Skip to content

census: transport 33 unexamined candidates, MC-001 re-pinned, census film re-rendered - #41

Merged
Cubits11 merged 3 commits into
mainfrom
claude/census-transport-2026-11-02
Nov 3, 2026
Merged

Cubits11 merged 3 commits into
mainfrom
claude/census-transport-2026-11-02

Conversation

@Cubits11

@Cubits11 Cubits11 commented Nov 3, 2026

Copy link
Copy Markdown
Owner

The census carries 33 more unexamined candidates, MC-001 is re-pinned to the new census bytes, and the census film shows the new count. This transports candidates; it does not verify them. No candidate is examined, and no classified row, count or criterion changes.

Commits

  • 5797caf census, registry, film (31 files):
    • 33 entries appended under unexamined_candidates. They come from the weekly sweeps, discovered 2026-09-21 to 2026-11-02. +132 lines, 0 deleted.
    • MC-001 re-pinned to the new census.yaml sha256, with last_reviewed set to 2026-11-02.
    • The census film follows the count: films/the-missing-column-census/manifest.yaml 25 → 58, its limit line in films/curation.json, and its master and vertical renders, as in 8eb6997.
    • Every page generated from the registries is regenerated.
  • 043d70a distribution (6 files): drafts and events re-bound to that commit. draft-history.json gains 4 revisions, as in registry: MC-002, MC-003 and MC-004 owner re-review 2026-11-02, evidence unchanged #40. Every event binding is committed.
  • 5584b48 sitemap: lastmod follows the commits above.

Owner attestation, 2026-11-02, for MC-001: "the census counts on this branch are 20/14/5 with 58 unexamined."

Verification on this branch

  • python3 scripts/verify_census.py: N/M/K 20/14/5, M ladder 14/12/0; 20 examined, 9 excluded, 58 unexamined (25 on main + 33 here). Exit 0.
  • python3 scripts/verification_manifest.py: ok deterministic verification manifest: 90 unique checks, exit 0. Each of the 90 checks was also run individually: 90 exit 0, none exit 2.
  • python3 scripts/films/verify_films.py: exit 0. The census film's manifest is bound to 18 facts, both receipts are fresh, there are 0 overflows, and the 14 px phone text audit passes.

Render variance, and the acceptance rule used

  • At value 25, before any change, the master renderer was observed not to be run-to-run deterministic. One of two renders differed from the committed render in 246 of 1201 frames, at sub-visible magnitude (worst frame PSNR 77.9 dB). The other render was byte-identical to the committed one.
  • verify_films.py's five-frame determinism check passed on the deviating render. It cannot detect this variance class.
  • The renders here were accepted under a stricter rule, fixed before any 58 render was inspected:
    • Master: two consecutive runs with an identical frames_digest, and a captured-frame diff against the committed 25 render confined to the count-bearing frames (937–1200, from 31.2 s) inside one digit slot (59 × 41 px). At most 4 runs. Accepted on runs 1 and 2.
    • Vertical: two runs with an identical digest. Accepted.
  • The comparison used captured PNG frames, not decoded video, because the encoder's 250-frame GOPs spread any change through the rest of a group.
  • The owner inspected the changed stills before the commits.

What this does not change

  • The publication gate. scripts/publication_gate.py still reports the same three blockers: title sources, the Social Square phone composition, and 0 of 5 reader records. No gate code, criteria, reader records, title sources or Social Square files change.
  • Deployment stays blocked.
  • This re-render lands before any reader-test candidate is bound, so no bound bytes move.

Order

#40 merged first. This is the second PR. The cadence-rows PR follows after this merges. Do not merge until CI is green and the owner has reviewed.

…film re-rendered

Transport, not verification: 33 entries appended under
unexamined_candidates (weekly sweeps discovered 2026-09-21 to 2026-11-02).
No candidate is examined and no classified content changes.
verify_census.py recomputes N/M/K 20/14/5 and M ladder 14/12/0; the
unexamined count moves from 25 to 58.

Owner attestation, 2026-11-02, for MC-001: the census counts on this
branch are 20/14/5 with 58 unexamined. MC-001 is re-pinned to the new
census sha256 and its last_reviewed moves to 2026-11-02.

the-missing-column-census reads MC-001.unexamined, so its manifest
(25 -> 58), its limit line in films/curation.json, and its master and
vertical renders follow, as in 8eb6997. The renders were accepted under a
stricter rule than verify_films.py applies: two consecutive runs with an
identical frames_digest, and a captured-frame diff against the committed
25 render confined to the count-bearing frames (937-1200, from 31.2 s)
inside one 59 x 41 px digit slot. At value 25 the master renderer was
observed not to be run-to-run deterministic: one of two runs differed in
246 of 1201 frames at sub-visible magnitude, and the other was
byte-identical. verify_films.py's five-frame check does not detect that
class. The owner inspected the changed stills before this commit.

No publication-gate code, criteria, reader records, title sources or
Social Square files change.
@Cubits11
Cubits11 merged commit 7e5a667 into main Nov 3, 2026
6 checks passed
@Cubits11
Cubits11 deleted the claude/census-transport-2026-11-02 branch November 3, 2026 04:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant