Parallel rollup audit — September 25, 2026

The audits found a new Table III omission and confirmed the private-law state and compiled-hearing date defects. The remaining outputs have different levels of proof: some qualify at their frozen pins, some have only publication or conservation checks, and others remain partly audited. The campaign is open.

The repair proposal maps these findings to concrete changes, owners, reprocessing rules and acceptance evidence. Its changes are proposed; this audit's dispositions remain unchanged.

Three independent auditors ran at medium effort, covering congressional people and votes, congressional documents, and external sources. The parent agent audited regulatory outputs. Each workstream followed the semi-formal code review skill, read native inputs and public outputs, and retained scripts and receipts. Cross-reviews checked the congressional findings, external-source claims and regulatory replay methods. This audit changed documentation and retained local evidence; it did not repair or publish production data.

Scope and results

The audit froze the public index at 17:13:48 UTC, code 03541c7d917b3ea5ed2eb88071590c13b45e34f6. Its SHA-256 is 0f7b0dc06a1f59d131de15d5ca679a4037c5dac1593712057e77aa2b9caa7301. The normalized results preserve every workstream disposition and the hashes of the underlying result files.

Disposition	Audited output keys	Meaning
Qualified with limits	34	The stated source, transformation or carried-forward scope passes; broader completeness is not implied.
Partial	27	Public evidence and some source checks pass; specified source or semantic proof remains missing.
Verified publication	3	Comments, their index and the agency partition pattern pass publication/coverage checks; wider native-source qualification remains open.
Verified empty, uncomputed	6	Empty bill model/backfill outputs are unchanged. Processing and coverage remain unfinished.
Failed	3	The frozen output reproduces a specific defect described below.

These are distinct audited output keys, including one agency partition pattern. They are not producer counts, ledger-row counts or a campaign completion percentage. Unchanged FEC and court bulk families retain their earlier audits; this pass did not repeat their full source qualification.

At the 17:36:54 UTC end check, the base object ETags and rulemaking pointer were unchanged. CRS alone had advanced from the audited 35a43885… to 7076f282…; that newer generation remains unaudited. The ledger retains the qualified frozen pin. See the end check.

Confirmed defects

Output	Evidence	Required correction
`table3_records` at `c9d35661…`	Act 119-37 contains 110 native rows but only 106 published rows. Four meaningful code-reference rows lack an act-section label and are filtered out. Both retained and fresh HTML reproduce the omission.	SpicyDocs must preserve rows with missing act-section labels. SpicyRegs must retain their native absence and verify source-row conservation before publishing.
`laws` at `c9d35661…`	Private laws 119-1 and 119-2 still say `uslm_outcome=not_requested`, although retained native USLM and the acquisition log prove captured-but-refused reads.	Preserve captured bytes, available title/date/citation fields and a truthful partial/refused outcome. Do not invent the missing Statutes at Large citation.
`hearing_bill_links` at `b188e1b3…`	`CHRG-117shrg56721` links to `117-s-2792` with one `held_date`; native MODS lists eleven dates.	Preserve all dates and populate a scalar only when unique; invalidate affected read checkpoints and reprocess.

The omitted Table III references are 7 USC 2254 (139 Stat. 511), 42 USC 1769g and 1758 (139 Stat. 534), and 2 USC 60a note (139 Stat. 563, Elim.). The traversal repair did publish a larger population; the old early-stop diagnosis no longer describes this generation. The current defect is row loss within a successfully read page.

Today's attempted private-law XML requests returned HTTP 200 HTML, not USLM. Independent review caught and corrected the initial interpretation. The finding rests on separately parsed, digest-verified prior USLM captures and the captured-but-refused run log. HTTP status alone did not qualify those new reads.

The congressional document report traces each defect through source parsing and publication and records the inspected tests. No production test suite was run for this documentation-only reconciliation.

What the passing checks establish

Regulatory transformations: independent full replays reproduce feed summary, agency statistics, monthly volume, discovery signals, docket search, organization/committee links and Federal Register docket links. These prove the stated computations over exact parents. Organization names remain heuristic matches, not verified identities or money flows.
Regulatory sources: every previously qualified Federal Register row is unchanged; all added September 24 issue rows match native JSON. CFR and the retained Agenda edition carry exactly identical table bytes, preserving their earlier qualification limits.
Congressional people and votes: members/terms and committee assignments preserve their substantive populations. Added Senate votes and positions match official XML; the complete member-term join matches an independent replay. Meetings, record issues, press, nominations and treaties qualify at their recorded scope. The community crosswalk, party-history, stale vote-action index and carried-source freshness limits remain explicit.
External sources: FCC proceedings and filings, the frozen CRS generation, GAO and lobbying qualify through complete carried-row conservation plus the documented native checks of new or changed records. This does not establish full history or eliminate later source restamps.
Congressional documents: law_code_sections, report metadata and bodies, hearing transcripts and read checkpoints qualify at their stated scope. All six previously refused hearing detail reads now have retained valid responses. Report sections remain partial because the changed heading/body decomposition was not independently reconstructed exactly.

Workstream reports retain the exact counts, pins, cell comparisons, samples and limitations: regulatory, people and votes, documents, and external sources.

Remaining work, in execution order

Repair the three reproduced defects. Preserve the broader public populations, add meaningful regression checks from the retained native examples, then independently replay, publish and read back each affected family. The source owner is SpicyDocs; SpicyRegs owns output/state handling.
Finish the partial congressional audits. Check new bill bodies, sections, differences and metadata beyond the sampled native cohort; older-Congress subject additions; new print documents/actions/citation spans; exact changed report-section decomposition; and the remaining communications detail fills. Amendments need capture-time source evidence to resolve later source drift. Empty models and backfills still require actual computation and acquisition.
Retain missing source evidence for external refreshes. SAM's scheduled 2002 selection succeeded and preserved the qualified 2026 population, but its complete new extract was not retained. USAspending mixes observation times without per-row dates. Court dockets have later native field drift without capture-time evidence. Keep these current generations partial; retain raw responses in ordinary refreshes so later audits can reproduce them.
Complete regulatory semantic and native-source checks. The new dockets and documents pass bounded native samples, not a whole-population audit. Rulemaking passes all recorded digest/reference checks, but retirement and identity semantics require a complete explanation against decisions 32–33. Removed no-action shells may legitimately have no successor; their absence alone is not a defect. Wider comments source repair and the dated partition layout remain open.
Qualify operations and consumers separately. Local comments publication and dependent refreshes succeeded. Hosted export capacity, the resumed incremental sweep, hosted MCP and representative cross-source joins remain separate gates in the execution plan. This audit does not establish their completion.

The output ledger owns qualified pins and delivery state. The execution plan owns the backlog. The receipt directory owns detailed audit measurements and exact source bytes.

Independent reconciliation review found no material overclaim in these status changes. The ledger pin check reports OK=39, NO-PIN=7, DRIFT=12, NOT-LIVE=0, MALFORMED=0. Its nonzero exit reflects the explicitly pending generations. These ledger-row checks compare pins or base ETags; an OK result does not mean the corresponding source audit is complete.