The nightly card check stays red for Mechs that publish after a refresh's pins

[realmarcin]

From the review of #109, which adds the check but deliberately does not refresh the figures.

Live run on 2026-09-22:

  DRIFTED  CommunityMech        card 392, site 405
  DRIFTED  CultureMech          card 6,286, site 6,288
  DRIFTED  MediaIngredientMech  card 2,951, site 2,952
  DRIFTED  TraitMech            card 723, site 751

The check exits 1 on drift, so the scheduled workflow will fail nightly until the cards are refreshed. #109's own rationale for keeping fetch failures out of the red signal — a standing red teaches people to ignore it — applies here too.

The refresh belongs with the full pipeline rerun owed for #84 and #87, where the card figures in _fleet/mechs_template.md and the MECHS block in _fleet/fleet_fragment.html are updated together. This issue tracks that the nightly is expected red until then.

[realmarcin]

Rerun at 2026-09-22T07:28Z, before #109 merged: CellStructureMech has moved too, card 421 vs site 422. Five of ten cards now drift. That it changed within an hour of the first run is the case for keeping the check off pull_request.

[realmarcin]

Update from #120. The four figures this issue listed are refreshed there, but the condition it tracks still holds after that PR: the scheduled card check stays red. CellStructureMech and TraitMech published new records after the refresh's pins, and the page is kept a consistent snapshot at those pins by design. At the last check the sites showed 547 and 766 against cards of 542 and 763. Keeping this open. #148 asks for a check that can tell growth past a pinned card from a wrong card, which is what would let this close for good.

[realmarcin]

Status on 2026-09-25, with #120 still unmerged. After #120's refresh, two cards drift because their Mechs keep publishing: CellStructureMech card 542 against 575 on its site, and TraitMech 763 against 768. The page is kept as one snapshot at the refresh's pins by design, so this recurs after every refresh of a fast-moving Mech; #148 asks for a check that can tell growth past a pinned card from a wrong card, which is what would let this close. Separately, CultureMech currently reads as unreadable because its pages/ index is gone again; that is #175, which #120 fixes by reading the committed README instead. Retitled from 'Four card figures have drifted...' since the condition, not the original four figures, is what stays open.