Disclosures / Postmortem Log

Seven regulatory citations wrong in proven artifacts

Discovered 2026-07-31 · Fixed 2026-07-31 · 6 nodes re-proven

What happened

A parity-checker triage pass found 7 regulatory citations that were wrong in kernel source, in most cases already shipping inside a cryptographically proven (groth16) artifact. Examples: an ECOA adverse-action notice duty cited to the administrative-enforcement section (15 USC §1691c) instead of the actual duty section (§1691(d)); an OECD Transfer Pricing Guidelines cash-pooling reference cited to the cost-contribution-arrangements chapter (Ch. VIII) instead of the correct chapter (Ch. X); a FinCEN CDD Rule applicability date invented as "July 2018" where no such date exists in the primary source; and one node (art-240) where the citation was wrong on both the page and the kernel simultaneously, so a page↔kernel agreement check could not have caught it.

Affected artifacts

art-59, art-227, art-228, art-240, art-260, art-268: 6 nodes, 7 citation fields.

Fix

Each citation was re-verified against its primary source before editing (not taken on the triage table's word). Every fix was additive; no citation was stripped to force two surfaces into agreement. All 6 affected nodes were re-proven on GPU (groth16-bn254): kernel digest, proof journal, and seal moved together, and 15 golden fixtures were re-pinned. The one node with no correct citation on either surface (art-240) was resolved from the regulator's own published record rather than invented.

Five citation defects wrong on both page and kernel simultaneously

Discovered 2026-07-31 · Fixed 2026-07-31 · 3 kernels re-proven

What happened

A follow-up source-verification pass found defects that a page↔kernel agreement checker structurally cannot detect: the same fact hand-authored identically wrong on both surfaces. Included a CFR subsection quoted with the wrong letter designation, a CBAM carbon-tariff calculator applying the wrong percentage ramp to fertiliser inputs (it was applying the cement/iron/steel/aluminium/hydrogen ramp instead of the flat rate that applies to fertilisers, so a live calculator produced a wrong number, not just a wrong citation string), and a reference version string naming an EU instrument that does not exist.

Affected artifacts

art-231, art-70, plus 3 further checker-invisible fields across the same batch.

Fix

Each defect re-verified against its primary source (Cornell LII text, the actual Commission Implementing Regulation) before editing. The CBAM calculator was made sector-aware in both the kernel and the live page calculator, so the badge, schedule, and advice copy all follow the corrected ramp. Affected kernels re-proven with matching seals.

Artifact asserted an inapplicable regulatory regime

Discovered 2026-08-02 · Fixed 2026-08-02 · 1 node re-proven

What happened

Node art-452 cited SR 11-7 (model risk management) as its regulatory basis. SR 26-2 excludes deterministic, non-statistical tools from the definition of "model" entirely, so a tool that falls outside that definition cannot be described as SR 11-7 compliant; the citation asserted a regime that did not apply to the tool it was attached to.

Affected artifacts

art-452, regulatory_basis field (this field is part of the hashed execution_hash preimage on this node, so the correction moved the artifact's hash).

Fix

Citation corrected from SR 11-7 to SR 26-2. The node was re-proven with a fresh seal; image_id, journal.kernel_digest, journal output, and seal moved together. A sibling node was independently diagnostic-run afterward and returned the same maturity score as before the change.

Entries are drawn from this project's own commit and build-record history (search terms: citation-correction commits, kernel re-prove records) and are added as they are found and confirmed fixed. This is not an exhaustive list of every development-time defect, only defects that reached a shipped, receipted artifact. A defect caught in review before shipping is not logged here. This page will grow as the correctness-verification work that produces these findings continues; it carries no commitment about how often it is updated.