Freeze point: elder-care/docs/research-inquiry.md as committed at 3d1785b (2026-08-03). Every deviation from protocol-specified method is logged here with a reason and its effect on findings. Per M6, this log is part of the record, not an appendix to it.
| # | Date | Deviation | Reason | Effect on findings |
|---|---|---|---|---|
| 1 | 2026-08-03 | Anchor verification for rows 1, 2, 3, 6, 7, 11 delegated to six parallel subagents; results QA-reviewed by the primary session before entry, with two independent WebSearch spot-checks on the two highest-stakes claims (row 6 WA Cares payout date, row 7 CMS rule status) | Search fan-out is parallelisable per housing's precedent; the two-source and root-trace rules were carried into each brief verbatim | No subagent figure was entered unverified. Rows 6 and 7 independently confirmed by the primary session against separate outlets (AARP/DSHS/KOMO for WA Cares; AHA/Federal Register/CANHR/Center for Medicare Advocacy for the CMS rule) |
| 2 | 2026-08-03 | Row 7's moratorium end-year is unreconciled: CMS's own repeal cites the H.R. 1 moratorium as running to 2035; one law-firm summary states 2034 | Primary Federal Register text not independently pulled (redirect-blocked in subagent session; not re-attempted) | Recorded both figures rather than picking one. Resolved 2026-08-10 by Phase 1 verification (#32): the answer is 2034. Pub. L. 119-21 §71111 ends the moratorium "September 30, 2034," and CMS's repeal rule (90 FR 55687) says "September 30, 2034" and "before October 1, 2034" throughout — the 2035 figure this entry attributed to CMS does not appear in CMS's rule |
| 3 | 2026-08-03 | Rows 4, 5, 8, 9, 10, 12 not attempted in this pass | Phase 0 time budget went to the ★ rows first, per M6 | Rows left blank, not guessed. Queued for full §2/§4/§5/§6/§8 execution |
| 4 | 2026-08-03 | §2 baseline pulled population 65+/85+ by state (ACS B01001) only. LTSS spending by payer (Medicaid/Medicare/OOP/private LTC/imputed unpaid care) vs. utilization by setting was not pulled | No single bulk-file CMS/Census source publishes this cross-tab directly (unlike ACS population data); it likely requires combining CMS National Health Expenditure Accounts, CMS-64 Medicaid expenditure reports, and NHE-derived LTSS-specific breakdowns (e.g. the kind KFF/MedPAC compile), which is a heavier pull than Phase 0's time budget covers | Queued as baseline next-increment. §2's population leg is committed and auditable in elder-care/baseline/; the spending leg is the larger remaining lift for full execution |
| 5 | 2026-08-03 | §12's precedent scan (WA Cares + CLASS Act + one HCBS waiting-list state case) delegated to subagents; WA Cares was already covered in depth by the anchor-6 work (deviation #1) so only CLASS Act and the state case were dispatched fresh | Avoids redundant research spend | No correction needed on QA; both reports disclosed their own confidence gaps rather than presenting unverified figures as settled (see #6) |
| 6 | 2026-08-03 | The HCBS waiting-list case study (Texas) could not verify a current STAR+PLUS HCBS-specific headcount or Texas's institutional-vs-HCBS spending mix — HHSC's primary reporting pages returned HTTP 403 to automated fetch, and Texas is documented as excluded from both CMS's Mathematica rebalancing report and AARP's 2023 LTSS State Scorecard's spending-balance indicator for data-accuracy reasons | Primary source inaccessible in session; the state itself has a documented reporting gap to the two leading national trackers | Recorded as an explicit non-finding rather than filled with a secondary-source guess: "cannot confirm whether Texas is institutionally biased" is itself entered as the finding for that half of the case study, per M1's rule that a named data gap is a finding |
| 7 | 2026-08-03 | §3's H3.1/H3.2 were tested without pre-registered M3 adjudication criteria (refuted if / supported if / indeterminate if) — the seed protocol states the hypotheses but not their adjudication conditions, unlike housing's research/ws03-adjudication-criteria.md |
Full execution had not yet reached the point of writing per-hypothesis criteria before this pass ran | Verdicts in research/ws03-findings.md are reported conservatively (INDETERMINATE-BUT-SUGGESTIVE for H3.1; split exposure/effect verdict for H3.2) rather than retrofitting criteria to match the result. Writing H3.1/H3.2 adjudication criteria before any further §3 evidence collection is queued as a pass-2 job |
| 8 | 2026-08-03 | Anchor row 8 (HCBS waiting-list total) filled in from data surfaced incidentally during the §3 ARPA §9817 research, not a dedicated §5 pass | The KFF waiting-list tracker came up directly in the workforce-funding search | Filled opportunistically rather than left blank; §5 still needs its own dedicated pass for the duplicate-counting and institutional-bias questions the anchor doesn't cover |
| 9 | 2026-08-03 | §8's Credit for Caring Act cost comparison has no CBO/JCT score to cite — the bill has never reached committee markup in any Congress since 2016 | JCT convention only scores bills at markup; none has occurred | Recorded as an explicit non-finding (per M1, absence of data is a finding) rather than filled with an unsourced guess. The "bigger than NFCSP, smaller than HCBS" framing in ws08-findings.md is labeled an extrapolation, not a sourced figure |
| 10 | 2026-08-03 | The self-direction-vs-agency-workforce interaction (does Cash & Counseling growth partly reflect agency labor shortages, connecting §8 back to §3's H3.1) was not resolved — the one source that raises it (PMC12605748) explicitly calls it an untested hypothesis | No dedicated causal study located in this pass | Reported as an open gap in ws08-findings.md rather than treated as supporting evidence for H3.1. Queued as a joint §3/§8 job for full execution |
| 11 | 2026-08-03 | §6/§9's state staffing-mandate list (California, New York, Florida, Arkansas, DC HPRD figures) is sourced from a legal/MACPAC secondary aggregator, not independently checked against each state's current statutory text | Session scope; verifying ~10 states' current codes directly was not attempted | Recorded with the caveat stated in ws06-09-findings.md rather than presented as admin-data-verified. Primary-text verification of each figure is queued, mirroring housing's deviation #12 (PPI scores from secondary summaries) |
| 12 | 2026-08-03 | §6/§9's enforcement-durability finding rests on a single state case (New York) presented as illustrative, not representative | Only one state's enforcement history was found documented in sufficient depth in this pass | Explicitly labeled as one case, not a general pattern, in ws06-09-findings.md. Whether NY's enforcement gap generalizes to other mandate states is queued for full execution |
| 13 | 2026-08-03 | §5's H5.1 test (Indiana) found one of its two spending-split figures (AARP 2023 Scorecard, ~23% HCBS) could not be re-verified against AARP's primary page directly — the state's own legislative report figure (34%) was independently confirmed | AARP's site rate-limited automated re-fetch in this pass | Both figures recorded with the discrepancy stated in ws05-findings.md rather than silently picking one; the directional finding (well below the 53% national average either way) does not depend on which figure is exact |
| 14 | 2026-08-03 | §11's candidate-architecture scorecard is a single-scorer first pass, not the independently re-scored, reconciled version M6 requires before a scorecard can be cited as adjudicated. PACE and private-LTC-insurance market reform were left unscored (no dedicated research this session) | Full scoring of all 9 candidates plus an independent re-score was beyond this session's remaining scope | Recorded explicitly in ws11-scorecard.md's closing section rather than presented as final. Independent re-score + reconciliation log, and dedicated PACE/reinsurance research, are queued for full execution |
| 15 | 2026-08-03 | Four remaining workstreams (§2 spending, §4 insurance-reform, §7 assisted living, §10 international precedents) were dispatched in one parallel batch rather than sequentially, to pick up pace at the user's request | User explicitly asked how to speed up given other filings due this week | No QA shortcuts taken — each report still independently reviewed before being written into a findings file; parallelizing changed wall-clock only, not verification rigor |
| 16 | 2026-08-03 | §2's LTSS spending baseline could not isolate Medicare-alone or private-LTC-insurance-alone shares of the $415B KFF total — every source found bundles them into a combined "other public and private" category | No source located that separates these two payers for a current year | Recorded as a named gap (~21% combined, unsplit) rather than guessed at. Third instance this session of "the US doesn't measure this domain the way its policy debate assumes," after §5 (Texas) and §8 (caregiver valuation) — flagged as a candidate cross-cutting whitepaper theme |
| 17 | 2026-08-03 | §7's H7.2 (Medicaid AL footprint growth) and the AL-vs-SNF incident-rate comparison both rest on directional evidence with no clean longitudinal or rate-comparison dataset found | Neither appears to exist in public research at present, per the dispatched agent's search | Recorded as explicit data gaps in ws07-findings.md rather than filled with the plausible-but-unverified inference |
| 18 | 2026-08-03 | §10's Japan findings rely partly on a methodological critique (Geyer, PMC7054649) whose core claim — that cross-country LTC spending comparisons undercount informal care inconsistently — was not independently verified against the critique's own primary data | Session scope | Recorded as the critique's claim, attributed to its source, not adopted as this protocol's own finding |
| 19 | 2026-08-03 | Per M6, a full red-team pass (Workflow, 11 agents: 4 attack/defend pairs + independent re-score + reconciliation + whitepaper synthesis) was run against every workstream finding, producing 22 resolved items (see red-team-log.md) and 3 reconciled scorecard cells. The corrections are fully applied in red-team-log.md, ws11-scorecard.md, ws05-findings.md, and the anchor table (rows 2, 11) — but NOT line-edited into the prose of ws03, ws04, ws06-09, ws07, ws08, and ws10-findings.md, which still read pre-red-team in places (e.g. ws03 still says H3.1 is "directionally suggestive" in its own body text, though the corrected verdict is plain INDETERMINATE per red-team-log #5) |
Full line-edit pass across 6 files was deprioritized in favor of completing the whitepaper first, per explicit user request | This is a real, disclosed inconsistency, not a hidden one: red-team-log.md is the authoritative correction layer and whitepaper-07-content.md / the final HTML report incorporate every correction, but a reader opening ws03-findings.md etc. directly, without also reading red-team-log.md, will see the uncorrected verdict language. Queued: line-edit pass across the 6 affected files to bring prose in line with the log |
| 20 | 2026-08-03 | First whitepaper build used the wrong design system. Built elder-care/report/gubment-eldercare-report.html against brand/gubment.css and childcare/report/ (Ultra/Special Elite/Public Sans fonts, single-column manila sheet) — both are stale, superseded artifacts. The live site (gubment.com) and the actual current template (site/childcare/index.html, site/fonts/) use a different, later system (Bricolage Grotesque/Newsreader/IBM Plex Mono, two-column paper-sheet layout with masthead/abstract/TOC/aside). User caught the mismatch against the live site |
brand/BRAND.md and brand/gubment.css were never updated after the site's design evolved past them, and childcare/report/ is an earlier draft that was superseded by site/childcare/ without being removed — nothing in the repo flags either as deprecated |
The incorrect build was deleted, not left alongside the correct one. Rebuilt at site/elder-care/index.html against the verified live template, screenshot-checked against the actual rendered live site. Follow-up recommended, not done here: brand/ and childcare/report/ should either be updated to match site/ or marked deprecated, so this mistake doesn't repeat on GBMT-8/9 |
| 21 | 2026-08-03 | The published whitepaper (site/elder-care/index.html) references og-card.png at the site root and its own canonical URL, but no elder-care-specific OG image exists — it will fall back to childcare's og-card.png, which describes childcare, not elder care, if shared on social media before a real one is made |
Per M8, plain-talk artifacts (social card, PDF) are a separate, later publication stage; image generation was out of scope for this pass | Not a broken link (a card will render), but a wrong one until a proper elder-care og-card.png is generated. Queued as the plain-talk-artifact stage, mirroring childcare/report/og-card.html+.png's pattern once childcare/report/'s deprecation (see #20) is resolved |
| 22 | 2026-08-06 | Late PACE/§12 evidence pass deliberately stops before editing the scorecard, report, or site | Two open PRs currently touch the elder-care site; adding a newly researched architecture requires fresh ranking arithmetic and reconciliation, not a one-row patch | ws12-pace-sequencing.md records the evidence and a provisional row. PACE remains excluded from published rankings until the full board, whitepaper, and site can move together |
| 23 | 2026-08-06 | Completed the line-edit pass queued in #19 across ws03, ws04, ws06-09, ws07, ws08, and ws10 | The red-team log had already adjudicated the corrections; leaving the original prose in place made the research record internally inconsistent | Underlying files now carry the authoritative downgraded scopes and confidence tiers. No new empirical claim was added by this edit |
| 24 | 2026-08-06 | The late PACE evidence pass began without PACE-specific M3 adjudication criteria | PACE was a named candidate architecture but had been left as an unscored placeholder; no criteria were written before the evidence search | The result is labeled exploratory and provisional in ws12; it cannot be entered into the published board until a future pass pre-registers integration criteria and recomputes the rankings |
| 25 | 2026-08-06 | The PACE late pass added a CMS directory enumeration after its initial evidence review | The existing scorecard's “historically hard to scale” phrase lacked even a current administrative baseline; the extraction is descriptive rather than an attempt to infer causation | ws12 now reports contracts and reported enrollment with the file's masking and coverage limits. It does not upgrade the causal scalability verdict or the provisional scorecard row |
| 26 | 2026-08-06 | Wrote PACE integration criteria after, rather than before, the initial late evidence pass | The original protocol had no PACE-specific M3 criteria; retroactive criteria would misdescribe the earlier search | The criteria govern only future comparative/causal evidence and scoring; they do not upgrade the exploratory late-pass finding |
| 27 | 2026-08-06 | Corrected the CMS July 2026 PACE directory enrollment extraction after an independent recheck found 201 visible values totaling 79,758, with five suppressed/blank rows | The initial extraction undercounted the numeric enrollment field | The administrative baseline is now reported as a visible-enrollment lower bound; it does not change the causal scalability verdict or provisional scorecard treatment |
| 28 | 2026-08-06 | Prospective PACE evidence check produced no qualifying support or refutation for any of the four new M3 criteria | The newly located material is descriptive, cross-sectional, a payment brief, a survey, qualitative research, or a trial protocol rather than the criteria's required comparative/causal evidence | All four criterion-level cells remain explicitly indeterminate; this confirms that PACE should not be promoted from the provisional research record into the published scorecard |
| 29 | 2026-08-06 | Added a prospective HCBS-rebalancing criterion after the original workforce pass | Newly located BIP quasi-experiments evaluate a bundled, target-bound programme rather than the original generic wage-versus-funding claim | §13 records narrow package-level support; generic spending, wage-floor, and self-financing claims remain indeterminate or unsupported, and no scorecard cell changes without an independent re-score |
| 30 | 2026-08-10 | Ran the structurally blinded re-score this filing had been claiming since publication but never had, and amended the anchored scale in the course of it. The pass-2 re-score (deviations #19, red-team-log) was an in-workflow agent instructed not to read ws11-scorecard.md while holding filesystem access throughout — instruction-level blinding, not the structural blinding M6 requires. Pass 3 used scripts/batch-rescore.py: one tool-less request with the evidence inlined, so the scorer could not reach the scorecard, red-team log or deviations log by any route. Separately, the pass amended the anchored scale — anchor 3 widened to cover "no evidence either way", anchor 1 narrowed to require documented adverse effect |
The published pages stamped "Survives independent re-score" on the strength of a pass that did not meet the standard, and HANDOFF carried elder care's genuine re-score as still owed. On the scale: as authored, anchors 4 and 5 required cited evidence while anchors 1 and 2 accepted its absence, so an unresearched architecture was scored down and could never be scored up — the same one-directional evidence floor the housing re-score named on 2026-08-06 and the same correction applied to housing, media and drugs. Elder care was excluded from that pass (commit 40b7372) precisely because a re-score "already existed" |
The blind matrix differed on 26 of 40 ranked cells; 25 were corrected and 1 kept as a logged judgment split, so the board changes on 25 of 40. 14 cells matched outright. The filing's most-repeated negative finding — "private LTC insurance market reform ranks last under all three weightings, with no exceptions, the clearest most stable negative finding in the whole scorecard" — is withdrawn as stated: four of that row's five cells rested on no record evidence in either direction, and its last-place rank survives only under the unamended scale. Also withdrawn: "Cash & Counseling ties or trails nationalized WA Cares insurance and the wage floor." Surviving and strengthened: Cash & Counseling leads caregiver-relief weighting decisively, and does not win under every weighting. New and robust under both scales: the direct-care wage floor no longer leads workforce weighting and now ties the do-nothing cash comparator, so §12's workforce-first sequencing prior draws no support from this board. Because the scale amendment is load-bearing, both readings are published side by side per protocol S5. Full reconciliation: ws11-rescore-log.md; raw scorer output and every redaction: ws11-blind-scores-2026-08-10.md |
| 31 | 2026-08-10 | Three passages in the blinded evidence packet were redacted before it was sent to the scorer, and PACE was scored from an evidence-only extract of ws12-pace-sequencing.md rather than the whole file |
ws04-insurance-reform-findings.md states four of the private-LTC row's five cell values verbatim and points at the scorecard; ws12 names the cell its cost score should take and carries a full provisional row plus a sequencing recommendation. Left in, these would have leaked the prior board into a nominally blind pass — the contamination GBMT-10 recorded and M9/M10 warn about |
Each redaction was replaced with a visible [REDACTED FROM THIS PACKET: …] marker rather than silently deleted, and all three are itemised in ws11-blind-scores-2026-08-10.md with the residual risks that were not redacted (dangling links to the excluded logs; ws13's directional "stronger … than an ARPA-only record" phrasing, kept because scoring discipline 6 requires honouring an instruction not to resolve a tradeoff). PACE and the Credit for Caring Act both now have independent blind rows recorded; neither is entered on the published board — PACE because deviations #22/#24/#26/#28 bar it until the M3 integration criteria are met, the credit because reversing pass 2's split-and-unrank decision is outside a re-score's mandate |
| 32 | 2026-08-10 | Ran Phase 1 of the Verification Protocol against this filing — the first check of it against primary law and primary literature rather than against itself. 689 claims extracted into two ledgers (public, 425; record, 264), every one given a verdict, full adjudication in verification-log.md. Extraction, retrieval and adjudication all ran in one session on the adjudicating model; nothing was delegated, so the V3 guardrail is reported as a zero-delta self re-extraction of Part 5 rather than as a diff against a cheaper ledger | The check was owed to every filing and this one had never had it — the re-score log said so in terms ("This is a re-score, not a fact-check … Whether the underlying findings are true is Phase 1 of the Verification Protocol, which has not yet run on this filing") | 40 CORRECTED, 42 OVERSTATED, 90 UNVERIFIABLE, 4 dual as-of (V6). Three clusters carry the weight: the private-equity mortality study, which this filing calls its best evidence, had four of its six published figures wrong (sample 4.2M not >7M; OLS ~2% of the mean not +10%; ~22,500 excess deaths not ~20,150; +8%/+6% billed spending not +19%) and a fifth — "~50% higher antipsychotic use" — that appears nowhere in the paper at any vintage; the "ten-year" staffing-rule moratorium is nine years three months (Pub. L. 119-21 §71111, to 2034-09-30), which also closes deviation #2's open 2034/2035 dispute in favour of 2034; and the workforce evidence base is 63% under-65 and mostly I/DD by KFF's own text, a scope the sources page discloses in one clause and the whitepaper never did. Corrections landed on both public pages and across nine record files |
| 33 | 2026-08-10 | Phase 1 found that deviation #23's line-edit pass ("completed", 2026-08-06) had missed two of the six files: ws02-spending-findings.md's confidence tier still justified medium-high by "multiple KFF compilations converge" after red-team #6 established they share upstream NHEA assumptions, and ws04's §4 digest headline still read "Reinsurance was actually tried by a state, and it failed" after red-team #8 softened it |
A completion claim was checked against the files rather than taken on the log's word | Both lines corrected; #23's claim of completion is now accurate. Recorded rather than silently fixed, because "the line-edit pass is done" was itself a published claim |
| 34 | 2026-08-10 | Phase 1 found that the published deviations digest on site/elder-care/sources/index.html carried 21 entries against a committed log of 31, and omitted entries 22–29 entirely — the whole PACE late-pass sequence, including the M3-criteria-written-afterwards disclosures and the CMS-directory extraction correction. The whitepaper's receipts line described the record as "all public" |
M6 makes the deviations log part of the record, not an appendix to it; a two-thirds digest under a wrong total is not the log | Count corrected to 31, entries 22–29 published as digest rows, and the whitepaper's "all public" claim replaced with a statement of what was and wasn't published. The source-page regeneration that would have caught this had never been run against a moved deviations log |
| 35 | 2026-08-10 | Ran Phase 2 of the Verification Protocol — the steelman — in a separate session and worktree from the Phase 1 fact-check, as S1 requires, and after Phase 1's corrections had landed. Per S2 the tilt was derived from this filing's own text and committed before any steelman evidence was gathered; the three targets are the verdict printed in the filing rail, the architecture leading the board, and the heading over the scorecard. Unlike GBMT-1's Phase 2, all three cases were built and adjudicated in one session rather than in independent builder contexts | GBMT-1's separate-builder pattern buys independence of the builder from the adjudicator; this pass bought it differently, by fixing the targets in writing before looking for evidence and by recording the two points where the steelman lost outright. The strongest items are single primary documents a reader can check in a minute, so they do not rest on trusting the builder | Two published claims withdrawn. (1) Part 9's heading, "the best-evidenced fix isn't the one anyone's proposing," is false: Congress enacted the Cash & Counseling design as a permanent state-plan option in 42 U.S.C. §1396n(j) (Pub. L. 109-171 §6087(a), 2005) and again as Community First Choice in §1396n(k) (Pub. L. 111-148 §2401) with a 6-percentage-point FMAP bonus; ~1.5M people self-direct in all 50 states + DC, up 23% since 2019 — a fact already in this filing's own ws08-findings.md. (2) The filing-rail verdict "Medicare mostly doesn't pay for this" is right about custodial care at statutory tier (42 U.S.C. §1395y(a)(9)) and wrong as stated about paid elder care: CMS NHE 2024 puts Medicare as the largest single payer of home health care (32.9%, ahead of Medicaid's 22.6%) and at 21.5% of nursing-facility care; MA covers 54% of beneficiaries and may buy non-primarily-health-related supports for chronically ill enrollees under 42 U.S.C. §1395w-22(a)(3)(D); CMS's GUIDE model has paid Medicare dollars for caregiver respite nationwide since 2024-07-01. "Medicare Advantage" and "GUIDE model" each returned zero hits across the filing and record. Verdict narrowed rather than defended. Full case, including where the steelman lost: steelman-log.md |
| 36 | 2026-08-10 | Phase 2 found that Pub. L. 119-21 §71121 partially enacts this filing's leading architecture — the same public law the whitepaper cites by section number on the public page for the §71111 staffing moratorium. In a subchapter titled "Expanding Access to Care," §71121 adds §1915(c)(11): from 2028-07-01 the Secretary may approve standalone waivers covering HCBS for people who do not meet the institutional level-of-care test, on state needs-based criteria, with $50M (FY2026) + $100M (FY2027) appropriated | The source atlas fetched two sections of this act — §70422 (LIHTC, for housing) and §71111 — and stopped. Nothing in the record engages any other provision of the largest Medicaid financing change in a decade; "work requirement" and "provider tax" return zero hits | Part 4 now carries the enacted de-capping, and the three constraints Congress attached with it: §71121(B)(v) requires the state to attest per-capita spending will not exceed institutional per-capita spending — the same cost-neutrality condition §1396n(c)(2)(D) has imposed since 1981, which is the reason for the cap Part 4 called "backwards"; §71121(B)(ii) requires that the new waiver not "result in a material increase" in existing waiting times, writing queue competition between HCBS populations into statute; and §71121(C) bars the appropriated money from funding "health insurance, skills training, and other benefits customary for employees" for Medicaid-dependent practitioner classes — a direct statutory constraint on §12's workforce-first sequencing prior, in the same section that expands the architecture |
| 37 | 2026-08-10 | Phase 2 found the same reading failure three times: a source this filing cites contains a second finding the filing did not carry across. (a) Ne'eman, Health Serv Res 61(2) 2026 — the paper behind the board's only positive causal cell — reports no significant effect for Community First Choice (1.51%, CI −12.77% to 15.79%) beside BIP's +13.24%; only the BIP arm was published. (b) McGarry & Grabowski, JAGS 71(10) 2023 — the source of "$0.74" — is titled Is there a "woodwork" effect? and answers no, reporting that figure as an offset and finding more older adults served at lower cost per beneficiary; only the qualifying half was published, and "woodwork" appeared nowhere in the filing. (c) §71121, above. Phase 2 also caught two Phase 1 residuals: ws11-scorecard.md's "what this scorecard cannot yet claim" bullet still carried the corrected-everywhere-else ">7M patients / ~50% higher antipsychotic use" PE figures, and the sources page's §7 digest heading still asserted that two GAO reports "confirm the same undercounting gap" directly above a body paragraph withdrawing exactly that claim |
Not suppression — in each case the source is correctly cited and the omitted half is one paragraph away in the same document. The failure is a reading discipline: the half that fit the frame was carried across and the half that complicated it was not | ws13-rebalancing-evidence.md and ws11-scorecard.md now carry both halves of (a) and (b); both public pages carry the CFC null, the woodwork answer and §71121. Both Phase 1 residuals corrected. Owed, not done here: the homepage (site/index.html) and 404 page carry the old unqualified verdict wording in their GBMT-7 hooks; they were left alone because parallel sessions hold live branches against site chrome, and they should be updated the next time that file is touched |