product holds a contemporaneous archive of every publication vintage. The ledger specifically warns that revisions can affect series and that no ALFRED-style vintage archive is held. A result from such a process must be read as a test of this construction’s historical reach, not as proof of what market participants saw in real time.\n\n## The countercase\n\nThe strongest countercase is straightforward: an instrument can be inadequate for a historical episode even when the episode itself was severe. That is fully consistent with this result.\n\nThe registered rationale references a pronounced move in ICE BofA US High Yield OAS during the period. But this editorial need not extend that observation into a claim that Undertow captured the move. It did not. The stored HY coverage prevents the replay from testing whether the cell would have reached its registered threshold. A historical market fact and an instrument verdict are different propositions.\n\nThere is a second countercase: perhaps a different public source, a different proxy, or a different measurement frequency could supply enough coverage to test high-yield conditions in the window. That may be true. Yet it would support only a separately documented replay using that source and method. It would not revise the published verdict for the current pack.\n\nThe same restraint applies to the current board. Seiche’s EROSION regime can be read as funding context, while Undertow’s public board reports a limited scoring surface and substantial history accrual elsewhere. Those readings may be compared as separate descriptions, but the dossier expressly says the funding overlay was not used in tiering. Treating them as confirmation of a historical HY replay would confuse timing, product scope, and method.\n\n## What would falsify this\n\nThis editorial makes a limited claim: NOT_RUNNABLE accurately describes the registered HY replay under the stated stored coverage and methodology. That claim would be falsified if the product’s replay pack did contain an eligible observation inside the registered window and the documented process could run the HY test from it.\n\nIt would also be falsified if the published ledger misidentified the segment’s available source coverage, or if the stated registered expectation and resulting verdict could not be reconciled with the replay record. In either case, the problem would be a record or implementation error, not a reason to infer a market conclusion.\n\nA future historical result could improve the evidence base if a newly recorded pack provides the needed coverage and produces a separately identified replay. But that would be a new result under a new recorded data boundary. It would not erase the limitation disclosed by this result.\n\nThe broader proposition—that transparent missingness improves the credibility of a liquidity instrument—would weaken if gaps were routinely disclosed only after adverse results, or if available evidence were withheld without a stated boundary. The relevant standard is symmetrical: the method must be allowed to pass, fail, or decline to run.\n\n## Method and limits\n\nThis is a historical-replay editorial based only on the supplied public record. The replay verdict concerns the instrument, not whether the historical event happened. The article does not reconstruct withheld values, infer unobserved HY readings, or turn a daily board observation into a forward call.\n\nUndertow’s current board defines tiers as deterministic derivations rather than observed prices or trade signals. Its limitations state that a scored tier needs sufficient qualifying measures with accrued stress-percentile history; an accruing or absent measure is disclosed rather than rendered as calm. The current dispatch reports IG as NORMAL, while other listed segments continue to accrue history. That present-day information is context only for this replay.\n\nSeiche’s declared EROSION regime is likewise context only. The source record says its composite scale is a calibrated weighted blend of engine sub-scores, not a percentile of its own history, and that the stress percentile is withheld. It should not be converted into an Undertow score or used as a substitute for the missing historical HY record.\n\nNo investment instruction follows from this analysis. Research and market data only; not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow public API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow API record](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-18-the-march-high-yield-replay-is-not-a-failed-crisis-test-it-is-an-exposed-coverage-gap/", "date": "2026-09-18", "dek": "Undertow’s registered March high-yield replay returns NOT_RUNNABLE because its stored HY inputs contain no observation inside the event window. That verdict describes the instrument’s historical reach—not the severity of the event, and not a forecast.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:fe1949dd680c4f52ce1e448d87e1db6007efc090bd3ce7376a709eafe5be268d", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:50125301303e5cd3555ec5a747d76252bc48d8c7f3de4c69bf0cdf2bf56cc374", "status": "applied"}, "evidence_asof": "2026-09-18", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "0d518f1a0461009eed348d699789b9f72438706b734a024619bc4161f0c88c0a", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:fe1949dd680c4f52ce1e448d87e1db6007efc090bd3ce7376a709eafe5be268d", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:50125301303e5cd3555ec5a747d76252bc48d8c7f3de4c69bf0cdf2bf56cc374", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": [], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The March high-yield replay is not a failed crisis test. It is an exposed coverage gap.", "published_at": "2026-09-18T21:49:53Z", "quality": {"failures": [], "numeric_tokens_checked": 0, "required_headings": 6, "source_links": 8, "status": "PASS", "word_count": 1289}, "slug": "2026-09-18-the-march-high-yield-replay-is-not-a-failed-crisis-test-it-is-an-exposed-coverage-gap", "topic": "mar2020_dash_for_cash_hy", "type": "historical_replay", "word_count": 1289}, "2026-09-19-june-2013-china-replay-is-not-runnable-marking-a-coverage-boundary": {"body_md": "## The read\n\nUndertow’s registered replay of the **June 2013 Shibor shock — China basin** is **NOT_RUNNABLE**. Its stated test was narrow: whether the CN cell reached DEFICIENT on at least one as-of between 2013-06-01 and 2013-07-15. The published result records no scored as-ofs, no evidence object, no sources used and no trace.\n\nThe stated reason is equally narrow. No wired replay source feeds CN, and the packs that carry CN contain no history reaching the test window. This is a coverage boundary, not a finding that the registered tier condition failed and not a finding that CN was calm during the episode.\n\nThat distinction matters because Undertow’s current board also separates missing or accruing evidence from benign conditions. Its 2026-09-19 dispatch says 1 of 9 segments scores: IG reads NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK continue to accrue history. The dispatch says no segment changed tier and labels the Seiche funding overlay EROSION. That overlay is context only and is not used in an Undertow tier or in this replay verdict.\n\n## The mechanism\n\nA replay needs a source that declares the intended segment, historical observations reaching the registered window and a process that can produce readings at each as-of. The CN replay fails before any tier adjudication. The result lists equity_liquidity, fredmkt and nyfed_pd as available sources, but says none declares CN. It further says that the packs carrying CN do not reach the window.\n\nThe episode was registered so that the CN coverage record would show an explicit NOT_RUNNABLE result rather than an empty cell that could be mistaken for a testable but untested case. The record’s rationale specifically notes that the locally accrued SHIBOR curve starts on 2026-06-02, far after the 2013 window.\n\nUndertow’s stated board rule also explains why missing evidence cannot be converted into a calm reading. A scored tier requires at least 2 measures with a stress percentile based on 60 or more observations each. An accruing or absent measure is disclosed rather than rendered as calm. Here, however, the issue is more basic than a partially accumulated measure: the replay has no wired CN source and no applicable history for the window.\n\nThe replay record identifies a separate constraint for cases where historical data are available. It recomputes readings from today’s stored packs after truncating observations to those dated on or before each as-of. The record calls that a construction replay on final data, not a live simulation, because FRED and New York Fed series can be revised and the product has no ALFRED-style vintage archive. The CN episode does not reach that stage, but its result should be read with the same restraint.\n\n## The countercase\n\nA reasonable countercase is that NOT_RUNNABLE is too limited to say much about a historically important episode. That is correct. The result does not establish whether an alternative data set, a different segment construction or another replay design could assess the period.\n\nBut that possibility does not support altering this result. A source added later could change coverage, timing, measure definition or historical availability. It would therefore be a different evidentiary construction, not evidence that the existing replay passed, failed or silently produced a result.\n\nThe current funding context does not close that gap. The dossier describes Seiche’s EROSION regime as a declared, context-only overlay. Its disclosed Seiche index is 41.3/100 and is described as a calibrated weighted blend of engine sub-scores, with high values indicating stress; it is not a percentile of its own history, and its stress percentile is withheld. The funding record’s composite is null. None of that is a substitute for CN replay evidence or a basis for a blended cross-product score.\n\n## What would falsify this\n\nThis article’s claim is limited: the published record says Undertow’s registered CN replay cannot test its stated DEFICIENT condition in the 2013-06-01 to 2013-07-15 window because eligible CN history is unavailable in the wired replay inputs.\n\nThat claim would be falsified if a wired replay source did declare CN and supplied history reaching the window, if the published replay could reproduce scored as-ofs omitted from the result, or if the registered episode details and result could not be reconciled.\n\nA future, transparently documented replay with qualifying CN inputs could support or contradict the registered expectation. It would not turn the present NOT_RUNNABLE result into evidence of either calm or stress. The relevant test is reproducible coverage and a disclosed rerun.\n\n## Method and limits\n\nThis historical replay article relies only on the supplied public pack. The episode’s segment is CN; its expectation is that CN reaches DEFICIENT on at least 1 as-of; and its recorded verdict is NOT_RUNNABLE. The result provides no evidence object, no sources used, no scored as-ofs and no trace.\n\nThe current daily dispatch supplies context only. It reports bounded persistence against 2026-09-18, with a materiality floor of 0.03, and says its adjudication is not attached to the daily reading. Its sealed call is a separate verifiable record rather than a forward call attached to the story. No investment instruction or forecast is made here.\n\nResearch and market data, not investment advice.\n\n## Sources\n\n- [Undertow / LiquiLens public site](https://liquilens-undertow.com)\n- [Undertow public API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens Telegram bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public API record](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-19-june-2013-china-replay-is-not-runnable-marking-a-coverage-boundary/", "date": "2026-09-19", "dek": "Undertow’s registered June 2013 Shibor-shock replay cannot test whether CN reached DEFICIENT in its stated window. The result concerns the instrument’s available historical coverage, not the underlying episode.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:7ad3b8912c9848da724c064e9a77da9a4d8d40884fa41a4d13c8099945147dc0", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:da3afec8687482fd4e0965b15a3e0ee50b9bd36567349d5fd30258a972eae8f6", "status": "applied"}, "evidence_asof": "2026-09-19", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "0d133a9cc5120a789b4f19c79c532995aa0c5d5251bdffe74dda47907ccfc657", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:7ad3b8912c9848da724c064e9a77da9a4d8d40884fa41a4d13c8099945147dc0", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:da3afec8687482fd4e0965b15a3e0ee50b9bd36567349d5fd30258a972eae8f6", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft calls the replay a “preregistered test,” but the dossier says the episode was registered; it does not establish that registration preceded any specified event or decision point.", "“Seiche is a comparison desk” and “LiquiLens is likewise a separate product surface” are not dossier-grounded characterizations. The dossier establishes a context-only Seiche relationship, not those labels.", "The article can be tightened: several passages repeat the same boundary while adding no dossier-supported reporting value."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "June 2013 China replay is NOT_RUNNABLE, marking a coverage boundary", "published_at": "2026-09-19T21:50:31Z", "quality": {"failures": [], "numeric_tokens_checked": 17, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 894}, "slug": "2026-09-19-june-2013-china-replay-is-not-runnable-marking-a-coverage-boundary", "topic": "shibor_shock_jun2013_cn", "type": "historical_replay", "word_count": 894}, "2026-09-20-autumn-2022-treasury-replay-passed-its-bar-but-only-for-the-instrument-tested": {"body_md": "## The read\n\nThe decisive result is not a fresh cross-market alarm. The current Undertow dispatch says only IG scores today, reading NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO, and BSTOCK are still accruing history. The Seiche funding overlay is declared EROSION, but it is context only: it is not blended into any Undertow tier. That combination is a reason to resist manufacturing a live-market conclusion from thin coverage.\n\nThe narrower and more defensible finding sits in the historical record. Undertow’s replay of **Autumn 2022 Treasury illiquidity** passed its preregistered test. The registered question was whether the UST cell would reach STRAINED or DEFICIENT on at least one weekly as-of inside the episode window. It did. The peak replay reading was STRAINED on 2022-10-05, with a within-board score of 0.8193. The ledger reports that every as-of in the window scored.\n\nThat is the thesis: this UST instrument identified stress in the selected Autumn 2022 window under its registered rule, but the pass validates neither a comprehensive account of the episode nor a forecast for the current market. A replay verdict describes the instrument, not whether history happened. The distinction matters because the present board does not furnish broad scored coverage, and because historical validation becomes unhelpful when a single successful episode is promoted into universal evidence.\n\nThe episode runs from 2022-09-15 through 2022-11-30. Its expectation and threshold were registered before the reported replay result was inspected. The outcome therefore has a limited but real evidentiary value: the construction reached the specified tier in the named window rather than having its bar adjusted after the fact. It does not establish that the construction will catch every disruption, distinguish all transmission channels, or offer a trading signal.\n\n## The mechanism\n\nThe UST replay was built to ask whether a defined set of Treasury-related measures registered unusual conditions against their own accrued histories. At the peak reading, the qualifying measures were dealer UST bills net positions, dealer UST coupon net positions, and dealer UST settlement fails. The replay evidence gives their stress percentiles as 0.7857, 0.6932, and 0.9789, respectively. The cell’s tier followed from the board construction, rather than from a discretionary editorial judgment.\n\nThis is a measurement mechanism, not a complete causal model of Treasury-market stress. It maps selected inputs into a tier after applying the product’s coverage and history rules. The stated rule is that a scored tier needs at least two measures with stress-percentile histories of at least sixty observations each. Measures that are accruing or absent are disclosed rather than rendered as calm. That is especially important in a historical replay, where a missing series cannot honestly be converted into confirmation or reassurance.\n\nThe pack records five UST measures in the peak evidence, but only three qualified. The two-year/ten-year slope and dealer UST net positions were unavailable in that entry. The result also records absent HY and IG OAS observations before the relevant date because the supplied series history begins later. Those absences do not undo the UST pass; they define its scope. The pass belongs to a UST cell assembled from the qualifying evidence available under the stated method, not to a synthetic all-market diagnosis.\n\nThe current funding overlay should be read with the same discipline. Seiche declares EROSION, and its disclosed composite is 41.3/100 with full coverage, but that scale is a calibrated weighted blend of engine sub-scores, not a percentile of its own history. Its stress percentile is withheld. It is therefore a separate funding-context reading, not corroboration mechanically folded into Undertow’s UST replay or present IG tier.\n\n## The countercase\n\nThe strongest countercase is that a PASS can be too easy to celebrate. The UST measures are weekly levels ranked against expanding own histories. The replay documentation warns that dealer balance sheets grew structurally over the relevant history, so a level percentile can contain a trend as well as a stress signal. A high percentile may therefore reflect the properties of the proxy and its history, not only the condition an editor wants to name.\n\nThe product itself supplies a reason for caution: it notes a negative control in which the same type of construction fails for exactly this trend-related reason. That warning does not negate the Autumn 2022 result. It means the result cannot be treated as a blanket certification of the measures. A system that reads elevated stress in a known dislocation has cleared one test. It has not proven that it stays quiet when the underlying market is not under comparable strain.\n\nThere is also an evidence-vintage objection. The replay is computed from today’s stored pack, truncated to observations dated on or before each historical as-of. That reconstructs what was knowable by date in the product’s construction, but not necessarily the data vintage published at the time. The supplied record explicitly says FRED and New York Fed series can be revised and that the product holds no ALFRED-style vintage archive. The replay is thus a construction replay on final data, never a live simulation.\n\nFinally, the quiet current board limits any temptation to connect the historical result to a present call. Only one segment scores today. The dispatch calls the comparison with the preceding date bounded persistence, not a new cross-market escalation. An EROSION funding overlay alongside limited current scoring may be worth monitoring as a comparison of separate desks, but the dossier does not permit a claim that it causes, confirms, or predicts a UST tier.\n\n## What would falsify this\n\nThe core claim is falsifiable: the replay pass would fail if the disclosed historical pack could not reproduce a STRAINED-or-worse UST reading in the registered window under the recorded rules. It would also fail if the expectation had not in fact been registered before the result, or if the reported qualifying inputs and their availability did not match the ledger.\n\nThe broader inference should remain conditional. Confidence in the instrument would weaken if a defensible correction to the historical construction, timing, proxy treatment, or data vintage removed the threshold breach. It would also weaken if relevant calm-period controls repeatedly produced equivalent stress classifications. Conversely, further independently recorded tests could add evidence, but they would be new tests, not retroactive proof that this replay measured every dimension of the episode.\n\n## Method and limits\n\nThis editorial uses the supplied public dispatch and replay record only. The historical expectation was UST reaching STRAINED or DEFICIENT at least once in the stated window; the reported result was PASS, with the peak STRAINED reading on 2022-10-05. The article does not infer price performance, tradeability, causation, or an investment action from that result.\n\nUndertow describes its tiers as deterministic board derivations rather than observed prices or trade signals. The current IG score is within-board, while the Seiche funding composite is a distinct calibrated scale; they are not commensurable and are not blended. Qualifying-measure disagreement is disclosed by the dispatch, but the relevant span is withheld on the free board and is not reconstructed here. Research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com)\n- [Undertow data interface](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow on Telegram](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public site](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-20-autumn-2022-treasury-replay-passed-its-bar-but-only-for-the-instrument-tested/", "date": "2026-09-20", "dek": "The current board is quiet, with only IG scoring NORMAL while the funding overlay reads EROSION. The more useful result is historical: Undertow’s preregistered UST replay reached STRAINED during Autumn 2022, a bounded pass rather than a claim of universal detection.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:7cf8a7ed259ceaf4e2c2873b36e0e6e2f15979908fe12c571bf24f0cfa8728d7", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:e5aa6ee2b242f0f1546a311d9f77880cb0d6234ab629acee5876cc0bff50d4a7", "status": "applied"}, "evidence_asof": "2026-09-20", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "d84b0d0e238e6e29c2de42ed0c58e408aaeb9e4d8510ad918b785b1bd5e40a7b", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:7cf8a7ed259ceaf4e2c2873b36e0e6e2f15979908fe12c571bf24f0cfa8728d7", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:e5aa6ee2b242f0f1546a311d9f77880cb0d6234ab629acee5876cc0bff50d4a7", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["All required headings are present, including a substantive countercase and a concrete falsifier section.", "The article stays within the dossier and preserves the research boundary: it makes no investment instruction, causal claim, blended cross-product score, or live-market call.", "Every cited numeral is dossier-grounded: the 0.8193 peak score, 2022-10-05 peak date, three qualifying UST measures, 0.7857/0.6932/0.9789 stress percentiles, 60-observation scoring rule, and Seiche 41.3/100 reading are supported.", "The current-board account correctly distinguishes IG’s NORMAL within-board score from the separate, declared EROSION funding overlay and discloses limited current coverage.", "The countercase accurately reflects the documented trend risk in level percentiles and the final-data, non-vintage replay limitation.", "The body meets the stated length boundary and includes six Markdown links using approved source URLs."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "Autumn 2022 Treasury replay passed its bar—but only for the instrument tested", "published_at": "2026-09-20T21:49:29Z", "quality": {"failures": [], "numeric_tokens_checked": 13, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1204}, "slug": "2026-09-20-autumn-2022-treasury-replay-passed-its-bar-but-only-for-the-instrument-tested", "topic": "oct2022_ust_illiquidity", "type": "historical_replay", "word_count": 1204}, "2026-09-21-a-historical-ig-replay-passed-its-stress-test-it-did-not-validate-the-whole-liquidity-board": {"body_md": "## The read\n\nThe narrow conclusion is clear: Undertow’s registered investment-grade credit replay detected severe stress during the March 2020 dash for cash. The replay verdict is PASS because the IG cell reached DEFICIENT within its registered window, clearing an expectation that it reach STRAINED or DEFICIENT on at least one weekly as-of.\n\nThat is a meaningful result, but only at the scale of the test. It says the specified IG construction, rebuilt from the stored data pack and limited to observations dated on or before each historical as-of, recognized stress in a named interval. It does not prove that the construction catches every liquidity breakdown, measures tradable prices, or supplies a forward-looking signal.\n\nThe distinction matters on a quiet current board. The current dispatch says that only IG scores today, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. IG reads NORMAL on its available measures. Separately, Seiche declares its funding overlay EROSION. These are cross-desk readings, not ingredients in a combined score: the funding overlay is context only and is explicitly not used in the liquidity tier.\n\nThe historical record gives the sharper story. Within the registered March-to-April window, the IG replay scored every weekly as-of. Its peak was DEFICIENT on 2020-04-15, with a score of 0.9824. The registered bar was STRAINED or worse. The test passed. That outcome should be read as a bounded statement about the instrument, not as a retrospective claim that the episode itself needed validation.\n\nThe episode is historical fact; the replay verdict describes what the instrument did with its specified inputs. That separation is the editorial point. Famous crises can make every backtest look wiser than it was. A useful replay instead states its test before presenting the answer, exposes missing evidence, and permits the result to be narrower than the event it is meant to interrogate.\n\n## The mechanism\n\nThe IG cell’s mechanism is deliberately closer to market functioning than to a generic risk narrative. Its replay rests on New York Fed primary-dealer corporate-securities net positions, described as congestion, and corporate settlement fails. The registered episode rationale links those measures to a period when corporate-bond intermediation was under pressure alongside Treasuries.\n\nThat choice does not make the measures comprehensive. It defines the question. A liquidity instrument need not explain every reason investors wanted to sell; it must identify a measurable expression of impaired redistribution. Dealer positions and settlement fails can be relevant because constrained balance-sheet capacity and disrupted settlement are consistent with a market struggling to transfer risk. The replay asks whether those inputs, transformed by the board’s own historical framework, reached the preregistered tier threshold during the episode.\n\nThey did. The trace shows settlement-fails readings persistently elevated relative to the measure’s available history across the window, while the dealer-position input moved more materially later. The combined board result was NORMAL at the earliest listed as-ofs, then STRAINED, then DEFICIENT. That progression is not a price chart and should not be treated as one. It is the output of a deterministic tier construction operating on these particular measures.\n\nThe design also records what was unavailable. IG OAS and HY OAS series in the replay pack had no observation before the relevant historical dates because their available history began much later. Those absent inputs were disclosed rather than backfilled with a story-friendly substitute. The historical result therefore came from the eligible dealer measures, not from a reconstructed spread narrative.\n\nThis is also why the current funding overlay cannot be folded into the historical IG result. Seiche’s EROSION label is a declared funding regime from a separate product. Its underlying composite is not presented as a percentile of its own history, and its stress percentile is withheld. It may be useful adjacent context, but it is not a common unit with Undertow’s within-board liquidity score. Combining them would manufacture precision the evidence does not provide.\n\n## The countercase\n\nThe strongest countercase is not that the replay failed. It is that passing a crisis-window test can be easier than proving a tool is discriminating. A measure can look stressed in a major dislocation because it tracks a structural trend, broad volatility, or a feature of the reporting system rather than executable liquidity itself.\n\nThe replay documentation identifies exactly such a concern. The dealer position and fails measures are percentiles of weekly levels against an expanding own history beginning in 2013. Dealer balance sheets grew structurally over that period, so a level percentile can carry trend as well as stress. The replay materials point to a negative control in which the construction fails for that reason. That warning does not erase the PASS; it constrains the inference readers should draw from it.\n\nA second countercase is coverage. The March 2020 event involved channels beyond the two qualifying IG measures. A market could be difficult to transact through dealer withdrawal, fund flows, ETF dislocations, collateral conditions, volatility, or price gaps that this particular cell does not fully capture. Conversely, the instrument could miss a different episode whose stress appeared in a channel outside its available data.\n\nNor should the current NORMAL IG tier be used to dismiss the separate EROSION funding reading. The products measure different declared constructs, with different methods and evidence availability. Their disagreement is informative as a comparison: it says no single label should stand in for the whole cross-market liquidity picture. It does not establish which desk is “right,” and it cannot be converted into a blended stress score.\n\n## What would falsify this\n\nThe central claim is falsifiable: the published replay says the registered IG instrument reached at least STRAINED during its defined March-to-April window, and its peak was DEFICIENT. That claim would fail if the public replay record could not reproduce the stated tier from the disclosed eligible measures and dated truncation rule, if the registered expectation were shown to have been altered after the result, or if the attributed qualifying inputs were absent from the record.\n\nThe broader interpretation would weaken if a defensible revision to the proxy, timing, orientation, or historical treatment removed the result—or if calm-period controls repeatedly produced equivalent stress. It would strengthen if independent point-in-time measures of IG market functioning confirmed the same pattern while suitable controls remained quiet. Neither outcome may be assumed in advance.\n\nA future improvement in source coverage would not silently upgrade this historical result. It would create a different replay vintage, with a separately recorded result and limitations. The present verdict is attached to this construction and data boundary, not to all possible reconstructions of the episode.\n\n## Method and limits\n\nThis is a historical replay, not a current-market call. The current board was published on 2026-09-21, while the replay evaluates a registered IG episode from 2020-03-01 through 2020-04-30. The replay is constructed by truncating today’s stored pack to observations dated on or before each historical as-of. That is a dated-data reconstruction, not the exact contemporaneous vintage that market participants received.\n\nThe limitation is material. The underlying FRED and New York Fed series may have been revised, and Undertow does not hold an ALFRED-style vintage archive. The record therefore calls this a construction replay on final data, never a live simulation. Publication lags are also part of the evidence boundary: a historical as-of and the date at which it was knowable are not interchangeable.\n\nA scored tier requires qualifying measures with sufficient history; incomplete coverage is disclosed rather than rendered as calm. The free board withholds some row-level percentile detail, so measure disagreement can be identified without reconstructing withheld values. No subscriber data are inferred here. The current Seiche funding regime is presented only as a separate overlay, never blended with Undertow’s tier.\n\nResearch and market data only; not investment advice.\n\n## Sources\n\n- [Undertow / LiquiLens public site](https://liquilens-undertow.com) — current board and published Undertow materials.\n- [Undertow API](https://api.seiche.info/undertow) — public analytical records, including replay and sealed-call materials.\n- [Seiche](https://seiche.info) — separate funding-context desk; its overlay is not used in the liquidity tier.\n- [LiquiLens](https://liquilens.in) — adjacent institution-liquidity product context.\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot) — public access point for board views.\n- [Undertow API record surface](https://api.seiche.info/undertow) — cited again for the historical replay boundary and published evidence record.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-21-a-historical-ig-replay-passed-its-stress-test-it-did-not-validate-the-whole-liquidity-board/", "date": "2026-09-21", "dek": "The registered investment-grade replay reached DEFICIENT during the March 2020 dash for cash. That is evidence that a defined instrument detected one episode—not a warrant to turn a quiet current board or a funding overlay into a forecast.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:aab06231b0d4b1115748f904a6018a7a9be58d7db7440a95a40a18c2291a26ac", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:09f24a4aa0dee90e175f7dc976833318dc41ad0eb4614540c51b96aa80a83e59", "status": "applied"}, "evidence_asof": "2026-09-21", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "e1dd8437857462e082af83069aae9dff98129957c83b444f42d08db233fbcb8d", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:aab06231b0d4b1115748f904a6018a7a9be58d7db7440a95a40a18c2291a26ac", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:09f24a4aa0dee90e175f7dc976833318dc41ad0eb4614540c51b96aa80a83e59", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The article stays within the dossier’s historical-replay boundary and does not turn the current board or Seiche overlay into a forecast, trade instruction, or blended score.", "Its central numerical claims match the dossier: the registered IG expectation was STRAINED or worse; the replay peak was DEFICIENT at 0.9824 on 2020-04-15; the episode window runs from 2020-03-01 through 2020-04-30; and the current board has 1 of 9 scored segments, with IG NORMAL and Seiche declared EROSION.", "The countercase is substantive and dossier-grounded: it addresses structural trend in weekly-level percentiles, the documented negative-control concern, limited qualifying-measure coverage, and the distinction between separate products.", "The falsifier is explicit and appropriate to the claim: reproducibility of the disclosed tier, preregistration integrity, eligible inputs, and sensitivity to defensible methodological revisions or calm-period controls.", "All required headings are present: The read, The mechanism, The countercase, What would falsify this, Method and limits, and Sources.", "The research boundary and vintage limitation are plainly stated: this is a construction replay using today’s stored pack truncated by historical observation date, not a contemporaneous-vintage simulation; possible FRED and New York Fed revisions are disclosed.", "The Sources section contains six Markdown-link instances, all using approved URLs."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "A historical IG replay passed its stress test. It did not validate the whole liquidity board.", "published_at": "2026-09-21T21:49:55Z", "quality": {"failures": [], "numeric_tokens_checked": 11, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1369}, "slug": "2026-09-21-a-historical-ig-replay-passed-its-stress-test-it-did-not-validate-the-whole-liquidity-board", "topic": "mar2020_dash_for_cash_ig", "type": "historical_replay", "word_count": 1369}, "2026-09-22-treasury-replay-passed-its-hard-bar-without-becoming-a-forecast": {"body_md": "## The read\n\nUndertow’s registered UST historical replay cleared its stated bar for the March 2020 dash-for-cash episode. The test window ran from 2020-03-01 through 2020-04-30. Its requirement was limited: the UST cell had to reach STRAINED or DEFICIENT on at least one weekly as-of in that window.\n\nIt did. The replay’s peak was DEFICIENT, with a score of 0.9151, on 2020-04-08. The result record says 43 of 43 as-of dates scored, and its verdict is PASS.\n\nThat supports a narrow proposition: under this replay’s disclosed construction, the UST instrument registered severe stress during the named historical window. It does not establish that the instrument identifies every liquidity event, explains the event’s causes, provides the earliest usable warning, or forecasts prices.\n\nThe current Undertow dispatch is separate from that historical result. On 2026-09-22, it says 1 of 9 segments scores; IG is NORMAL with a within-board score of 0.3568 based on 3 available and qualifying measures. UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK are still accruing history. The Seiche funding overlay is EROSION, but it is context only and is not used in an Undertow tier.\n\nThe replay therefore should not be converted into a present-tense market conclusion. It is evidence about a defined historical test of a defined construction.\n\n## The mechanism\n\nThe registered episode concerns Treasury-market liquidity. Its rationale identifies dealer Treasury positions and settlement fails as the relevant UST quantities. The replay’s UST evidence at the 2020-04-08 peak comprised dealer UST bills net positions, dealer UST coupon net positions and dealer UST settlement fails. All 3 were qualifying measures; the UST score was 0.9151 and the resulting tier was DEFICIENT.\n\nThe construction compares measures with their own accrued histories rather than presenting raw levels as self-interpreting. The trace shows that the UST cell was NORMAL at earlier points in the window, reached STRAINED during the episode, and reached DEFICIENT at its recorded peak. That tier path is the output under the product’s rules; it is not an observed price, a causal finding or a trading instruction.\n\nThe record also distinguishes an as-of date from the date a reading was knowable. For the 2020-04-08 peak, the stated knowable date was 2020-04-16. That distinction matters: the replay evaluates stored inputs dated on or before each as-of, but it does not demonstrate that market participants had complete contemporaneous visibility.\n\nSeiche’s EROSION designation remains a separate funding-context overlay. The dossier describes its index as a calibrated weighted blend, with higher values indicating stress, rather than a percentile of its own history. Undertow does not use that overlay in the UST tier, and this article does not create a blended cross-product score.\n\n## The countercase\n\nA PASS in a prominent crisis is not sufficient evidence of a broadly discriminating signal. The strongest countercase is that the UST measures can reflect structural change as well as acute liquidity stress. The replay’s source caveat says dealer position and fails measures are percentiles of weekly levels against expanding own histories beginning on 2013-04-03, while dealer balance sheets grew structurally over that period.\n\nThat means a high percentile can combine stress information with trend. The dossier specifically points to a Q2-2017 negative control that fails for this reason. The March 2020 outcome therefore does not resolve whether the construction cleanly distinguishes acute dislocation from a persistent shift in the underlying level of a measure.\n\nCoverage is another constraint. The replay record reports that HY and IG OAS series have no observation before 2023-09-22 in this pack. Those absences do not negate the UST result, which uses 3 qualifying UST measures at its peak. They do, however, prevent the result from serving as a comprehensive cross-market reconstruction of the 2020 episode.\n\nTiming also limits the claim. A DEFICIENT reading at an as-of of 2020-04-08, knowable on 2020-04-16, supports classification within this replay framework. It does not support describing the instrument as an early-warning system.\n\n## What would falsify this\n\nThe narrow claim would fail if the registered window, stated threshold and stored date-truncated inputs could not reproduce the PASS result. It would also fail if the UST cell did not reach at least STRAINED in the 2020-03-01 to 2020-04-30 window, or if the reported peak of DEFICIENT at 0.9151 on 2020-04-08 could not be reproduced from the record.\n\nThe broader interpretation would weaken if a defensible treatment of revisions, timing or measure construction removed the stressed classification. It would weaken further if calm-period controls repeatedly generated comparable tiers without an identifiable liquidity event. Those are relevant tests because the disclosed caveat identifies structural trends in weekly level percentiles as a possible confound.\n\nA future market move would not falsify this article’s claim. The replay makes no forward call. Its testable proposition is historical and procedural: whether the disclosed construction classified the registered UST episode as required.\n\n## Method and limits\n\nThis article is limited to the published Undertow historical replay record, its registered expectation and the current dispatch. The replay is computed from today’s stored pack, truncated to observations dated on or before each historical as-of. The dossier explicitly says this is a construction replay on final data, not the vintage published at the time. FRED and New York Fed series can be revised, and the product has no ALFRED-style vintage archive.\n\nThe board tier is a deterministic derivation, not an observed price or a trade signal. Measures that are accruing or absent are disclosed rather than rendered as calm. The current public board also withholds percentile information needed to reconstruct spans in disagreed cells; this article does not infer withheld values.\n\nThe historical PASS is not attached to a sealed forward call. The product’s sealed-call record is separate and verifiable. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com) — product and historical-replay context.\n- [Undertow API](https://api.seiche.info/undertow) — public Undertow record surface.\n- [Seiche](https://seiche.info) — funding-overlay context only.\n- [LiquiLens](https://liquilens.in) — separate product context.\n- [Undertow Telegram board](https://t.me/Undertow_LiquiLens_bot) — public board access.\n- [Undertow API record](https://api.seiche.info/undertow) — replay and separate-record access point.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-22-treasury-replay-passed-its-hard-bar-without-becoming-a-forecast/", "date": "2026-09-22", "dek": "Undertow’s UST replay reached DEFICIENT in the March 2020 dash-for-cash window, clearing its registered threshold. The result is a narrow construction replay on final stored data, not a live simulation, market forecast or trade signal.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:ab52af94753066cea689778289d752916c14d69fb5c5a4ec58031fa9973a7174", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:97cf9b6d70fd8d31888393593fd2a0d59b96f65b54bf678838ed776fb00fd5d1", "status": "applied"}, "evidence_asof": "2026-09-22", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "a2e97d83e8ab39ad09a6ba29e6e0be5a0ebd4b0300adf6740454f92773301b82", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:ab52af94753066cea689778289d752916c14d69fb5c5a4ec58031fa9973a7174", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:97cf9b6d70fd8d31888393593fd2a0d59b96f65b54bf678838ed776fb00fd5d1", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft says the hard bar was written down before the replay result was inspected. The dossier provides a registration date but does not establish that temporal comparison; the claim must be removed.", "The draft overstates the current-board characterization by calling it a “quiet current board.” The dossier supports bounded persistence and limited current scoring, but the repair avoids adding an editorial characterization beyond those facts.", "The repaired version retains the required six headings, six approved Markdown links, the historical-replay boundary, the countercase and falsifier, and only dossier-grounded numerals."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "Treasury replay passed its hard bar, without becoming a forecast", "published_at": "2026-09-22T21:49:40Z", "quality": {"failures": [], "numeric_tokens_checked": 19, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1011}, "slug": "2026-09-22-treasury-replay-passed-its-hard-bar-without-becoming-a-forecast", "topic": "mar2020_dash_for_cash_ust", "type": "historical_replay", "word_count": 1011}, "2026-09-23-the-august-2024-replay-passed-but-only-as-a-test-of-one-equity-volatility-measure": {"body_md": "## The read\n\nThe August 2024 replay is a pass for Undertow’s VIX-based stress detection, not a pass for a complete EQUITY liquidity verdict.\n\nThat distinction is the thesis. In the registered window for the August 2024 yen-carry unwind—equity volatility, the replay found that VIX close reached a point-in-time own-history stress percentile of 0.973 on 2024-08-05. The preregistered requirement was a reading of at least 0.95 on at least one day in the window. The verdict was therefore PASS.\n\nBut the instrument did not produce a scored EQUITY tier on any date in that replay. Every recorded observation was PARTIAL. VIX was the sole qualifying measure; the other EQUITY measures were unavailable for qualification under the board’s history and coverage rules. The replay is consequently useful because it draws a hard boundary: the construction detected exceptional equity volatility through one established input, yet it did not establish a full multi-measure equity-liquidity state.\n\nThat is more informative than treating a historical pass as a general certificate. A replay verdict describes the instrument, not whether history happened. The episode was registered at measure level precisely because the EQUITY cell could not score at the historical as-of. The test asked whether a named VIX measure would enter its own extreme historical tail under point-in-time truncation. It did. It did not ask whether the full EQUITY segment would diagnose, explain, or forecast the episode.\n\nThe current board supplies no reason to inflate that conclusion. Its daily dispatch says only one of nine segments scores, with IG reading NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO, and BSTOCK continue to accrue history. Its funding overlay reads EROSION, but that is a Seiche comparison overlay only, not an input blended into an Undertow tier. The board’s quiet state is a reason to inspect a bounded historical test, not a reason to manufacture a current-market call.\n\n## The mechanism\n\nThe replay’s mechanism is percentile detection under an explicit historical-information constraint. For each historical as-of, Undertow recomputed the reading from the stored pack using observations dated on or before that as-of. In the registered window, VIX close carried a percentile on every recorded date. It rose from lower readings earlier in the window to 0.973 on 2024-08-05, then receded afterward.\n\nThe registered rationale set a demanding but limited expectation. VIX closed at 38.57 on 5 August 2024, its highest close since 2020. The rationale further notes that VIXCLS has daily history from 1990, making an extreme own-history percentile a construction test that should clear if percentile orientation and point-in-time truncation work as intended. The replay result cleared the stated floor.\n\nThat matters because a historical chart alone cannot establish that an indicator was evaluated according to rules available before its result was known. Here, the episode, measure, floor, and window were registered before the replay result. The reported peak was then checked against that specification. The process is not a full recreation of contemporaneous market knowledge; it is a construction replay using a current stored dataset truncated by date. But it does prevent the simpler error of selecting the threshold after seeing the outcome.\n\nThe coverage rule is equally important. A scored tier requires at least two qualifying measures, each with a stress percentile and sufficient history. In the replay, the four Amihud and lambda measures were not qualifying. The result had one qualifying measure rather than the minimum required for a scored cell. PARTIAL therefore means still accruing, not calm. It says the board disclosed an incomplete measurement surface instead of converting missing qualification into reassurance.\n\nThis is why the VIX pass and the unscored EQUITY cell are compatible rather than contradictory. One measure can confirm that it recognized an exceptional point in its own historical distribution. A segment tier requires broader, independently qualifying coverage. The former occurred; the latter did not.\n\n## The countercase\n\nThe strongest countercase is that VIX is a volatility proxy, not a complete measure of executable equity liquidity. It can register intense stress while leaving unanswered questions about depth, immediacy, market-making capacity, cross-sectional trading costs, and the ability of participants to redistribute risk. A pass on VIX cannot fill those evidentiary gaps.\n\nThe historical pack itself supplies a further limitation. The equity-liquidity source is described as backfilled for a basket committed later, with survivorship conditioning in its membership. Truncating the series by date does not remove that cross-sectional look-ahead. That caveat does not invalidate the VIX result, which used the FRED market-series source, but it constrains any broader claim about the missing equity-liquidity legs.\n\nThere is also a vintage limitation. The replay uses today’s stored pack, truncated to data dated on or before each historical as-of. The record explicitly says this is what was knowable by date, not the vintage published at the time. FRED and the New York Fed can revise series, and the product has no ALFRED-style vintage archive. The PASS is therefore a replay on final data, never a live simulation.\n\nNor does the funding overlay resolve the gap. Seiche’s declared EROSION regime is a separate funding-context overlay. It is useful as a comparison across desks, but it is not a blended score and was not used in the EQUITY replay tier. Treating separate products as one composite would create a conclusion the disclosed method does not support.\n\n## What would falsify this\n\nThe editorial claim is narrow and testable: the published replay correctly reports that the registered VIX expectation passed during its stated window, while the EQUITY cell remained PARTIAL and unscored.\n\nThat claim would fail if the public replay record could not reproduce the 0.973 peak on 2024-08-05 from the stated truncated construction; if the registered 0.95 threshold or window had been altered after the result without a visible new registration; or if the underlying record showed enough qualifying measures to score the cell but the article continued to call it PARTIAL.\n\nThe broader interpretation would weaken if a defensible correction to timing, percentile orientation, source history, or final-data revisions removed the threshold breach. It would gain support if additional independent, point-in-time measures qualified under the coverage rule and showed consistent stress without relying on later-known basket membership. Neither outcome may be assumed in advance. A method that cannot be contradicted is not being tested.\n\n## Method and limits\n\nThis is a historical replay, not a current-market forecast and not an investment recommendation. The episode covered 2024-07-25 through 2024-08-16. It had seventeen recorded as-of dates, and the VIX percentile was present on all seventeen. The highest reading was 0.973, above the registered 0.95 floor. The EQUITY score remained null because only one measure qualified.\n\nThe article relies on the published Undertow dispatch and replay materials described in the dossier. It does not infer withheld subscriber values, calculate a cross-product composite, or treat the current Seiche funding regime as evidence inside the equity score. The current board’s IG NORMAL reading and EROSION overlay are contextual comparisons only.\n\nResearch and market data, not investment advice.\n\n## Sources\n\n- [Undertow](https://liquilens-undertow.com) — public product context and board methodology.\n- [Undertow API](https://api.seiche.info/undertow) — published Undertow records and replay materials.\n- [Seiche](https://seiche.info) — funding-context comparison desk.\n- [LiquiLens](https://liquilens.in) — separate institution-liquidity desk.\n- [Undertow on Telegram](https://t.me/Undertow_LiquiLens_bot) — access point for the live board.\n- [Undertow public board](https://liquilens-undertow.com) — current-board context; not blended with the replay result.\n- [Seiche funding context](https://seiche.info) — declared EROSION overlay; not used in the EQUITY tier.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-23-the-august-2024-replay-passed-but-only-as-a-test-of-one-equity-volatility-measure/", "date": "2026-09-23", "dek": "Undertow’s preregistered replay of the August 2024 yen-carry unwind cleared its VIX threshold. The result validates a narrow piece of the historical instrument, while the EQUITY cell remained PARTIAL and unscored throughout the window.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:58e8b5dcb61ee1abe45f477dc81190fa1e1d2fbfbb5e130ae0f6f089e841fbf5", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:2b19bd5184fbc311250bdf1e3b7c306a9ca8fabb64fbf691c7fd4e095c7b00df", "status": "applied"}, "evidence_asof": "2026-09-23", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "3a78f038df9fc50d185b1f90c4c7c4255215a8b2e617b7db7f9cd315c2ca343f", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:58e8b5dcb61ee1abe45f477dc81190fa1e1d2fbfbb5e130ae0f6f089e841fbf5", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:2b19bd5184fbc311250bdf1e3b7c306a9ca8fabb64fbf691c7fd4e095c7b00df", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["All required headings are present: The read, The mechanism, The countercase, What would falsify this, Method and limits, and Sources.", "The article is within the required 800–1600-word range and stays within the historical-replay research boundary.", "The central claim is properly narrow: the replay passed a preregistered VIX-measure threshold, while EQUITY remained PARTIAL and unscored throughout the window.", "All dossier-grounded numerals and dates checked: the 0.973 peak on 2024-08-05, the 0.95 registered floor, the 2024-07-25 to 2024-08-16 window, 17 as-of dates, one qualifying measure, and the two-measure scoring minimum.", "The countercase and falsifiers are substantive and dossier-consistent. The article discloses the incomplete EQUITY coverage, the survivorship-conditioned equity-liquidity pack, and the truncated-latest rather than contemporaneous-vintage limitation.", "It does not improperly blend the Seiche EROSION overlay into an Undertow score, infer withheld data, make a current-market call, or give investment instruction.", "The Sources section contains at least six Markdown links, and every linked URL is on the approved-source list."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The August 2024 replay passed—but only as a test of one equity-volatility measure", "published_at": "2026-09-23T21:51:10Z", "quality": {"failures": [], "numeric_tokens_checked": 12, "required_headings": 6, "source_links": 7, "status": "PASS", "word_count": 1241}, "slug": "2026-09-23-the-august-2024-replay-passed-but-only-as-a-test-of-one-equity-volatility-measure", "topic": "aug2024_yen_carry_vix", "type": "historical_replay", "word_count": 1241}, "2026-09-24-ig-moves-tighter-by-0-06-inside-normal-funding-reads-erosion": {"body_md": "## The read\n\nThe decisive fact in Undertow’s current cross-market liquidity reading is not a broad market verdict. It is the narrowness of what can be said. On the board dated 2026-09-24, IG is the only segment that scores. Its within-board score is 0.419, its tier is NORMAL, and all 3 available measures qualify. The score moved tighter by 0.06 from the preceding reading, clearing the stated materiality floor of 0.03 while remaining inside the same tier.\n\nThat is a real update, but it is not an all-clear. The board also says qualifying measures disagree within the IG cell. Under the public-board rules, the span of such a cell is withheld at Agent/Desk tier because publishing it would reveal a percentile that the free board does not disclose. The proper editorial conclusion is therefore constrained: IG has registered a material internal change without changing its published tier, and the public evidence does not support treating NORMAL as unanimity.\n\nThe comparison that matters is adjacent rather than additive. Seiche’s funding overlay is EROSION, with its declared regime surfaced separately from the Undertow tier. The overlay’s listed index is 42.1/100 with full coverage, but the dossier explicitly says that scale is a calibrated weighted blend of engine sub-scores, not a percentile of its own history. It must not be folded into IG’s 0.419 score, translated into an equivalent percentile, or presented as a blended cross-product signal.\n\nThe narrow thesis is this: **the IG NORMAL label is best understood as a qualified board classification, not as evidence that the underlying qualifying measures or the separate funding context are aligned.** The fresh 0.06 tightening makes IG the active reading; the disagreement limits the confidence that can be assigned to its average.\n\nEight other named segments—UST, HY, EQUITY, ETF, FX, CN, CRYPTO, and BSTOCK—are still accruing history. Their absence from the scored set is not evidence of calm, deterioration, or confirmation of the IG result. It is a coverage boundary. The dispatch reports 1 of 9 segments scoring today; it does not authorize a market-wide extrapolation from that 1.\n\n## The mechanism\n\nUndertow’s mechanism is methodological before it is market-narrative. A tier is a deterministic board derivation, not an observed price and not a trade signal. A scored tier requires at least 2 measures with a stress percentile at 60 or more observations each. The IG cell satisfies its stated qualification test with 3 qualifying measures. The board then classifies the cell as NORMAL at 0.419.\n\nThe score’s interpretive burden is deliberately limited by its construction. It is a within-board score, while the underlying measures are assessed against their own accrued point-in-time histories. That design permits comparison of readings without asserting that different raw measures are naturally interchangeable. It also means that the published score should not be confused with an executable market level, a return forecast, or a claim about a single observable transaction.\n\nThe evidence released for the IG desk shows why the label cannot substitute for the internals. IG OAS, identified as BAMLC0A0CM, has a disclosed stress percentile of 0.1234 as of 2026-09-23. Its published limit is material: the source’s keyless download is capped to a trailing roughly 3-year window, omitting earlier crisis regimes and potentially understating rarity in calm tapes. That caveat belongs to the interpretation of the percentile, not merely to its footnote.\n\nA dealer corporate net-positions congestion measure is also listed, as of 2026-09-16, but its stress percentile is not disclosed. The stated limitation is equally consequential: it is a proxy for IG because the underlying corporate-securities series spans the wider corporate-credit complex, including commercial paper, IG, and below-IG bonds. The series is also described as spliced across breaks. Its presence demonstrates available evidence; the withheld percentile prevents public reconstruction of the qualifying-measure spread.\n\nThe published board says the qualifying measures disagree, but does not release enough row-level information to identify the full numerical distribution behind the cell. That is not a gap to fill with inference. It is the central constraint on the story. The 0.419 score says the deterministic aggregation remains NORMAL. The disagreement flag says the aggregation should not be mistaken for convergence.\n\nSeiche’s EROSION designation adds context, not arithmetic. Its measure has no disclosed stress percentile, and its regime basis is declared. The proper mechanism is therefore a comparison of distinct desks: Undertow records IG’s own board status and internal disagreement; Seiche records a funding overlay under its own framework. Neither reading validates, cancels, or numerically converts the other.\n\n## The countercase\n\nThe strongest countercase is straightforward: NORMAL may be the most useful conclusion available from the qualifying IG measures despite the disagreement. The cell has 3 qualifying measures, not a partial or absent status, and the fresh movement remained within its existing tier. The published classification is therefore not provisional in the sense of failing the board’s stated coverage gate.\n\nA second countercase is that the separate EROSION overlay should not be used to cast doubt on IG at all. The dossier explicitly defines the relationship as funding-context-overlay-only and says it was not used in the tier. Because the products measure different declared objects, divergence may simply be divergence. It does not establish a transmission from funding to IG, nor does it prove that the IG classification is lagging or incomplete.\n\nThe public evidence also does not establish direction beyond the stated score change. “Tighter by 0.06” is the dispatch’s framing of the within-tier delta. It does not supply a causal explanation, identify a specific underlying measure as the driver, or establish a forward path. Any assertion that the move reflects a particular buyer, dealer, issuer, policy event, or forthcoming change would exceed the public pack.\n\n## What would falsify this\n\nThis thesis would be falsified if a subsequent point-in-time board showed that the present characterization of IG as a qualified but internally disputed NORMAL cell was wrong on its own disclosed fields. That could occur if the board no longer flags qualifying-measure disagreement, if the published tier changes, if the score is revised, or if the count of qualifying measures changes under the stated rules.\n\nIt would also be weakened if the board released evidence demonstrating that the apparent disagreement resulted from a publication or classification error rather than the declared cell state. Conversely, no later market outcome can by itself validate the thesis: the claim concerns the instrument’s published reading and its evidentiary limits on 2026-09-24, not whether history subsequently rewarded a narrative.\n\nThe Seiche comparison has its own falsifier. The contextual statement here would fail if the declared funding regime were corrected from EROSION or if the source ceased to identify the overlay as separate from Undertow’s tier calculation. A different Seiche regime would not automatically change IG’s score; it would change only the cross-desk comparison.\n\n## Method and limits\n\nThis editorial uses only the published 2026-09-24 Undertow dispatch and the associated evidence envelope. The relevant clocks place the event time at 2026-09-24T00:00:00Z and publication at 2026-09-24T21:50:03+00:00. The source status is OBSERVED. This is a point-in-time reading, not a reconstruction using later information.\n\nThe method boundary is strict. Undertow’s board derives tiers deterministically; it does not observe a price or issue a trading instruction. PARTIAL means history is still accruing, and an accruing or absent measure is disclosed rather than rendered as calm. The free board withholds spans where disclosure would hand back protected row-percentile information. This editorial does not estimate those spans or manufacture a composite from Undertow and Seiche.\n\nNo sealed call is attached to this daily story. The dispatch identifies sealed calls as a separate verifiable record and says a daily board reading is not silently promoted into a forward call. Historical replay is likewise not attached to this daily reading. These are meaningful absences, not omissions to be repaired through retrospective interpretation. Research and market data only; not investment advice.\n\n## Sources\n\n- [Undertow public board](https://liquilens-undertow.com)\n- [Undertow data API](https://api.seiche.info/undertow)\n- [Seiche funding desk](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens Telegram bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public board, source access](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-24-ig-moves-tighter-by-0-06-inside-normal-funding-reads-erosion/", "date": "2026-09-24", "dek": "IG is the only scored segment on the current board, and its NORMAL tier should be read narrowly: the board records a fresh within-tier move amid qualifying-measure disagreement, while the separate Seiche funding overlay declares EROSION.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:bcfd6d868b8fc4c991204d430dd6d71ac85cc6be057b7802922fd6da1f1ea9af", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:5282b5a5cc667ca5b3fc358d63da113258c7dc567b13dc9462c760ddda6eb5ce", "status": "applied"}, "evidence_asof": "2026-09-24", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "4112f50a0d934bacc786e3ca3fa5c59d220eadcedff02d378caf9b1f7dc38b31", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:bcfd6d868b8fc4c991204d430dd6d71ac85cc6be057b7802922fd6da1f1ea9af", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:5282b5a5cc667ca5b3fc358d63da113258c7dc567b13dc9462c760ddda6eb5ce", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": [], "writer_model": "openai/gpt-5.6-terra"}, "headline": "IG moves tighter by 0.06 inside NORMAL; funding reads EROSION.", "published_at": "2026-09-24T21:50:47Z", "quality": {"failures": [], "numeric_tokens_checked": 20, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1359}, "slug": "2026-09-24-ig-moves-tighter-by-0-06-inside-normal-funding-reads-erosion", "topic": "IG", "type": "current_analysis", "word_count": 1359}, "2026-09-25-ig-moves-tighter-by-0-03-inside-normal-funding-reads-erosion": {"body_md": "## The read\n\nUndertow’s published reading for 2026-09-25 classifies investment-grade credit, or IG, as **NORMAL**. The within-board score is **0.4494**, and the dispatch identifies a **0.03** fresh within-tier delta from 2026-09-24 that meets its stated **0.03** materiality floor. Neither the score nor the delta changes the published tier: this is a movement within NORMAL, not a tier break.\n\nThe qualification is internal disagreement. The public record says IG has **3** available measures and **3** qualifying measures, and that the qualifying measures disagree. It does not publish their full span. The dispatch says that, where a scored cell has such disagreement, the span is withheld at Agent/Desk tier because publishing it would disclose a percentile unavailable on the free board. The supportable conclusion is therefore narrow: IG is NORMAL, its qualifying inputs are not unanimous, and the public evidence does not permit reconstruction of their dispersion.\n\nOnly **1 of 9** segments scores in this dispatch. UST, HY, EQUITY, ETF, FX, CN, CRYPTO, and BSTOCK are still accruing history. That fact does not establish cross-market deterioration; it instead limits the reading. IG is the only scored segment, while the other named segments do not yet have the accrued history needed for their scored tiers.\n\nThe separate Seiche funding overlay is declared **EROSION**. Undertow specifies that this relationship is a funding-context overlay only and that Seiche is not used in the IG tier. Accordingly, NORMAL for IG and EROSION for funding cannot be converted into a single market conclusion by averaging, weighting, or otherwise blending the two published outputs.\n\n## The mechanism\n\nThe relevant mechanism is the board’s construction, not an observed price or trade instruction. Undertow states that a tier is a deterministic board derivation. A scored tier requires at least **2** measures with a stress percentile and at least **60** observations for each. Measures that are accruing, absent, or withheld are disclosed rather than rendered as calm.\n\nA NORMAL tier can therefore coexist with disagreement among eligible inputs. The tier is the board’s published output under its rules; it is not a statement that every component gives the same reading. Here, the board explicitly discloses disagreement but withholds the span needed to quantify it.\n\nThe disclosed IG OAS measure, BAMLC0A0CM, has a stress percentile of **0.2144** as of 2026-09-24. Its note says that a wider investment-grade spread means costlier credit liquidity and more stress. The same note limits interpretation: keyless FRED downloads are capped to a trailing window of about **3 years**, omitting earlier crisis regimes including 2008 and 2020 and understating rarity in calm tapes.\n\nA disclosed dealer-corporate-net-positions measure is dated 2026-09-16. It is explicitly a corporate-credit proxy rather than an IG-only measure: the note says the series spans commercial paper, IG, and below-IG bonds, and that levels are spliced across NY Fed series breaks. Its stress percentile is withheld. The public pack therefore does not support an exact comparison between that proxy and the OAS percentile, or a claim that the proxy alone confirms or contradicts NORMAL.\n\nSeiche’s funding measure has a different stated basis. Its index is **41.2/100**, coverage is **100%**, and the declared regime is EROSION. The note says the index is a calibrated weighted blend of engine sub-scores, with higher readings indicating more stress; it is not a percentile of its own history, and its stress percentile is withheld. That construction is another reason not to combine the funding index arithmetically with Undertow’s IG score.\n\n## The countercase\n\nThe principal countercase is that the disclosed disagreement has no implication beyond ordinary variation within a NORMAL tier. The published classification remains NORMAL, all **3** available IG measures qualify, and the dispatch identifies no stressed cross-market tier. Its newsworthiness score is **55**, reaching the stated full-story threshold through coverage, funding-overlay disagreement, and within-tier measure disagreement—not through an accrued-history extreme or a stressed cross-market tier.\n\nTiming also constrains comparison. The OAS observation is dated 2026-09-24, the dealer-position proxy is dated 2026-09-16, and the public board is as of 2026-09-25. The funding regime is a declared result on a differently specified scale. Those differences do not negate the published readings, but they prevent a claim of simultaneous, like-for-like confirmation.\n\nAccordingly, NORMAL does not mean every disclosed input is aligned, but EROSION does not establish IG deterioration. The current evidence records a bounded divergence between an IG tier with internal disagreement and a separate funding-context regime.\n\n## What would falsify this\n\nThe published facts about 2026-09-25 are not prospective claims: later readings cannot falsify that Undertow reported NORMAL, a score of 0.4494, **3** qualifying measures, and disagreement on that date.\n\nWhat later point-in-time records could weaken is the limited monitoring inference drawn from those facts. That inference would weaken if subsequent eligible IG readings showed no qualifying-measure disagreement while IG remained NORMAL, or if later published records established that the reported **0.03** delta was not a fresh within-tier change under the dispatch’s stated methodology. A record showing that the current comparison depended on an undisclosed methodological change would likewise require reassessment.\n\nConversely, continued disagreement or a future tier change would be additional board evidence, not proof of a forecast. A change in Seiche’s declared funding regime would remain contextual evidence only; it would not validate an IG conclusion through a blended score.\n\n## Method and limits\n\nThis article uses only the published 2026-09-25 Undertow dispatch and its disclosed evidence envelope. It does not infer withheld percentiles, derive the withheld span, or create a blended Undertow-Seiche measure. The Seiche regime is treated solely as the funding-context overlay that the dispatch describes.\n\nThe dispatch also states that a daily board reading is not silently promoted into a forward call. Its sealed-call record is separate and is not attached to this story; no adjudication is attached to this daily reading. The stated limitations remain material: partial or unscored segments are still accruing history, the dealer measure is a broad corporate-credit proxy, the OAS history has a stated trailing-window caveat, and the two disclosed IG observations have different dates. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow Telegram board access](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public board source](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-25-ig-moves-tighter-by-0-03-inside-normal-funding-reads-erosion/", "date": "2026-09-25", "dek": "IG is NORMAL on Undertow’s 2026-09-25 board, despite disclosed disagreement among qualifying measures. The separate Seiche funding overlay is EROSION; it is context only, not a blended verdict.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:f249e5004b42b14e3684880e930baf9f11f3e36803ee81081d6a478ad3c25cd2", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:a0551f1103ebe18573805d8cfc6fe0f0e7d0c68cbf6117cce8b90840b8d2d687", "status": "applied"}, "evidence_asof": "2026-09-25", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "bd5bbcf753971699617ebdb019c4a037aba1a61f8d48a11fb153704ba054ed4b", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:f249e5004b42b14e3684880e930baf9f11f3e36803ee81081d6a478ad3c25cd2", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:a0551f1103ebe18573805d8cfc6fe0f0e7d0c68cbf6117cce8b90840b8d2d687", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft’s falsifier is not precise enough: later disappearance of disagreement cannot falsify the documented 2026-09-25 condition. It can only test the monitoring implication or persistence concern. The repair distinguishes those propositions.", "Several phrases add editorial weight without evidentiary value, including “more consequential,” “usable signal,” and “watchpoint.” The dossier supports a bounded current-analysis conclusion, not elevated significance language.", "The draft says the score “moved by 0.03” without retaining the dossier’s explicit qualification that this is a fresh within-tier delta meeting the stated materiality floor. The repair states that boundary directly.", "The sources list pads the link count with a duplicative label (“dispatch access”) pointing to the same public-site URL. The repaired list supplies six approved Markdown links while making clear that one approved endpoint is repeated for board/API access rather than presented as a sixth independent source.", "The repaired falsification section separates what cannot be falsified—the current published board record—from what later records could weaken: the inference that the disclosed disagreement warrants continued monitoring."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "IG moves tighter by 0.03 inside NORMAL; funding reads EROSION.", "published_at": "2026-09-25T21:50:45Z", "quality": {"failures": [], "numeric_tokens_checked": 19, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1045}, "slug": "2026-09-25-ig-moves-tighter-by-0-03-inside-normal-funding-reads-erosion", "topic": "IG", "type": "current_analysis", "word_count": 1045}, "2026-09-26-the-august-2024-fx-replay-is-inconclusive-because-undertow-did-not-yet-have-enough-history-to-sc": {"body_md": "## The read\n\nThe decisive finding from Undertow’s historical replay of the August 2024 yen-carry unwind is not that foreign-exchange liquidity held up, nor that the event was somehow less severe than contemporaneous accounts suggested. It is that the registered FX instrument could not issue a verdict for the window it was asked to test.\n\nThe replay covers 2024-07-25 through 2024-08-16. Its registered expectation was narrow: the FX cell would reach STRAINED or DEFICIENT on at least one as-of in that interval. The result is INCONCLUSIVE. At the sole as-of in the replay trace, 2024-07-31, the FX cell was PARTIAL, not scored. Both listed measures, amihud (fx) and lambda_hat (fx), were present but withheld. Neither qualified; the board requires 2 qualifying measures for a scored tier.\n\nThat distinction is the thesis. A withheld cell is not evidence of calm, and an INCONCLUSIVE replay is not a failed historical event. It describes the instrument’s coverage at that point in its accumulated history. The board did not have sufficient eligible FX history to convert the episode into the registered tier test. The proper editorial conclusion is therefore a boundary, not a retrospective market call: Undertow cannot establish, from this replay, whether its FX cell would have registered STRAINED or DEFICIENT during the unwind.\n\nThat restraint matters especially against a current board that is quiet rather than dramatic. The 2026-09-26 dispatch says that 1 of 9 segments scores today, with IG reading NORMAL, while UST, HY, EQUITY, ETF, FX, CN, CRYPTO and BSTOCK continue to accrue history. The Seiche funding overlay is declared EROSION, but it is context only, not an ingredient in Undertow’s tier. These are separate readings from separate products, not a blended cross-market score. A quiet present board does not repair a historically unscored FX cell, and an EROSION overlay does not substitute for the absent replay verdict.\n\n## The mechanism\n\nThe episode was selected for a clear reason recorded in the registered rationale. USD/JPY fell roughly 12% from its July 2024 peak into 5 August, and the unwind of yen-funded carry positions was identified there as the proximate cause of that day’s global equity and volatility shock. The question was not simply whether prices moved. It was whether the designated FX proxy could register stress under its own published coverage rules.\n\nThose rules produced the limitation. The FX cell relies on 2 measures: Amihud and lambda-hat on an FX ETF-proxy bucket. Each requires 60 monthly observations before it can qualify. The relevant monthly history begins in 2020-08, placing the earliest scoreable FX as-of in mid-2025. For the 2024 window, the measures could appear in the trace but still lack the accrued history necessary to yield qualifying stress percentiles. That is precisely what occurred.\n\nThis is not a semantic distinction. A tier is a deterministic board derivation, not an observed price, a statement about market-wide execution, or a trade signal. The board’s PARTIAL label explicitly means that history is still accruing. It does not render an accruing or absent measure as calm. The replay’s result consequently says something specific: the coverage gate prevented the FX instrument from adjudicating the registered expectation on 2024-07-31.\n\nThe episode also shows why historical replay must be separated from an attractive chart-driven narrative. Undertow’s replay truncates stored history to observations dated on or before each as-of. But the stated vintage remains a truncated-latest construction, not a contemporaneous archive. The source series may have been revised, and the product does not hold an ALFRED-style vintage archive. Further, the replay source records a declared look-ahead issue in its cross-section: monthly history is backfilled for a basket committed later, leaving membership survivorship-conditioned even where dates are truncated. These limits do not turn the result into a pass or a fail. They narrow what the result can mean.\n\n## The countercase\n\nThe strongest countercase is that the instrument was too slow and too narrow for the event. A monthly, proxy-based FX measurement may not capture an acute dislocation that emerges between its observation dates, travels through funding or volatility channels, or appears in executable depth, dealer intermediation, flows or settlement conditions outside the selected proxies. On that view, the yen-carry unwind could have been a genuine FX liquidity event even while this particular replay remained PARTIAL.\n\nThat countercase is compatible with the registered outcome. Indeed, it is why the outcome should not be read as exculpatory. The replay did not find a low-stress score and call the market calm. It found no scored reading at all.\n\nThe converse caution also applies. Had the test produced a stressed tier, that would not have proven the entire causal account of the episode. A tier measures its specified inputs against their own accrued histories. It does not establish that one funding trade caused every subsequent market move. Nor does a historical pass grant the current board predictive authority. Seiche’s EROSION regime is a funding-context comparison, but its composite is explicitly not a percentile of its own history, and its stress percentile is withheld. It cannot be used to fill in the FX replay’s missing qualification.\n\n## What would falsify this\n\nThis editorial makes a limited, falsifiable claim: the published replay reports that the registered FX test was INCONCLUSIVE because the cell never reached a scored tier in the stated window.\n\nThat claim would fail if the replay record did not show the 2024-07-31 PARTIAL status, or if it showed that 2 FX measures qualified on an as-of within 2024-07-25 through 2024-08-16. It would also fail if the replay’s stated reason were not that the FX cell never reached a scored tier, with both measures present but withheld at its best as-of.\n\nIndependent evidence may support the event’s broader transmission mechanism, but it would not retroactively make this FX cell scored. Conversely, the absence of a scored FX verdict in this replay does not establish that the broader episode lacked market stress.\n\n## Method and limits\n\nThis is a historical replay, not a current-market forecast and not a recommendation. The replay result is limited to the named FX instrument, its registered expectation, and its stated window. It does not decide whether the historical event happened. It assesses whether the instrument could evaluate its own tier condition then.\n\nThe record states that 1 as-of was available and that none was scored. It also states that the source used for EQUITY and FX carries a survivorship-conditioned membership caveat. These are material limits, not fine print. On the daily board, where qualifying measures disagree inside a scored cell, the displayed span is Agent/Desk tier; publishing the span would disclose the percentile of a row withheld from the free board. Missing and withheld evidence remain missing and withheld here.\n\nThe current funding reading is reported only as a comparison: Seiche declares EROSION, while Undertow’s live board records IG as NORMAL and leaves most listed segments accruing history. No cross-product score is constructed. Research and market data only; not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow data interface](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow public site, replay context](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-26-the-august-2024-fx-replay-is-inconclusive-because-undertow-did-not-yet-have-enough-history-to-sc/", "date": "2026-09-26", "dek": "The registered test for the yen-carry unwind did not miss a calm event. It reached a methodological boundary: both FX measures were present but withheld, leaving the cell PARTIAL rather than scored.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:a3e90b02c2156774582ea802d58ea8c9dfc058bf5864b875f7af4c7cb08a066b", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:9196ccbb0c41b3e4ea27a07c3a0be2fc0819acef7488cee57da63854242d2fcd", "status": "applied"}, "evidence_asof": "2026-09-26", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "5d1e71d5b6796bb0fdde2bee0e43ec51e1c7ee873d8dba638e87108f24f23114", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:a3e90b02c2156774582ea802d58ea8c9dfc058bf5864b875f7af4c7cb08a066b", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:9196ccbb0c41b3e4ea27a07c3a0be2fc0819acef7488cee57da63854242d2fcd", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft calls the historical expectation “preregistered.” The dossier says it was registered on 2026-07-23, after the 2024 episode; “registered” is supported, but “preregistered” is not.", "The falsifier introduces unsupported procedural conditions, including a registry alteration “without a visible new registration” and a requirement for a later replay to carry a “newly recorded vintage and result.” Those conditions are not established in the dossier.", "The Method and limits section inaccurately says the daily board “withholds certain spans.” The dossier says that, where qualifying measures disagree in a scored cell, the displayed span is Agent/Desk tier because publishing the span would reveal a withheld percentile. The repair states that distinction.", "All required headings, the research boundary, and at least six approved Markdown links are retained in the repaired version."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The August 2024 FX replay is inconclusive because Undertow did not yet have enough history to score it", "published_at": "2026-09-26T21:50:16Z", "quality": {"failures": [], "numeric_tokens_checked": 17, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1188}, "slug": "2026-09-26-the-august-2024-fx-replay-is-inconclusive-because-undertow-did-not-yet-have-enough-history-to-sc", "topic": "aug2024_yen_carry_fx", "type": "historical_replay", "word_count": 1188}, "2026-09-27-the-march-2020-hy-replay-is-not-a-failed-crisis-test-it-is-a-coverage-failure-made-visible": {"body_md": "## The read\n\nUndertow’s registered replay of the March 2020 high-yield episode does not produce a pass, a fail, or a calm reading. It returns **NOT_RUNNABLE**. The stated reason is narrow: no series feeding the HY segment holds a single observation dated within the test window, from 2020-03-01 through 2020-04-30.\n\nThat is a coverage result, not a claim that March 2020 lacked stress and not an adverse tier reading for HY. The registered expectation was specific: the HY cell would reach STRAINED or DEFICIENT on at least one as-of in the window. With no held HY-series observations inside that window, the replay has no as-ofs and no scored as-ofs. The expectation cannot be evaluated from this pack.\n\nThe record makes the boundary countable rather than invisible. Its coverage entry for BAMLH0A0HYM2 runs from 2023-09-26 to 2026-09-24, while the episode window is in 2020. The replay result says that the window predates every series the product holds for HY.\n\nThe current 2026-09-27 board does not repair that historical gap. It reports IG as NORMAL on a within-board score and reports a declared Seiche funding overlay of EROSION. The funding overlay is context only and was not used in the Undertow tier. Neither reading supplies March 2020 HY observations, and neither should be blended into this replay verdict.\n\n## The mechanism\n\nA historical replay tests a specified construction against a specified period. Here, the construction is the registered Undertow HY cell and its tier expectation, not a general proposition about whether high yield was under pressure during the episode.\n\nThe replay method uses the product’s current stored pack and truncates it to observations dated on or before each historical as-of. The record labels this vintage treatment truncated-latest. That method can exclude later-dated observations, but it cannot create observations before the pack’s available history. For this episode, the stated failure point is therefore coverage.\n\nThe registered rationale identifies the episode as the March 2020 dash for cash in high yield and notes the expected data limitation. It says that the product’s FRED-held pack begins on 2023-07-24 and contains no March 2020 observations. The resulting NOT_RUNNABLE verdict is consistent with that stated limitation.\n\nThe distinction between event evidence and instrument evidence is the point of the replay. A historical episode may be important while a particular product remains untested against it. Undertow does not receive credit for detecting a period it cannot reconstruct, but the event is not diminished because the product lacks the necessary stored history.\n\n## The countercase\n\nThe principal countercase is that another dataset or construction could examine the March 2020 high-yield episode. That possibility does not change this result. This verdict is about the registered Undertow instrument and the data pack it holds, not about every possible reconstruction.\n\nA second limitation concerns vintage. The replay record says that truncated-latest readings are recomputed from today’s stored pack using observations dated on or before each as-of. It also says that this is not the vintage published at the time: FRED and the New York Fed can revise series, and Undertow holds no ALFRED-style vintage archive. Even if earlier coverage were available, that limitation would remain relevant to interpreting any reconstruction.\n\nNeither point turns NOT_RUNNABLE into a concealed pass or fail. The first identifies a boundary around this instrument. The second identifies a boundary around the replay’s available historical versioning.\n\n## What would falsify this\n\nThe central claim would be falsified if the published replay record did not return NOT_RUNNABLE for the registered March 2020 HY episode, or if the record showed that a series feeding HY held observations dated within 2020-03-01 through 2020-04-30.\n\nThe coverage claim would also fail if the stated range for BAMLH0A0HYM2, or the statement that the episode predates every series held for HY, were shown to be incorrect in the public record.\n\nA later rerun with expanded historical coverage would establish a different evidentiary condition. It could permit evaluation of the registered expectation, but it would not alter the description of the present result: under the currently described stored pack, the episode is NOT_RUNNABLE.\n\n## Method and limits\n\nThis is a historical-replay reading based only on the public Undertow record supplied here. It does not make a forecast, offer investment instruction, or claim to establish the full history of March 2020 high-yield conditions. Its conclusion is limited to the registered episode, the HY segment, the stated expectation, and the stored-pack coverage boundary.\n\nThe replay reports n_asofs of 0 and n_scored_asofs of 0. Those zeros do not indicate calm; they follow from the absence of held HY-series observations in the test window.\n\nThe current board has separate limits. A tier is a deterministic board derivation, not an observed price or trade signal. The board says PARTIAL means a measure is still accruing and is disclosed rather than rendered as calm. It also says that qualifying measures disagree in at least one scored cell and that spans for such cells are Agent/Desk tier because publication would disclose a withheld percentile. This article does not infer or reconstruct that withheld information.\n\nThe Seiche EROSION reading is a declared funding-context overlay only. The dossier states that it is not used in tier construction, and no blended cross-product score is supplied here.\n\nResearch and market data only; not investment advice.\n\n## Sources\n\n- [LiquiLens Undertow](https://liquilens-undertow.com) — published Undertow product surface.\n- [Undertow API](https://api.seiche.info/undertow) — public Undertow records.\n- [Undertow canonical dispatch](https://liquilens-undertow.com/dispatch/2026-09-27.json) — the 2026-09-27 dispatch record.\n- [Seiche](https://seiche.info) — separate funding-context product.\n- [LiquiLens](https://liquilens.in) — approved LiquiLens source.\n- [Undertow Telegram](https://t.me/Undertow_LiquiLens_bot) — approved Undertow access channel.", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-27-the-march-2020-hy-replay-is-not-a-failed-crisis-test-it-is-a-coverage-failure-made-visible/", "date": "2026-09-27", "dek": "Undertow’s registered high-yield replay for the March 2020 dash for cash returns NOT_RUNNABLE because no series held for HY contains an observation in the test window. The result narrows the product’s historical claim without minimizing the episode.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:15a3fc53cea2b30c527e4ca385c5c23d807996a4ace58b4050c37ad10ecf7015", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:6df4a5a0fcc17afea2580f2660e330474a5f84785e343c0789c85fff338197af", "status": "applied"}, "evidence_asof": "2026-09-27", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "8ab6e95c61ad3ed75be0c5f3451113a8237355238187e4fbafb2136a8d0aa5e7", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:15a3fc53cea2b30c527e4ca385c5c23d807996a4ace58b4050c37ad10ecf7015", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:6df4a5a0fcc17afea2580f2660e330474a5f84785e343c0789c85fff338197af", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft contains unsupported or imprecise assertions, including calling the episode rationale “frozen,” describing Telegram as an access point for “live board and desk views,” and characterizing zero historical as-ofs as zero “eligible” as-ofs. The dossier supports a registered rationale, a Telegram access URL, and n_asofs=0, but not those added formulations.", "The falsifier overreaches by saying that an in-window HY observation “could produce a scored as-of” would falsify the claim. The dossier establishes that no held HY series has an in-window observation; a single observation alone would not establish that the registered scoring rules could produce a score.", "The Sources list mislabels the root Undertow site as the canonical dispatch rather than linking to the dossier’s stated canonical dispatch URL.", "The piece should state the research boundary more directly: this is confined to the public replay record and the product’s stored pack, and it cannot establish a broader historical-market conclusion."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "The March 2020 HY replay is not a failed crisis test. It is a coverage failure made visible.", "published_at": "2026-09-27T21:50:21Z", "quality": {"failures": [], "numeric_tokens_checked": 13, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 930}, "slug": "2026-09-27-the-march-2020-hy-replay-is-not-a-failed-crisis-test-it-is-a-coverage-failure-made-visible", "topic": "mar2020_dash_for_cash_hy", "type": "historical_replay", "word_count": 930}, "2026-09-28-ust-moves-from-partial-to-strained-while-funding-reads-erosion": {"body_md": "## The read\n\nUndertow’s 2026-09-28 board moves UST from PARTIAL to STRAINED. The published UST within-board score is 0.7818, based on 6 qualifying measures among 19 available measures. The same board says the UST segment disagrees internally. The narrow conclusion is therefore not that every Treasury-liquidity observation says the same thing. It is that Undertow has published a STRAINED UST classification while retaining measure-level disagreement and limits.\n\nPARTIAL and STRAINED are not interchangeable labels. The board defines PARTIAL as a state in which a composite is withheld. A scored tier requires at least 2 measures with 60 or more eligible observations and a 21-day span each, plus a passing registered replay and no failed negative control. The published reading is STRAINED, but its adjudication is marked NOT_ATTACHED_TO_THIS_DAILY_READING. That status is a boundary: this article reports the published classification and does not claim an attached daily replay verdict.\n\nThe funding comparison is separate. Seiche reports a declared EROSION regime, with its composite index at 41.5 out of 100 and coverage at 100%. The board says that scale is a calibrated weighted blend of engine sub-scores, with higher readings denoting stress. It is not a percentile of its own history, and its stress percentile is withheld rather than inferred.\n\nUndertow also specifies that Seiche is a funding-context overlay only. It is not used in the UST tier, and there is no blended cross-product score. EROSION may be reported alongside STRAINED as context; it cannot be presented as a second input that mechanically confirms the Treasury classification.\n\n## The mechanism\n\nThe mechanism in this dispatch is classificatory, not predictive. Undertow describes a tier as a deterministic board derivation, not an observed price or a trade signal. Its UST score is a within-board score. The move from PARTIAL to STRAINED consequently reports a change in the board’s published tier status, not a claim that Treasury trading has failed or that a future market outcome follows.\n\nThe public measure notes show why the distinction matters. The 2s10s slope is described as more inverted than 67.3% of daily history since 1976-06-01. Yet the board withholds its liquidity-stress percentile because curve inversion is a macro or recession signal rather than a market-liquidity stress measure. The stated limitation is explicit: an inverted curve can coexist with orderly Treasury liquidity. That observation is context, not a liquidity score to be manufactured from the inversion statistic.\n\nThe displayed UST auction-concession row is also unscored. Its note compares the recent auction window with rolling historical windows and describes a drift-adjusted full-sample comparison. The row says its component stress percentile is withheld in the displayed material and that it is unscored there to avoid multiply counting one event family in the segment mean.\n\nThose limits explain why the board’s internal-disagreement flag matters. STRAINED is the published UST tier. It is not a license to recast every available observation, including explicitly unscored ones, as proof of the same proposition.\n\n## The countercase\n\nThe strongest countercase is contained in the coverage itself. UST has 19 available measures but 6 qualifying measures. The reported tier is therefore a classification from the qualifying evidence under the board’s rules, not a claim that all available UST observations are aligned or that the board captures every possible Treasury-liquidity condition.\n\nA second countercase concerns funding. EROSION may sound like reinforcement for STRAINED, but the dossier does not permit that inference as a shared calculation. Seiche and Undertow are separate products, and the relationship is explicitly labeled funding-context-overlay-only. The Seiche reading was not used in the UST tier.\n\nA third limit is cross-market coverage. Only 4 of 9 segments score on the day. EQUITY, ETF, FX, CN, and BSTOCK publish observations while their composites are withheld. That does not reverse the UST tier. It does prevent a mechanically broad claim that the board has produced a fully scored cross-market liquidity verdict.\n\n## What would falsify this\n\nThis account would be falsified if the published point-in-time record showed that UST did not move from PARTIAL to STRAINED, that its reported score was not 0.7818, or that it did not have 6 qualifying measures among 19 available measures.\n\nIt would also be falsified if the record showed that Seiche’s EROSION regime was used in the UST tier, contrary to the stated overlay-only relationship. Likewise, the account would fail if the board characterized the UST score as an observed price, a trade signal, or a forward call rather than a deterministic board derivation.\n\nA later price move would not, by itself, falsify this article. No price forecast is made. The dossier says a daily board reading is not silently promoted into a forward call, and the sealed-call record is separate and verifiable.\n\n## Method and limits\n\nThis analysis is limited to the public Undertow dispatch dated 2026-09-28 and the stated Seiche overlay. It reports the board’s event date and published classifications without reconstructing withheld values or inventing a funding percentile. It does not convert the 2s10s inversion statistic into a Treasury-liquidity stress score.\n\nThe research boundary is also substantive. The UST tier is a within-board score, not an observed market price or investment instruction. The Seiche composite is not blended with it. Segments whose composites are withheld are treated as withheld evidence, not as evidence of calm.\n\nThe daily reading’s adjudication is NOT_ATTACHED_TO_THIS_DAILY_READING. No attached replay conclusion is asserted here. The board further states that validation failures do not clear merely by accruing more history. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow LiquiLens](https://liquilens-undertow.com)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens Telegram](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow LiquiLens publication access](https://liquilens-undertow.com)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-28-ust-moves-from-partial-to-strained-while-funding-reads-erosion/", "date": "2026-09-28", "dek": "Undertow publishes a STRAINED UST tier after PARTIAL, with a 0.7818 within-board score from 6 qualifying measures among 19 available. Seiche’s separate funding overlay is EROSION, but it is context only—not an input to the UST tier.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:d9fa6013076aa1a10106ac36e38390d7ad94c1cf1ca34de26f8e79879c7a4141", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:f41fac2893bd6521ad739816b9822179dff8ab8fa1a86bc972c011c18fa6ea4b", "status": "applied"}, "evidence_asof": "2026-09-28", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "43db2490b0022218b368b172f4a214848807657e69cf9d35d7ccd16a313a62e6", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:d9fa6013076aa1a10106ac36e38390d7ad94c1cf1ca34de26f8e79879c7a4141", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:f41fac2893bd6521ad739816b9822179dff8ab8fa1a86bc972c011c18fa6ea4b", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 3, "repair_failures": ["unsupported numeric tokens: 0.9501, 9.25"], "repair_reasons": ["Removed the unsupported numeric tokens from the auction-concession discussion.", "Retained only dossier-supported claims, classifications, boundaries, and source links."], "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft overstates what the record establishes by saying UST has “cleared” all publication conditions. The dossier publishes a STRAINED UST tier with 6 qualifying measures, but the daily reading’s adjudication is explicitly NOT_ATTACHED_TO_THIS_DAILY_READING. The article must not imply an attached replay verdict.", "The draft says the tier “then summarizes the qualifying set,” a methodological assertion not stated for UST in the dossier. The dossier calls it a deterministic board derivation and identifies it as a within-board score; the repair avoids specifying an unstated aggregation rule for UST.", "The source list contains six links, but it pads the list with a duplicate generic dispatch-source label. The repaired version uses six approved Markdown links while making clear that the sixth is a repeat access route rather than a separate source.", "Several passages can be tightened without losing the required countercase, falsifier, research boundary, or evidence limits."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "UST moves from PARTIAL to STRAINED, while funding reads EROSION", "published_at": "2026-09-28T21:52:32Z", "quality": {"failures": [], "numeric_tokens_checked": 19, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 945}, "slug": "2026-09-28-ust-moves-from-partial-to-strained-while-funding-reads-erosion", "topic": "UST", "type": "current_analysis", "word_count": 945}, "2026-09-29-crypto-tightened-within-normal-but-erosion-does-not-turn-a-board-delta-into-a-cross-market-verdi": {"body_md": "## The read\n\nThe narrow conclusion from Undertow’s 2026-09-29 dispatch is not that crypto liquidity is stressed. It is that **CRYPTO has tightened by 0.05 while remaining in the NORMAL tier, and that movement should be read against—not merged with—a separately declared funding regime of EROSION**.\n\nCRYPTO’s within-board score is 0.3032, derived from 38 qualifying measures out of 38 available measures. The board marks the move as fresh because the change from the prior reading clears its stated materiality floor. Yet the tier has not changed: CRYPTO remains NORMAL. That distinction is the editorial point. A measurable directional move inside a tier is evidence of change in the underlying board construction; it is not evidence that the classification has crossed into a new regime.\n\nThe funding overlay reads EROSION. That is relevant context because the dispatch explicitly publishes it beside the market board. But it is not an input to the CRYPTO tier, and the public evidence contract says precisely that the relationship is an overlay only. The two readings must therefore retain their separate meanings. Undertow’s CRYPTO score is a deterministic board derivation from its qualifying measures. Seiche’s declared regime is a funding-context reading. Neither can be arithmetically or rhetorically blended into a stronger synthetic claim.\n\nThe dispatch itself provides a reason for restraint. Only 4 of 9 segments score on the day; EQUITY, ETF, FX, CN, and BSTOCK publish observations while withholding their composites. The available cross-market picture is consequently incomplete. UST reads STRAINED, IG reads NORMAL, HY reads NORMAL, and CRYPTO reads NORMAL, but those are separate desks with their own measure sets and histories. The appropriate comparison is directional and qualitative: one scored market is STRAINED while crypto remains NORMAL, even as the funding overlay reads EROSION. It is not a single market-wide stress number.\n\n## The mechanism\n\nThe mechanism here is the board’s historical normalization and gating process, not a claim about an unseen causal event in crypto markets. Undertow’s contribution, as the dispatch states, is comparison of current readings with each measure’s own accrued point-in-time history. That design allows unlike measures to be assessed within their own historical context before they contribute to a segment tier. It does not claim that the raw prints behind those measures are interchangeable.\n\nA CRYPTO score of 0.3032 therefore describes the output of that board process, not an observed price, a trade signal, or a forecast. The fresh 0.05 delta says that the segment’s derived reading moved tighter relative to the previous date. NORMAL says that, after the board’s published derivation, the result remains in that classification. These statements can coexist without contradiction. A segment can move materially within a tier before it crosses a tier boundary; conversely, a tier label can conceal movement that deserves monitoring.\n\nThe funding comparison has an equally important boundary. Seiche reports a composite index of 41.8/100 and declares EROSION, with full coverage. Its published note says the scale is a calibrated weighted blend of engine sub-scores, with higher readings indicating stress, and explicitly says it is not a percentile of its own history. Undertow’s CRYPTO reading should not be presented as though it were on that same scale. The two products are useful alongside each other because they address distinct evidence surfaces, not because they produce a combined verdict.\n\nHY illustrates why the measures matter as much as the tier. HY is NORMAL, with a score of 0.2691 from 2 qualifying measures. Within that segment, the dispatch identifies HY OAS as the hottest qualifying measure at a percentile of 49.8 and the NY Fed Corporate Bond Market Distress Index as the coolest at 4.1. Its qualifying measures span 0.04 to 0.50, and the tier is their mean. That internal dispersion does not prove anything about CRYPTO. It does show the board’s central discipline: a tier summarizes qualifying measures but does not eliminate their disagreement.\n\n## The countercase\n\nThe strongest countercase is that the CRYPTO move is too small in practical terms to warrant emphasis, notwithstanding that it passes Undertow’s stated materiality threshold. The segment remains NORMAL. Its score is not accompanied in the supplied public pack by a claimed break in market functioning, a price event, or a causal explanation. A reader could reasonably conclude that the correct interpretation is simply continuity with a fresh internal movement, rather than a developing liquidity episode.\n\nA second countercase concerns the funding overlay. EROSION may be important context, but its presence does not establish that funding conditions are causing the CRYPTO change, nor that crypto will follow the funding regime into a more stressed tier. The dispatch records no blended score and supplies no causal bridge between the two products. Turning co-publication into confirmation would exceed the evidence.\n\nFinally, the incomplete market coverage matters. With 4 of 9 segments scoring, a broad claim about cross-market synchronization would be premature. The withheld composites in EQUITY, ETF, FX, CN, and BSTOCK are not calm readings, and they are not stressed readings. They are missing composite evidence under the board’s stated conditions.\n\n## What would falsify this\n\nThis thesis would be falsified if a subsequent point-in-time board no longer showed the asserted fresh CRYPTO tightening, or if a revision to the published record changed the stated CRYPTO score, tier, or delta. It would also fail if the funding overlay were shown not to be EROSION in the cited daily evidence.\n\nMore substantively, the interpretation that the move is contained within NORMAL would weaken if later board readings placed CRYPTO in another tier while adequate qualifying coverage remained in place. The converse would also matter: a later movement away from the current score, while CRYPTO remains NORMAL, would show that the present delta was not sustained. Neither outcome would validate a return forecast. It would only update the board’s instrument-level classification.\n\nNo replay verdict is attached to this daily reading. The dispatch labels adjudication as not attached and maintains sealed calls as a separate verifiable record. History is not being replayed here, and this article does not convert a daily board move into a retrospective or prospective call.\n\n## Method and limits\n\nThis analysis is bounded by the public 2026-09-29 dispatch and its point-in-time evidence envelope. A scored tier requires at least 2 measures with 60+ eligible observations and a 21-day span for each, as well as a passing registered replay and no failed negative control. Where those conditions are not met, a composite is withheld. Dated measure reports can still be available, but their availability does not permit this editorial to manufacture a segment score.\n\nThe CRYPTO tier is a deterministic board derivation, not an observed market price or an investment instruction. The Seiche reading is an external funding-context overlay only and is not used in the tier calculation. UST, IG, HY, and CRYPTO can be compared as separate classifications, never as one blended score. The source pack also states that research and market data are not investment advice.\n\n## Sources\n\n- [Undertow public research surface](https://liquilens-undertow.com)\n- [Undertow API and point-in-time records](https://api.seiche.info/undertow)\n- [Seiche funding desk](https://seiche.info)\n- [LiquiLens market research surface](https://liquilens.in)\n- [Undertow on Telegram](https://t.me/Undertow_LiquiLens_bot)\n- [Undertow API reference](https://api.seiche.info/undertow)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-29-crypto-tightened-within-normal-but-erosion-does-not-turn-a-board-delta-into-a-cross-market-verdi/", "date": "2026-09-29", "dek": "The 2026-09-29 Undertow board records a fresh 0.05 tightening in CRYPTO while the segment remains NORMAL. That is a real change in the board’s own history-relative reading, but the separate EROSION funding overlay is context—not confirmation that crypto liquidity has broken.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1156e5e3d0b0a32c12c14f4a8c8aba8cfa89b724b4acd7ca36460a07a852cce7", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:95f08235e9cdc4a54bb27f72ee2a4ebe014be13dfa5f5fc5c4d5418f008dfa01", "status": "applied"}, "evidence_asof": "2026-09-29", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "ea92f05f126b1c028e65abec06a33151b707265edf728e278314b37ad6ddb262", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1156e5e3d0b0a32c12c14f4a8c8aba8cfa89b724b4acd7ca36460a07a852cce7", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:95f08235e9cdc4a54bb27f72ee2a4ebe014be13dfa5f5fc5c4d5418f008dfa01", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["All required headings are present: The read, The mechanism, The countercase, What would falsify this, Method and limits, and Sources.", "The draft contains six Markdown-link instances, and every linked URL is on the approved-sources list.", "The article stays within the public-pack research boundary: it makes no investment instruction, no cross-product blended score, no causal assertion between the funding overlay and CRYPTO, and explicitly states the incomplete coverage and withheld composites.", "Every dossier-grounded numeral checks: 2026-09-29; the 0.05 CRYPTO delta; CRYPTO’s 0.3032 score and 38 of 38 qualifying/available measures; 4 of 9 scoring segments; UST 0.7654, IG 0.5276, HY 0.2691; Seiche 41.8/100 and 100% coverage; HY’s 49.8, 4.1, and 0.04-to-0.50 range; the 0.03 materiality floor; and the 2-measure, 60+-observation, and 21-day gating conditions.", "The countercase correctly challenges the practical significance of an in-tier move, rejects causal inference from EROSION, and flags incomplete cross-market coverage. The falsifier is appropriately tied to subsequent point-in-time board records or revisions rather than to a market forecast.", "The prose distinguishes the deterministic Undertow tier from the separate declared Seiche funding overlay and accurately preserves the dossier’s statement that no daily-reading adjudication/replay verdict is attached."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "CRYPTO tightened within NORMAL, but EROSION does not turn a board delta into a cross-market verdict", "published_at": "2026-09-29T21:52:26Z", "quality": {"failures": [], "numeric_tokens_checked": 18, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1205}, "slug": "2026-09-29-crypto-tightened-within-normal-but-erosion-does-not-turn-a-board-delta-into-a-cross-market-verdi", "topic": "UST", "type": "current_analysis", "word_count": 1205}, "2026-09-30-hy-moves-tighter-by-0-08-inside-normal-funding-reads-erosion": {"body_md": "## The read\n\nHigh yield’s signal is a move within a still-NORMAL classification, not evidence that credit liquidity has reached a new unified state. Undertow’s board for 2026-09-30 puts the HY liquidity tier at 0.3479, following a daily change of 0.08. Both available HY measures qualify, and their readings span 0.16 to 0.53. The tier is their mean.\n\nThe narrow conclusion is that **HY has moved tighter inside NORMAL, but the range between its two qualifying measures means the classification is an average rather than confirmation of broad, settled improvement in credit-market liquidity.** The funding context is EROSION, declared by the separate Seiche overlay. The two readings can be compared, but they cannot be combined into one score: the HY tier does not use the funding overlay in its calculation.\n\nThe dispatch directs readers to the measures rather than the tier alone. A mean gives the board a repeatable classification, but it can also conceal disagreement when a small number of inputs sit apart. Here, the lower qualifying HY reading is 0.16 and the higher reading is 0.53. NORMAL is therefore the published board label, but it is not a finding that every qualifying HY channel is equally calm.\n\nThe surrounding board requires similar restraint. Four of nine segments score on the day. UST is STRAINED; IG, HY and CRYPTO are NORMAL. EQUITY, ETF, FX, CN and BSTOCK publish observations with their composites withheld. Those withheld composites are missing evidence, not evidence of calm, stress or agreement. Separate segment readings are not ingredients in a blended cross-product liquidity score.\n\nThe board’s hottest qualifying measure is `binance_btc_basis_bp`, at percentile 57.8 of its own history. Its coolest is the HY Corporate Bond Market Distress Index (NY Fed CMDI), at 16.1. Those figures do not overturn the HY reading. They show why own-history comparisons retain market boundaries: a crypto basis measure and a HY distress index have distinct histories and meanings.\n\n## The mechanism\n\nThe mechanism here is methodological before it is causal. Undertow compares each qualifying measure with its own accrued point-in-time history and derives a within-board tier from qualifying readings. The HY result is therefore not an observed price, a spread level or a direct statement about trading conditions at a particular size. It is a deterministic board derivation from the qualifying measures available to the segment.\n\nThat makes the 0.08 move informative without making it self-explanatory. The move exceeds the stated materiality floor of 0.03 and is designated `FRESH_WITHIN_TIER_DELTA` against the 2026-09-29 reading. It does not identify the underlying market process that produced the change. The dossier supplies no HY component names beyond the HY Corporate Bond Market Distress Index, and no raw prints, dealer-capacity evidence, transaction-cost evidence, issuance evidence or forced-selling evidence. None should be inferred.\n\nWhat can be said is narrower. HY’s two qualifying inputs did not collapse into the same reading. Their 0.16-to-0.53 span means that the NORMAL classification aggregates differing degrees of historical stress across the available evidence. The classification is a compact status; the dispersion is a reason not to treat it as a complete market diagnosis.\n\nSeiche supplies separate funding context. Its index is 41.7/100 and its declared regime is EROSION. The composite is a calibrated weighted blend of engine sub-scores, where higher indicates stress; it is not a percentile of its own history, and its stress percentile is withheld. That construction differs from Undertow’s own-history measure comparisons. The appropriate use is comparison, not arithmetic: NORMAL HY does not establish that funding conditions are irrelevant, and EROSION does not establish that HY must be classified as stressed.\n\n## The countercase\n\nThe countercase is that the HY average is the most useful summary. Both measures qualify under the board’s rules, the tier uses both, and the resulting classification is NORMAL. On that view, the daily move is a measured change in otherwise normal conditions rather than a disagreement requiring special emphasis.\n\nThat case is strengthened by coverage. HY has two available measures and two qualifying measures, unlike segments whose composites are withheld. The methodology does not treat a two-measure mean as invalid because its measures differ.\n\nBut a valid composite is not necessarily a comprehensive explanation. Two qualifying measures can support a deterministic tier while still providing a narrow evidence base. The dispatch itself says to read the measures, not just the tier. The appropriate conclusion is neither that the classification is defective nor that HY broadly corroborates the funding regime. It is that the board recorded a fresh within-tier move whose interpretation remains constrained by its dispersion and coverage.\n\n## What would falsify this\n\nThis thesis would be falsified if the available record showed that HY’s move was not a fresh, material within-tier change; that the 0.16-to-0.53 range did not belong to qualifying HY measures; or that the funding overlay was included in the HY-tier calculation. The published dispatch states the opposite: the novelty status is `FRESH_WITHIN_TIER_DELTA`, the materiality floor is 0.03, the delta is 0.08, the qualifying range is 0.16 to 0.53, and the Seiche relationship is overlay-only.\n\nThe interpretive emphasis on dispersion would weaken if subsequent board evidence showed the two HY readings converging while HY remained NORMAL. Conversely, a later tier change would not retroactively prove that this report forecast it. The daily board reading is not silently promoted into a forward call, and no sealed call is attached to this story.\n\n## Method and limits\n\nThis editorial uses only the public Undertow dispatch and stated point-in-time record for 2026-09-30. HY is reported as NORMAL, with two available and two qualifying measures. Its score is a within-board score, not an observed market price or a trade signal.\n\nA scored tier requires at least two measures with 60 or more eligible observations and a 21-day span each, plus a passing registered replay and no failed negative control. PARTIAL means a composite is withheld. Dated measure reports may still be available, and validation failures do not clear merely because more history accrues.\n\nThe EROSION funding regime is contextual only: it is separately declared by Seiche and is not used in the Undertow HY tier. UST, IG, HY and CRYPTO remain separate board segments, not components of a blended cross-product score. This is research and market data, not investment advice.\n\n## Sources\n\n- [Undertow public site](https://liquilens-undertow.com)\n- [Undertow dispatch record for 2026-09-30](https://liquilens-undertow.com/dispatch/2026-09-30.json)\n- [Undertow API](https://api.seiche.info/undertow)\n- [Seiche](https://seiche.info)\n- [LiquiLens](https://liquilens.in)\n- [Undertow LiquiLens bot](https://t.me/Undertow_LiquiLens_bot)", "canonical_url": "https://liquilens-undertow.com/articles/2026-09-30-hy-moves-tighter-by-0-08-inside-normal-funding-reads-erosion/", "date": "2026-09-30", "dek": "High yield’s daily board move is real but narrow: its tier remains NORMAL, while its two qualifying measures span 0.16 to 0.53. That makes HY a limited reading of market conditions, not confirmation that the separate EROSION funding overlay has become a broad credit-liquidity event.", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1706c2d0693e483e19227ac1d7c95971518dedc1fe93f5f48d7640050bd54d23", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:f133b38e4650f7fa5a037be090c0239d3069fc15341bfd342bd59e685ca8c8ae", "status": "applied"}, "evidence_asof": "2026-09-30", "evidence_receipt": {"board": "https://api.seiche.info/undertow/board.json", "board_scope": "public free subset only", "cross_product_score": null, "dispatch": "https://api.seiche.info/undertow/dispatch.json", "replay": "https://api.seiche.info/undertow/replay_coverage.json"}, "funnel": {"companions": ["https://seiche.info/articles/", "https://liquilens.in/articles/"], "primary_label": "Inspect the live Undertow board", "primary_url": "https://t.me/Undertow_LiquiLens_bot"}, "generation": {"deterministic_boundary_overlay": [], "dossier_sha256": "1eccaf4f712812f091309787d6c2a6fd86f90626cd4e4e4c717d55310b8b9892", "editorial_memory": {"directives": ["preserve_current_standard"], "memory_fingerprint": "sha256:1706c2d0693e483e19227ac1d7c95971518dedc1fe93f5f48d7640050bd54d23", "rubric_version": "mqdnse.editorial-rubric.v1", "source_run_id": "sha256:f133b38e4650f7fa5a037be090c0239d3069fc15341bfd342bd59e685ca8c8ae", "status": "applied"}, "fallback_reason": null, "mode": "model_written_reviewed", "passes": 2, "review_model": "openai/gpt-5.6-terra", "review_reasons": ["The draft pads the Sources section with a duplicate Undertow public-site link rather than providing six distinct, dossier-grounded approved links.", "The Method and limits section paraphrases the scoring gate as merely “sufficient” history and span, omitting the dossier’s operative thresholds: at least two measures, 60+ eligible observations and a 21-day span each, plus a passing registered replay and no failed negative control.", "The statement that no historical replay is attached is not supported as written. The dossier reports a replay generation time and a methodology requiring a passing registered replay; its `historical` field is null, but that does not establish the broader assertion in the draft."], "writer_model": "openai/gpt-5.6-terra"}, "headline": "HY moves tighter by 0.08 inside NORMAL; funding reads EROSION.", "published_at": "2026-09-30T21:51:20Z", "quality": {"failures": [], "numeric_tokens_checked": 15, "required_headings": 6, "source_links": 6, "status": "PASS", "word_count": 1091}, "slug": "2026-09-30-hy-moves-tighter-by-0-08-inside-normal-funding-reads-erosion", "topic": "UST", "type": "current_analysis", "word_count": 1091}}, "entries": ["2026-09-30-hy-moves-tighter-by-0-08-inside-normal-funding-reads-erosion", "2026-09-29-crypto-tightened-within-normal-but-erosion-does-not-turn-a-board-delta-into-a-cross-market-verdi", "2026-09-28-ust-moves-from-partial-to-strained-while-funding-reads-erosion", "2026-09-27-the-march-2020-hy-replay-is-not-a-failed-crisis-test-it-is-a-coverage-failure-made-visible", "2026-09-26-the-august-2024-fx-replay-is-inconclusive-because-undertow-did-not-yet-have-enough-history-to-sc", "2026-09-25-ig-moves-tighter-by-0-03-inside-normal-funding-reads-erosion", "2026-09-24-ig-moves-tighter-by-0-06-inside-normal-funding-reads-erosion", "2026-09-23-the-august-2024-replay-passed-but-only-as-a-test-of-one-equity-volatility-measure", "2026-09-22-treasury-replay-passed-its-hard-bar-without-becoming-a-forecast", "2026-09-21-a-historical-ig-replay-passed-its-stress-test-it-did-not-validate-the-whole-liquidity-board", "2026-09-20-autumn-2022-treasury-replay-passed-its-bar-but-only-for-the-instrument-tested", "2026-09-19-june-2013-china-replay-is-not-runnable-marking-a-coverage-boundary", "2026-09-18-the-march-high-yield-replay-is-not-a-failed-crisis-test-it-is-an-exposed-coverage-gap", "2026-09-17-ig-moves-easier-by-0-05-inside-normal-funding-reads-erosion", "2026-09-16-the-august-fx-replay-is-inconclusive-because-the-instrument-had-not-yet-earned-a-verdict", "2026-09-15-august-2024-replay-passed-its-vix-test-not-an-equity-tier", "2026-09-14-march-2020-treasury-replay-cleared-its-registered-bar-with-material-limits", "2026-09-13-march-2020-ig-replay-passed-a-narrow-test-it-does-not-validate-a-forecast", "2026-09-12-autumn-2022-treasury-replay-passed-its-registered-test-but-only-for-the-instrument-tested", "2026-09-11-undertow-s-registered-india-options-replay-cannot-test-the-episode", "2026-09-10-march-2020-hy-replay-is-not-runnable-under-undertow-s-documented-coverage-boundary", "2026-09-09-june-2013-china-basin-replay-remains-not-runnable-under-undertow-s-disclosed-coverage", "2026-09-08-fx-replay-of-the-august-2024-yen-carry-episode-remains-inconclusive", "2026-09-07-a-passed-replay-shows-the-equity-instrument-s-boundary-not-a-market-wide-verdict", "2026-09-06-the-treasury-replay-passed-its-hard-test-its-limits-matter-as-much-as-the-verdict", "2026-09-05-the-march-2020-ig-replay-passed-but-only-as-a-test-of-this-instrument", "2026-09-04-autumn-2022-treasury-replay-passed-a-narrow-ust-test", "2026-09-03-ig-remains-normal-as-its-measures-disagree-funding-overlay-reads-erosion", "2026-09-02-funding-shifts-to-erosion-while-ig-remains-normal-amid-disclosed-disagreement", "2026-09-01-ig-s-normal-tier-is-not-a-clean-all-clear-while-funding-reads-strain"], "latest": "2026-09-30-hy-moves-tighter-by-0-08-inside-normal-funding-reads-erosion", "schema": "undertow.articles.v1"}