Found in the adversarial review of #120.
'Card and graph totals reflect the published Mech pages checked on September 24' is false for CellStructureMech and TraitMech
Where: _fleet/mechs_template.md:18
The card paragraph says the totals reflect the published pages checked on September 24, 2026, and that each Browse link is the source for its card. For two Mechs the cards keep pinned figures that differ from what those pages showed at check time: CellStructureMech 542 against 544 (545 now) and TraitMech 763 against 764. The drift is disclosed in the PR body, site_audit.json and _fleet/README.md, but not on the page. The same figures are repeated in index.md and resources.md.
Failure scenario. A reader clicks Browse on the CellStructureMech card and sees 545 structure records where the card says 542, even though the page said the card reflects that site as checked on the stated date.
Evidence. python3 scripts/fleet/check_cards.py → 'DRIFTED CellStructureMech card 542, site 545' and 'DRIFTED TraitMech card 763, site 764'. site_audit.json has site_figure_at_check 544 for CellStructureMech and 764 for TraitMech, with notes that the card keeps the pinned figure.
Verifier (refuted, low). The drift numbers reproduce, but the sentence the finding calls false is literally true, and the behaviour is intended and documented. The page says the totals "reflect the published Mech pages checked on September 24, 2026". The GitHub Pages deployment history shows that at the 02:06Z pin (19:06 PDT on Sept 24) the deployed CellStructureMech page (42cc23b, deployed 02:03Z) read 542 and the deployed TraitMech page (a21c5aa) read 763. Those are exactly the card figures. The 544 and 764 in site_audit.json come from a second check about 85 minutes later on the same local day, after new records had been deployed. A later same-day check that saw newer pages does not make the statement false.
The failure scenario, a reader clicking Browse and seeing a bigger number than the card, is what any dated snapshot of these fast-moving corpora produces. Within days it will apply to most Mechs, not just these two. The page is explicit that its figures come from a stated check date. Keeping the pinned figures rather than mixing revisions is the documented design of the refresh, and the PR body, _fleet/README.md and site_audit.json all disclose it.
The same totals in index.md and resources.md carry no date, but that is true of every figure in them and is not specific to this drift.
Unrelated, but seen while verifying: both site_audit.json notes say "The live page is byte-identical to pages/index.html at the pin". The site_html_sha256 recorded next to each note is actually the hash of the later post-pin page, the 544 and 764 versions (see reproduction steps 2 and 4). The note and its hash therefore describe two different page versions. This is a provenance inconsistency in the audit file, not the defect this finding claims.
Suggested fix. Reword to say the totals come from each Mech's pinned revision (published pages checked September 24), or add one sentence naming the two Mechs whose sites have since moved on.
Found in the adversarial review of #120.
'Card and graph totals reflect the published Mech pages checked on September 24' is false for CellStructureMech and TraitMech
Where:
_fleet/mechs_template.md:18The card paragraph says the totals reflect the published pages checked on September 24, 2026, and that each Browse link is the source for its card. For two Mechs the cards keep pinned figures that differ from what those pages showed at check time: CellStructureMech 542 against 544 (545 now) and TraitMech 763 against 764. The drift is disclosed in the PR body, site_audit.json and _fleet/README.md, but not on the page. The same figures are repeated in index.md and resources.md.
Failure scenario. A reader clicks Browse on the CellStructureMech card and sees 545 structure records where the card says 542, even though the page said the card reflects that site as checked on the stated date.
Evidence. python3 scripts/fleet/check_cards.py → 'DRIFTED CellStructureMech card 542, site 545' and 'DRIFTED TraitMech card 763, site 764'. site_audit.json has site_figure_at_check 544 for CellStructureMech and 764 for TraitMech, with notes that the card keeps the pinned figure.
Verifier (refuted, low). The drift numbers reproduce, but the sentence the finding calls false is literally true, and the behaviour is intended and documented. The page says the totals "reflect the published Mech pages checked on September 24, 2026". The GitHub Pages deployment history shows that at the 02:06Z pin (19:06 PDT on Sept 24) the deployed CellStructureMech page (42cc23b, deployed 02:03Z) read 542 and the deployed TraitMech page (a21c5aa) read 763. Those are exactly the card figures. The 544 and 764 in site_audit.json come from a second check about 85 minutes later on the same local day, after new records had been deployed. A later same-day check that saw newer pages does not make the statement false.
The failure scenario, a reader clicking Browse and seeing a bigger number than the card, is what any dated snapshot of these fast-moving corpora produces. Within days it will apply to most Mechs, not just these two. The page is explicit that its figures come from a stated check date. Keeping the pinned figures rather than mixing revisions is the documented design of the refresh, and the PR body, _fleet/README.md and site_audit.json all disclose it.
The same totals in index.md and resources.md carry no date, but that is true of every figure in them and is not specific to this drift.
Unrelated, but seen while verifying: both site_audit.json notes say "The live page is byte-identical to pages/index.html at the pin". The site_html_sha256 recorded next to each note is actually the hash of the later post-pin page, the 544 and 764 versions (see reproduction steps 2 and 4). The note and its hash therefore describe two different page versions. This is a provenance inconsistency in the audit file, not the defect this finding claims.
Suggested fix. Reword to say the totals come from each Mech's pinned revision (published pages checked September 24), or add one sentence naming the two Mechs whose sites have since moved on.