# Triage of the 46-item GAN-culture catalog (2026-08-31)

Michael pasted the 2026-08-26 "GAN culture & bitGAN lore" catalog (A1-A20, B1-B26) and
asked for a full evaluation + plan, run down autonomously. The catalog predates
D1-D28, so a lot of it is already settled. This file is the item-by-item verdict
against the current DECISIONS log, registry (~103k keys), and on-disk artifacts.

Verdicts: **DEAD** (killed by a ruling or sweep since 8/26), **DONE** (the proposed
test already ran), **OPEN** (unworked, worth running), **PARTIAL** (some of it ran).

## Dead or already closed — no work
- **A6 re-save sessions** — DEAD. D28 ran it: piece-word reads (essence/piece#/mint#/
  name x 4 orderings) AND date-number reads all negative, plant-validated. This kills
  the catalog's two "untested" session word-sets too (they are the essence reads).
- **A10 mislabel eleven** — DEAD. D4: 9 of 11 labels actually match the art; the
  selection is closed. (Residue kept: #357 and #423 are real mislabels, both already
  worked into anchors/variants.)
- **B1 words-in-clue** — DEAD. Handoff: "no marking exists"; checked at page level.
- **B7** — folds into A6, dead. **B15** — retired; mint-order authority only.
- **A13 / A15 / A20** — fences, already honored.
- **Glyphs/letters as a selection** — D9 closed. But the painted-mark CENSUS finished
  after the catalog was written: all 512 scanned, class closed at 4 confirmed carriers
  (#165 "81", #179 "10.00?", #207 "4", #379 "AI") + #414 unsure. The catalog's hope
  that painted marks might supply members is now a settled small class (H-5 pointers).

## Done since (or before) the catalog
- **B23 case-anomaly census** — DONE 8/27 (`case_censor_results_2026-08-27.md`), on
  exact-case chain names. Result: small deliberate classes (suffix-Gan pair {175,205},
  Primoridal pair {16,100}, spooky double-space set), NOT an eleven. Residue open:
  apply the acrostic lens to the OTHER published lists (credits, solver lists).
- **B9 spooky micro-decision** — settled by D1: run {205,206,207} as variants.
- **A17 chain mismatches** — PARTIAL: verify_tags machine-checks the chain families;
  the substantive-residue taxonomy (which mismatches touch anomaly carriers) is a
  small open census.
- **B13 ternary fence** — near-done (Level column verified); one 10-min string scan
  confirms and closes.

## Open — locally computable (Wave 1, running now via subagents)
1. **A5 + B6 position-32 probe** — piece at position 32 of each drop, anomaly status;
   plus position-in-drop tabulation for all ratified glitch classes.
2. **B5 HODL partner rules** — the four cheap partner rules (same-number+offset,
   same position-in-drop, forge lineage, adjacency); does any close 1:1?
3. **B12 review pointer table** — each Review anomaly-carrier's three cheap pointers;
   score how many land on marked pieces.
4. **A19 forge inheritance violations** — verify the eye-addition rule on all 16
   forges; violations = candidate markers; the unminted pair as authored absence.
5. **B14 cross-contract census + A9 authored absences** — count pre-clue anomalies
   outside the 512 (crispy, #0, Noun GAN Study burn, reneGAN 0110 burn) and the
   authored-absence inventory (D26 already explained the Arweave holes: 339/512
   coverage — archive rot, not authored).
6. **A1 perfect-fakes census** — byte-touched (re-saved) but visually clean pieces;
   honest count vs 11.
7. **A8 + B26 event counting** — glitch EVENTS (pairs collapse) under provenance
   filters (intentional/natural/both); any (filter x definition) landing on 11.
8. **A2 walk-shape taxonomy** — all 512 classified {plain, reversed, palindromic,
   splice-revisit, held, frozen} from frame bytes; count non-plain walkers;
   distinguished-index list feeds B11.
9. **A4 stillness census** — bottom-30 movers, convention-violating stillness.
10. **B16 field-by-field depth census** — the Solstice template (one field, three
    hiding depths, 11 words) tabulated across every Act 1 field. Includes B21
    (distribution-anomaly depth).
11. **B18 odd-file bespoke reads** — #134, #228, #300, #390, #0, #446, #188 each
    examined alone for internal 11/12-structure; plus A18 disposal/palette deviants.
12. **B11 distinguished-index sequence** — all distinguished frame indices as ONE
    mint-ordered sequence; what is it? (Diffed against numcode/* — those swept
    Arweave-code constructions, not this; D28 killed per-piece geometry on fib11
    only.)
13. **Lore censuses** — A16 tributes, B8 crypto-roles, B24 count-the-counts,
    A11 architecture names, A14 SuperRare numbers, B2 thrice-stated + Fate-function,
    B10 trio-positions-as-date.
14. **A3 + B3 registry audits** — which historical NOHITs never got their 128-ending
    sweep; whether passphrase variants ever ran (they did not — harness supports
    `seed(mn, pw)`); enumerate the strong dead sets for `universe`-as-word-12 and
    `universe`-as-passphrase re-tests. NOTE lore fact D-log: Michael believes #0
    was published AFTER the puzzle launched — verify on-chain before leaning on it.
15. **B17 index transplants** — diff against multistep_sets.json first; the
    between-orderings transplant (index in list A -> entity in list B) may be novel.

## Open — external / visual (Wave 2)
16. **B20 gallery archaeology** ★ — web.archive.org captures of CryptoVoxels
    bitGALLERY / META Gallery / SomniumSpace near 2022-02-05; which pieces hung, in
    what order. The one precedent where 11 pieces' display order WAS a key.
17. **B25 journal-image audit** — inventory every img in the 109 pages' raw HTML;
    inspect Act-1-window images (the clue post above all) for barcodes/codes.
18. **A7 near-duplicate pairs** — visemb/framevecs cross-piece ranking, top ~30
    pairs for Michael's eye.
19. **B19 tile-vs-tile duplicate scan** — within published composites (montage
    SELECTION is dead; a duplicate-tile scan is new evidence, cheap).
20. **B22 eyes audit prep** — reconcile glitch_liars.json artifacts, build the
    all-frames review page for Michael (his eye required; also unblocks H-2
    eyes-as-binary).
21. **B4 Named-relations table** — 12 named pieces x {forge, burn date, gallery
    position, first sale, owner, montage, re-save}; hunt the column where #39
    "fits in".

## Standing protocol adopted from the catalog
- Every candidate eleven gets its full 128-checksum ending sweep (A3), plus the two
  cheap B3 variants: append `universe` as word 12; `universe` as passphrase.
- A12 dream-rule tie-breaker: prefer pattern-property reads over read-the-words
  candidates (consistent with every ruling to date).

## Sweep authorization note
Michael's 8/31 instruction ("work autonomously and continuously, run absolutely
everything down") + the D27 precedent re-opens local, free, plant-validated sweeps.
Nothing with external cost runs without his go-ahead. Every sweep is labeled
selection x register x ordering and logged to the registry.