# Read me first

This bundle is everything we know about the bitGAN Act 1 puzzle, in files you can download.
It mixes three very different kinds of thing, and the Data page now groups them:

1. **Pindar's own sources, verbatim.** His properties sheet, his Arweave files, the chain
   records, his journal pages, and the January 2022 working table with his own word-index
   column. Start here. Everything else is derived from these.
2. **Facts recomputed from those sources.** Hashes, frame counts, save times, transfers, number
   classes. Anyone can re-run `primary_facts.py` and get the same numbers.
3. **Our reading, and our attempts.** The anomaly ledger, the taxonomy, the collector's eye, and
   every recipe we have tried. This is interpretation. It has been corrected more than once.

## The one rule: nothing is closed (D70, 17 September 2026)

Older files in this bundle say things like *dead*, *closed*, *ruled out*, *do not reopen*. Read
every one of those as a note about **one recipe on one day**, never as a verdict on a glitch, a
word rule or an ordering. Two reasons:

- A sweep tests one exact combination — which pieces, how each becomes a word, what order. A
  miss says that combination missed. It says nothing about the pieces.
- Several sweeps ran on data we have since corrected. The description-anomaly set went from 7
  to 15. Two essences turned out to be transposed, not out of sequence. Two hidden-type
  letters turned out to be chain errors Pindar had fixed in his own sheet. A miss recorded
  against the old data may simply have been the wrong words.

`tested_sets_no_hit.txt` and `coverage.json` are coverage records so you can see where we have
looked. They are not a list of closed doors. The site's recipe builder shows every option open
and marks a past attempt with its date only.

## Statements in older files that later data overturned

| where | it says | what we now know |
|---|---|---|
| AUDIT-BRIEF (claim 3) | the only unmatched Arweave GIFs are two studies of piece #0 | they are NounGAN studies in a second drive; 64 NounGAN files match none of the 512 |
| AUDIT-BRIEF (claim 30) | no earlier revision of the sheet exists on the drive | true of the drive; an earlier table with four extra columns sits in the 1 Jan 2022 capture of journal 015, reached only through #272's on-chain link — see FINDINGS-2026-09-16 |
| AUDIT-BRIEF (claim 14), older tags | seven pieces carry a non-boilerplate description | fifteen; eight are defects in the boilerplate itself (#272, #301–#306, #400) |
| older tags, methods.json, pb_content.json | the ten plain colour words are each an essence once | still true, but rose (#446) and bronze (#493) are colour words on the list too; the tag was widened |
| older tags, glitch_dictionary | #21, #57, #259, #268 are "essences out of sequence" | #21 and #57 are; #259 and #268 are a transposed pair — Pindar's index column runs in order through them and only the words swapped |
| hiddentype_primer, older tags | five hidden-type misfiles, one class | two (#32, #119) are chain-side errors he corrected in his own sheet; three (#161, #187, #193) appear in both sources and were carried deliberately |
| methods.json | 27 selections with twenty different free-text statuses | statuses are now exactly ratified / candidate / dead / superseded; the live list is /api/principles |
| chain_mismatches.json | "the 20 sheet-vs-chain differences" | a raw 13 Sep snapshot including title-counter noise; the ruled set is 26 pieces (29 counting the Twin/Twins rename) |
| any file before 16 Sep | #251 could not be served | our own storage was missing the file; it is restored and byte-identical to Arweave |
| HANDOFF-2026-08-28 "closed / dead" | do not reopen without new evidence | superseded by D70 above; the January table is new evidence for most of them anyway |

## Where the current truth lives

- Pieces, tags and evidence: `/api/pieces`, `/api/tags`, `/api/export.csv` (live).
- Selections, word rules, orderings: `/api/principles` (live); the space swept so far: `coverage.json`.
- Rulings and their dates: `DECISIONS.md`. The date on a ruling matters more than its number.
- The January table and what it settles: `FINDINGS-2026-09-16.md`, `page015_jan1/`.