# SOLUTION SPACE — the three exhaustive catalogs + guardrails
Built 2026-08-27 from: glitch_type_catalog.md, confirmed_glitch_registry.md, word_sources.md,
extraction_theories_A/B.md, gan_lore_expanded.md, drop_registers.md, case_censor / drop_thematist /
template_mechanic lane reports, deck_512.json + deck_derived.json, CYCLE-SUMMARY-2026-08-27.md,
owner_facts_2026-08-24.md. This is a working reference: a theory = one row of S1 × one row of S2 ×
one row of S3, checked against S4.

---

## S0. THE TRIPLE DISCIPLINE (read first)

**"Dead" always means a specific TRIPLE died: (glitch-selection × word-derivation-register ×
ordering). It never means the glitch itself is dead.**

Why this matters mechanically: `_registry.json` keys are **sorted word-SETS** — they record that a
bag of words was swept, which collapses the selection and the register into one hash and loses the
ordering entirely (run labels carry the ordering, and only per-run). A NOHIT on
(union-eleven × essence × mint) says nothing about (union-eleven × wl[frame-index] × mint) or
(union-eleven × essence × upload-transplant). The essence register has now died on virtually every
principled selection — that is evidence against the ESSENCE COLUMN of the S2 catalog, not against
any selection in S1.

**Standing rule henceforth:** every sweep label must encode `selection-id × register-id ×
ordering-id` (use the S1/S2/S3 row IDs below). A negative kills exactly that triple. Also per lore
A3: a candidate killed without its full 128-checksum-ending sweep is NOT dead; audit provenance
before trusting any historical NOHIT.

---

## S1. GLITCH CATALOG — everything that might constitute "a glitch"

Legend: **CONF** = byte/structural-confirmed · **EYE** = the collector's-eye confirmed · **cand** =
candidate. Member lists live in glitch_type_catalog.md / glitch_types.json unless inlined.

### S1-V · VISUAL (in the art; OpenSea-visible — Tier 1)

| ID | Type | Def (one line) | Members / pointer | Status |
|---|---|---|---|---|
| V1 | frame_splice (standard_glitch) | foreign/self frame spliced into the loop | 46 (catalog); byte core #7,43,67,71 + montage-frozen 19,31,37,73,244 | CONF core, EYE rest |
| V2 | inverted_glitch | spliced frame is negative/inverted | #231,256,274,479 (#256 double) | CONF |
| V3 | color_flash_sinewave | flash morphs binary bg into sine waves | #130,133 | EYE |
| V4 | color_flash_plain | whole-frame color flash, binary intact | #462,463,464,497 | EYE |
| V5 | color_speedup | animation visibly accelerates | 16 members (catalog) | EYE |
| V6 | morai_echo | faint extra Moirai heads (template #1) | 11: 11,12,43,44,45,73,78,100,117,123,454 | EYE, FUZZY count |
| V7 | sits_behind_binary | GAN behind the bg instead of in front | 18 robo members | EYE (robo convention?) |
| V8 | naga_palindrome | visual/name palindrome robos | 8: 129,141,147,159,162,174,180,192 | EYE (family behavior) |
| V9 | larger_eyes | eyes abnormally large | 15 (catalog) | EYE, FUZZY |
| V10 | unique_eyes | singular eye rendering | #147,161,187,467 | EYE |
| V11 | eye_type_mixing | eyes blend skull/robo/ghost types | #187,191,206 | EYE, FUZZY |
| V12 | costume_spooky | costume over an underlying GAN | 10: 193,195-202,204 | EYE |
| V13 | hidden_number | number painted in the art | #165 "8941", #207 "#4" | EYE (channel nothing scans) |
| V14 | rainbow_morph_prelude | rainbow morph prefiguring #512 | #501-504 | EYE |
| V15 | line_background | new horizontal-line bg | #485-488 | EYE |
| V16 | scanline_removal | pixel-row removal across the GAN | #489,491 | EYE |
| V17 | day_night_pair | Solar/Lunar positive-negative pair | #209,300 | EYE |
| V18 | beige_background | first-ever beige bg (flags timing pair) | #479,480 | EYE+CONF |
| V19 | robo_skull_mixture | robo+skull feature mix | #175 | EYE |
| V20 | twin_entangled | twin / "Entangled" quantum pieces | 7 (catalog) | EYE, FUZZY |
| V21 | merge_convergence | two GANs converge to one | #50 | EYE |
| V22 | figure_resemblance | art resembles a figure (whale/shark/invader…) | #5,89,268,302,465,466 | EYE, FUZZY |
| V23 | color_gradient_off | wrong hue for position in ghost rainbow | **#357 — ANCHOR, 6-way over-determined** | CONF+EYE |
| V24 | quantum_glitch_capstone | #512's different-in-kind glitch | **#512 — ANCHOR, only Dim-11** | EYE |
| V25 | shimmer | shimmer effect | #7 | EYE, FUZZY |
| V26 | test_written_word | "Test" painted on body+bg | 39 ghost/review (catalog) | EYE |
| V27 | spooky_ghost_eyes | only artist-GAN with spookyGAN eyes | #447 | EYE |
| V28 | wrong_creature_for_drop | rare species inside its drop (theme-breaker) | 20: skull{5,8,13,89} robo{130,131,137,145} sol{209,300} ghost{327,407} rev{465,466,479,480,497,507,511,512} | cand, unswept as class |
| V29 | type_rule_breaker | piece fails to BE/DO what its type declares | 20, all six drops (type_rule_breakers.json) | EYE class; #281 CONF (1-of-8 glitchGAN non-glitcher) |
| V30 | mislabel (label-liar) | SubType-vs-art contradiction | exactly 11: strong 179,423,231,511,509; med 204,150,490,478,92,491 (instrument validated on #14) | cand, needs the collector blind-ratification |
| V31 | eyes_lie | Eyes trait ≠ counted eyes in art | #207 (2 vs 3) + unaudited; glitch_liars.json | cand, audit not run |

### S1-N · NAME-VISIBLE (in the displayed title/name)

| ID | Type | Def | Members | Status |
|---|---|---|---|---|
| N1 | case_anomaly suffix "Gan" | miscased GANg suffix | exactly {#175, #205} in all 512 | CONF (chain exact-case) |
| N2 | Primoridal transposition | letter-swap of Primordial | exactly {#16, #100} | CONF |
| N3 | double_space | two spaces before counter | 6/16 spooky {194,202,203,204,207,208} + #174 | CONF (invisible on OpenSea — tier 2) |
| N4 | leet_names | digit-for-letter | 14 robo (convention) + cross-drop outlier #194; #191 err0r | CONF; glitch only where cross-drop |
| N5 | radix_corruption | 0x where octal demanded | #465/466 "0x20/0x21" (octal→hex revalues 16/17→32/33) | CONF |
| N6 | error/joke names | err0r, Not Found, Gltich, skulGAN, xxxxxGAN | #191, #404, #73, (skulGAN 1pc), #512 | CONF |
| N7 | formula-breaker proper names | title breaks "SubType+GANg" formula | ghost-drop 12; 31 unique-title-word pieces (word_sources §2-3) | CONF census |
| N8 | reversal | SubType/name reads backward | #48 wobniaR; god↔dog #159, flow↔wolf #180 | CONF |

### S1-M · METADATA (trait-level, not displayed prominently)

| ID | Type | Def | Members | Status |
|---|---|---|---|---|
| M1 | non_bip_essence | essence not a BIP-39 word | exactly 3: #191 err0r, #363 TRUE, #473 FALSE | CONF |
| M2 | essence_displaced | essence off the wl[GAN-1] band | #21 bean, #57 mistake (skull); band decays in later drops | CONF |
| M3 | chain_corrupt_side | on-chain attr is the corrupted one | **exactly 11**: 16,17,44,48,58,62,66,81,173,281,309 | CONF census (word_sources) |
| M4 | hiddentype_violation | HiddenType breaks its own rule | #161 '32', #187 'C', #119 missing, F-class undocumented | CONF |
| M5 | satoshi_reference | bitcoin-ref essence (valid BIP) | #151 | CONF (flavor, not corruption) |
| M6 | meta_essences | puzzle-vocabulary essences (seed/random/final/key…) | ~19 ghost+review | CONF census; robo's discounted |
| M7 | filename_typo | Arweave filename typos/descriptors | gltich/raindow; hand-written skull descriptors | CONF (partial coverage: 339/512) |

### S1-F · FILE-STRUCTURAL (byte-level; browser-invisible — Tier 2/3)

| ID | Type | Def | Members | Status |
|---|---|---|---|---|
| F1 | timing_hold | one 100ms frame in a 0ms file | #479 f27, #480 f24; (#146 f0 — encoder-quirk cand) | CONF |
| F2 | wrong_speed | whole delay table wrong for drop | 0ms robos {155,176,180,182,190}; 0ms review {446,467,468,469,512}; #228 50ms | CONF |
| F3 | negative_frame_bytes | byte-level negative frames | trio 231/256/274 identical {37,43,47,50,51,52,53}; #479 {1,2,3,4,7,11,17,26} arrow | CONF |
| F4 | byte_identity | two files byte-identical | #14 = #75 (MD5 + shared XMP + CDN) | CONF |
| F5 | resave_sessions | post-export Photoshop re-saves | 77 files over **exactly 11 sessions** (xmp_timeline); silent re-saves #47,203,407,454,457,460,461 | CONF (session count = our clustering — verify stability) |
| F6 | creatortool_odd | XMP tool version differs within drop | 23.1 twelve in review = final session {446,447,453,454,457,460,461,465,466,479,480,512} | CONF |
| F7 | frame_count_anomaly | frame count off drop standard | review 88/81/89/63 vs 55; #228 120 | CONF |
| F8 | psd_lineage | unique PSD-derived file | #300 (saved 12/21, solstice day) | CONF |

### S1-R · RELATIONAL (pairs, pointers, cross-contract)

| ID | Type | Def | Members | Status |
|---|---|---|---|---|
| R1 | deliberate_pairs | authored 2-member classes | **the 10**: {14,75},{150,179},{479,480},{209,300},{489,491},{130,133},{165,207},{363,473},{16,100},{175,205} | CONF/EYE; deck event-arithmetic: 10 pairs + 1 anchor = the ONLY construction hitting exactly 11 events |
| R2 | label_swap | two pieces exchanged labels | #150↔#179 (misfit collapses on exchange) | CONF (statistical) |
| R3 | double_mint | one piece minted twice | #390 (slots 519 traitless + 583) | CONF |
| R4 | pointer/homage edges | piece points at another (precursor/prelude/akin) | relation_edges in catalog: 207→solstice, 501-504→512, 217→512, 191→130, … | EYE |
| R5 | signposts | name corruptions pointing at a register | 0x20/0x21→#32/#33 (#32 = "hold on to this info" promise piece); Entangled/Invading spookies | cand (extraction A §signpost) |
| R6 | opposite_mismatch_pair | GANg-vs-behavior mirror pair | #281 (glitchGAN, doesn't) ↔ #47 (not glitchGAN, does; silently re-saved) | CONF |
| R7 | HODL_eleven | Pindar's own Fibonacci list {0,1,2,3,5,8,13,21,34,55,89} + partner-hop rules | straight read dead; partner +128 rule dead; other partner rules open | mixed |
| R8 | cross_contract | anomalies outside the 512 proper | #0 Pandimensional (off-contract, SuperRare); #66 SubType chain-vs-Arweave differs; forge/burn wallet 32 items/16 forges; burnt-Flaming = exactly 11 | CONF facts, unread as signal |

### S1-A · ABSENCE (lost things — invisible to anyone looking at what's present)

| ID | Type | Def | Members | Status |
|---|---|---|---|---|
| A1 | missing_file | file that should exist doesn't | #134 (only missing robo GIF); #446 missing Arweave twin; batch reprise07 missing | CONF |
| A2 | lost_art | true art public nowhere | #14's rainbow animation | CONF |
| A3 | traitless_slot | mint slot with no traits | #390's slot 519 | CONF |
| A4 | numbering_jokes | 104→404 "Not Found" in-slot replacement; 120→420 | CONF |
| A5 | forge_missing_25th | pair burned 2021-10-29, never minted | forge_mapping §5 | CONF |
| A6 | skipped_bip_words | 4 BIP words skipped in sequential essence order | community item #33 | cand (verify) |
| A7 | absent_register | ghost drop's proven NO hidden trait | drop-level | CONF (absence as register) |
| A8 | mint_holes | solstice never-minted slots | **CLOSED** — process noise (owner facts #5) | dead |

**Counting doctrine (lore A8/B26):** the clue counts GLITCHES, not pieces. Pairs/trios may collapse
to ONE event. The deck's event arithmetic found exactly one 11-hitting construction: R1's 10 pairs +
one anchor (#357 or #512). Any S1 row can also be recounted under intentional-only / natural-only /
both filters.

---

## S2. WORD-DERIVATION CATALOG — how a selected piece/event yields a BIP-39 word

**This is the frontier section.** The essence register (W1) has died on essentially every principled
selection; the unexplored rows below are where the puzzle most plausibly lives. "TESTED ON" lists
the selections actually swept with that register (per registry + run labels); everything else is
UNTESTED **for that register** even if the selection is "dead" under W1.

| ID | Register | Mechanism | Tested on (selections × orderings) | Status |
|---|---|---|---|---|
| W1 | essence (straight) | piece's Essence trait | ~81k sets: all ruled-out families, SET-1..4, union-11, Fib-11, feeling-11, mislabel-11, session-firsts/lasts, Eleven-A/B/B2/C, pairs+anchor… | HEAVILY TESTED, all dead. Decoy by Solstice precedent + 511-distinct design |
| W2 | title/name BIP words | BIP words in exact-case chain name / Title | name-eleven + unique-title-31 candidates built (word_sources); sweep provenance UNVERIFIED — audit before calling dead | PARTIALLY TESTED at best |
| W3 | SubType word | SubType if BIP (caps at 8 distinct) | not sweepable as 11-distinct; silver-11 (repeat-word) unswept | UNTESTED (structurally weak) |
| W4 | Type word | Type if BIP (8 distinct) | same ceiling | UNTESTED (structurally weak) |
| W5 | true-class word | what the art actually IS (mislabel inversion — Pindar's own published idea) | NOTHING — **blocked on the collector's namings** for the mislabel 11 | **UNTESTED — top frontier** |
| W6 | partner's word | pair relations: first-speaks / last-speaks / other-member | pairs+anchor 11 in 4 variants (q0827b, mint/rev/sudoku) DEAD; HODL+128 partner DEAD (q0827c); pointer-11 (5 sub-rules) UNSWEPT | PARTIALLY TESTED |
| W7 | number→wordlist index | wl[GAN], wl[GAN±1], wl[mint], wl[mint±1] | swept on historical selections (dead); **NOT on union-11 targets / onef11** (ext-B §5: exhaustion ran against earlier selections) | PARTIALLY TESTED — re-run on live selections |
| W8 | frame-index→wordlist / →piece | held/negative frame index as wl index or GAN pointer | SET-1 extraction variants only (#207→above/about, #14→announce) DEAD; onef11/union frame-reads UNSWEPT | PARTIALLY TESTED |
| W9 | painted-number index | hidden painted numbers (#165 "8941", #207 "#4") as wl/piece indices | inside SET-1 variants only | MOSTLY UNTESTED; hunt for more members first |
| W10 | radix decode | binary/octal/hex titles → numbers → word or pointer | #465/466 hex-revalue read as signpost only; never as an 11-register | UNTESTED |
| W11 | drop-channel word | each drop's glitch MEDIUM names the word (skull=splice… → a word per drop/piece) | NOTHING — register not even fully specified | **UNTESTED — needs design** |
| W12 | description words | contract description field per piece | censused (descriptions.md / analyze_descriptions.py); no 11 swept | UNTESTED as register |
| W13 | artist-tribute words | person-tribute pieces → tribute-derived word (A16; clue-page GIFs = tribute class incl. prize) | census NOT LAUNCHED | **UNTESTED — top frontier** |
| W14 | transplant (index hop) | positions from list A read out in list B (podGALLERY grammar B17) | union→upload-order transplant DEAD (q0827c, fwd+rev); montage-order, title-list, drop-internal transplants UNSWEPT (diff vs multistep_sets.json first) | PARTIALLY TESTED |
| W15 | SLIP→BIP list swap | same index, other wordlist | swept straight (zero adherents, dead); SLIP-only values = rainbow/scholar/phantom | TESTED dead as stated |
| W16 | filename descriptor words | Arweave hand-written vocabulary (ancient/elder/heart/warrior/whale…) | never swept as an 11 (assignment not forced; 173 pieces uncovered) | UNTESTED |
| W17 | date/season decode | Solstice-style date→season/month; re-save SESSION dates | Solstice spent for its drop; session-date reads UNSWEPT | UNTESTED |
| W18 | letter-payload / acrostic | case-deviant letter+position (B23 verified precedent) | census done (small classes, no 11); payload READING never assembled | UNTESTED |
| W19 | Eyes/Dimensions decode | integer trait → wl index | analytic dead (degenerate bip39[0..10] prefix) | TESTED-BY-ANALYSIS dead |
| W20 | event-word | one word per glitch EVENT (which-member-speaks per R1) | only via W6's pairs+anchor sweeps | PARTIALLY TESTED |
| W21 | absence-word | each authored absence contributes via its drop channel (A9) | absence census incomplete; nothing swept | UNTESTED |

**Register discipline:** a set is only "tested under register Wn" if the sweep actually derived its
words through Wn on that selection. The registry cannot tell you this — check run labels.

---

## S3. ORDERING CATALOG — every conceived ordering principle

"Stateable" = emitted by `orders6.stateable_orderings` (sort any recorded column fwd/rev; group by
any 2-8-value column then sort inside; artefact orderings). Deck `orderings.ranks` carries per-piece
ranks for the starred ones.

| ID | Ordering | Definition | Stateable in orders6? | Status |
|---|---|---|---|---|
| O1 | mint order ★ | TokenIndex_mintorder ascending — **canon default** (both solved puzzles verbatim) | YES | default for every sweep |
| O2 | reverse mint | O1 descending | YES | standard variant |
| O3 | GAN number ★ | piece number ascending (=O1 for 1-444; Review breaks it) | YES | standard |
| O4 | drop-then-position ★ | drop sequence, then position inside drop | YES (group+sort) | standard |
| O5 | Dimensions/Sudoku | 8 distinct-dim pieces pin slots 2-8 & 11; 3 wilds fill 1,9,10 | **NO** — constraint, not a sort; dedicated sudoku harness | swept on SET-1..4, Eleven-A..C |
| O6 | alphabetical ★ | by title (or by word — = BIP-index order) | YES | standard |
| O7 | upload order ★ | act1_uploadorder.json manifest | PARTIAL — only 68 Review pieces have ranks | used in transplant q0827c |
| O8 | montage display order ★ | montage_display_seq (512) | YES (artefact ordering) | available |
| O9 | gallery hanging order | clue-era CryptoVoxels/bitGALLERY arrangement (B20: was literally a password once) | **NO — data unrecovered** (Archive Hunter not launched; gallery mutated, owner facts #6) | frontier |
| O10 | grouped-then-ordered | group by type/drop/trait, order inside by another column | YES (the orders6 group mechanism) | standard |
| O11 | ends-pinned variants | #512/`final` last; #0/Moirai first ("before the first, after the last", B3) | NO — a constraint layered on any base order | applied ad hoc |
| O12 | random-word-position | canon: random 12th LAST only; variants put it elsewhere | n/a (sweep-structure choice) | canon per clue; variants foreclosed (S4) |
| O13 | save/session order | XMP UUID-v1 timestamps (exact save sequence known for final session) | NO (derivable but hacker-grade for a 2022 solver) | stated in ext-A §2 |
| O14 | BIP-39 index order | words sorted by wordlist index | YES (=O6 on words) | coincides with O1 on skull selections |

Practical note: orders6 also emits dozens of column-sorts (Eyes, Dimensions-as-sort, frames,
essence-alphabetical…) — those are covered by "YES" rows and enumerated per run, ~14-18 real
orderings per set.

---

## S4. GUARDRAILS

### 4a. Verbatim canon (quote exactly; never paraphrase into new constraints)

1. **The clue:** "The seed phrase is 11 glitches ending with a random word."
2. **Meta-tweet (2022-08-25, x.com/VanArman/status/1562799956175757313):** "Can the Unsolved bitGAN
   Puzzle FINALLY be solved before the end of the FINAL Drop? / [clue restated] / I can't give any
   more clues but am here to tell everyone that they have been looking for glitches all wrong."
3. **Discord same day (the collector's relay, authoritative):** "I just resolved the Glitch Puzzle. The
   back testing worked again. I was beginning to worry that I had made a mistake, but I haven't. It
   was backtested and I can get into the wallet. But man I have to admit, even knowing how to solve
   it, it was hard." / "So don't feel bad for struggling with it. Its obviously a tough puzzle"
4. **the collector-relayed (Discord):** the puzzle is in ALL 512 pieces, not just Act 1 Review.
5. **Precedent statements (solved puzzles):** "use the minting order of the bitGANs" (both solved
   seed puzzles); "Their essence is just a distraction, they hide their true BIP39 nature elsewhere"
   (Solstice clue); "It begins with sunrise and it ends with sunset" (ends pinned when hinted);
   "it doesn't matter how you find the seedphrase and there is no judge" (byteGAN).
6. **No unlockable-content channel; nothing in postcards/Artsticks; bitGALLERY mutated** (owner
   facts 1, 6, 8).

### 4b. the collector's working hypotheses as guardrails — each with its explicit OPPOSITE

| # | Guardrail (working hypothesis) | Opposite | Does Pindar's own word decide? |
|---|---|---|---|
| G1 | Solution is visual / OpenSea-visible (Tier 1) | Solution is in file bytes no 2022 browser showed | **NEITHER FORECLOSED.** "Looking for glitches all wrong" cuts both ways; byteGAN precedent honored an unintended byte route ("no judge"). the collector's tiering is a search-order policy, not canon |
| G2 | The eleven span all drops | Concentrated in one drop | **Narrow opposite FORECLOSED:** "in all 512, not just the Act 1 Review" kills review-only. A one-drop eleven inside another drop is merely DISFAVORED (all-512 could mean the search context) |
| G3 | Order = mint order | A non-mint ordering | NOT foreclosed for Act 1 — mint order is stated only for the two SOLVED puzzles. Strong precedent, so opposite = deliberate-violation territory |
| G4 | Essences are decoys | Essences matter for some pieces | NOT foreclosed — the decoy quote was the SOLSTICE clue. But the empirical record (81k dead essence sets, 511-distinct by design) makes W1 near-dead as a register regardless |
| G5 | Random word is LAST only | Random word sits elsewhere | **FORECLOSED by the clue itself** ("ending with a random word") + Solstice precedent of honest end-pinning. Only surviving loophole: "ending" as a pun (e.g. drop-ending), which is deliberate-violation territory |
| G6 | Rules statable in ≤2 sentences, no heavy computation | An algorithmic/computed solution | Not Pindar's words (the collector's bias) — but "even knowing how to solve it, it was hard" says EXECUTION is fiddly, not that derivation is computational. Disfavors compute, forecloses nothing |
| G7 | Puzzle is solvable and intact | Puzzle broken / key lost | **FORECLOSED:** backtested at least twice ("worked again"), wallet opens |
| G8 | More clues may come | — | **FORECLOSED:** "I can't give any more clues." Everything needed exists now |
| G9 | Pairs may count as ONE glitch (event counting) | Every piece counts separately | Neither foreclosed; deck arithmetic favors events (only 11-hit = 10 pairs + anchor) |
| G10 | Multi-step / two-step chains are "very Pindar" | Single-step direct read | Not foreclosed either way; podGALLERY/byteGAN precedent favors indirection; "hard even knowing how" fits multi-step |
| G11 | #357 and #512 are anchors in the eleven | Neither is in the eleven | Our evidence (6-way over-determination), not Pindar's words — sets excluding #357 are suspect, not forbidden |

### 4c. How to use this sheet
Every future theory must declare, before sweeping: (a) its S1×S2×S3 triple by row IDs, (b) which
guardrails G1-G11 it obeys, and (c) which it DELIBERATELY violates and why the violation is
motivated (a violation of a foreclosed row — G2-narrow, G5, G7, G8 — requires extraordinary
justification; a violation of a disfavored row just needs a stated reason). Negatives are then
logged against the triple, never against the S1 row alone.

---
# ADDENDUM 2026-08-27 — new registers, class, and the anomaly ledger

## New glitch class (S1)
- **near-miss / incomplete homage** — a piece that ALMOST renders as another entity
  (another bitGAN, an artist-GAN, an invader, Sol/Luna) but never completes the
  transformation. Confirmed members so far: #357 (almost invader / almost Sol-Luna,
  spikes never form), #423 (recycling symbol of Eclectic Method appears briefly).
  Fits Pindar lore ("trying to come into focus but couldn't"; "transforming into a
  different bitGAN"). Reframes "glitch" as incomplete transformation, not corruption —
  a SELECTION idea, not just a word idea. Census pending the collector's eye. STATUS: candidate.

## New word-derivation registers (S2)
- **W22 image-association word** — a BIP-39 word for what is VISIBLE in the art beyond
  the label: dominant anomalous color (#357->yellow, over-determined), objects/symbols
  in the frame (recycling symbol->recycle), background content (skulls, binary). Reads
  what the eye sees, not the metadata. UNTESTED except #357->yellow (swept dead on the
  strong elevens, q0827y).
- **W23 homage-target word** — the word of the ENTITY a piece homages or half-becomes
  (Eclectic Method's motif=recycle; an invader; Sol/Luna). Distinct from W22 (what's
  drawn) — this is WHO/WHAT it references. UNTESTED as a set.

## The anomaly ledger (data hygiene, standing rule)
`theories/anomaly_ledger.json` — canonical per-GAN record. EVERY anomaly found for any
GAN accumulates here additively (797 records / 405 GANs as of 2026-08-27; 108 GANs still
clean). Standing rule: whenever a new anomaly is found (by eye, byte, or census), append
a record {type, desc, source} to that GAN. Never delete; supersede with a new record if
corrected. This is the master index for "what's weird about piece X".

---
# ADDENDUM 2 — 2026-08-27 (Twin/Entangled ratification + name-census discoveries)

## S1 additions — RATIFIED and NEW glitch classes
- **TWIN (ratified by the collector, 13 members)** — two side-by-side GANs whose bodies never
  merge/interact. Members: 47,59,73,121,124,132,139,157,298,323,373,379,395. Per-drop
  DIALECT: skull=two separate figures; robo=tiled repetition of identical units;
  ghost=one wide body with two eye-lobes; monster=two heads. #379 "Spin Twins" is the
  authored exception (they converge/interact — the NAME says so).
- **ENTANGLED (NEW class, the collector)** — bodies consistently INTERMINGLED (not merely
  side-by-side). Visual definition matches Pindar's own naming: 15 named pieces
  217,222,226,244,260,263,281,411,436,450,465,473,481,489,499 (7 solstice, 2 ghost,
  6 review). the collector: entangled = quantum vocabulary; #100 is "entangled before Pindar
  had language for it". Essences: bachelor balcony banner believe birth blade book key
  circle solution tragic false price draw sound — note KEY, SOLUTION, BOOK, BELIEVE.
- **PALINDROME / nagaGAN (8 members, EXACT and COMPLETE)** — verified this pass:
  129,141,147,159,162,174,180,192. Triple-encoded: (1) palindrome NAME (RoboboR,
  Dr Awkward, Taco Cat, Dog God, Star Rats, Evil Olive, Flow Wolf, CalcuclaC);
  (2) their 6-bit binary positions are ALL EIGHT possible 6-bit palindromes
  (000000,001100,010010,011110,100001,101101,110011,111111) — the class occupies every
  one, so it is mathematically complete and authored; (3) animation runs forward then
  reverse (the collector: "mirror images in how they loop"). Essences in binary order:
  alpha awkward cat dog twin olive wolf zero (brackets: alpha first, zero last; #162's
  essence is literally "twin", linking naga to the Twin class).

## S3 additions — ORDERING vectors from the palindrome finding
- **O15 mirror/palindromic order** — the naga class encodes read-forwards-then-backwards
  at three levels. Test: order a candidate eleven so it reads the same both ways, or
  read the phrase in reverse (already partially covered by reverse-mint, but NOT as an
  authored mirror principle).
- **O16 binary-palindrome position order** — order by position-within-drop expressed in
  the drop's own radix, palindromes first/last.

## S2 additions — word vectors
- **W24 class-name word** — a piece's CLASS supplies the word (twin, entangled,
  palindrome, echo). #162's essence IS "twin"; #512's is "final". Self-naming pieces.
- **W25 glyph-in-art word** — letters/numbers rendered INSIDE the art (the collector saw "AI"
  in #379 frames 24-30; #281 eyes read "10 10"; #165 "8941"; #207 "#4"; #470-472/#245
  glyph eye-bands). Full-collection census in progress. the collector's hypothesis: some may
  be HAND-EDITED by Pindar pre-mint = deliberate payload.

## New SELECTION vectors (candidate elevens)
- V1 TWIN 13 -> narrow to 11 (drop the 2 dialect outliers, or drop Spin Twins + one).
- V2 ENTANGLED 15 -> narrow to 11 (7 solstice / 2 ghost / 6 review; puzzle-meta essences).
- V3 NAGA 8 + 3 (needs 3 companions — e.g. anchors #357/#512 + one).
- V4 QUANTUM THREAD: qubitGAN #130 -> 15 Entangled -> #512 Quantum Glitch (narrative
  chain the collector identified, pointing at the next collection).
- V5 DECIMAL-PALINDROME piece numbers (59 in range) intersected with anomaly-richness:
  #191(13 types) #66(11) #1(10) #4(9) #474(9) #505(9) #2 #7 #9 #11 #121 #363(8 each).

---
# ADDENDUM 3 — 2026-08-27: the glyph channel and its DECISIVE CRITERION

Full-collection glyph sweep complete (3 workers, all 512 pieces, eye-band sheets).

## THE CRITERION (the pass's most useful output)
**Does the glyph survive the subject's dissolution?**
- Glyph PERSISTS where the creature has dissolved  => BACKGROUND WALLPAPER = family
  grammar, not payload. Key exhibit #173 (robo tiling continues over dissolved robot).
- Glyph is PAINTED ON the creature and dissolves with it => candidate PAYLOAD.
This separates texture from payload without guesswork and should gate every future
glyph claim.

## Family grammars identified (NOT payload)
- ROBO drop alphabet: 0/box, I, 3 (mirrored E), E, C, 8/B — used for eye bands AND
  background tiling across #129-#192.
- GHOST/TEST drop: the word "TEST" tiled as wallpaper (~30 pieces) + repeated in the
  mouth band. Family theme (matches prior T3 drop-normal ruling).
- REVIEW reversal cluster #457-464: multi-row character cells; per-piece structures DIFFER
  (#459 red two-row bar/slot vs #462 blue full-width bars) but likely the same grammar —
  UNRESOLVED between workers, needs the criterion applied.

## Genuine payload candidates (survive as painted-on marks)
- #165 "8941" (the collector's own find) — VERIFIED distributed across top row + eye row +
  lower band at full-frame zoom. NOTE: spans the whole frame.
- #207 "#4"; #379 "AI" (the collector, frames 24-30); #281 "10 10".
- #196 NEW: isolated red mark above/between eyes, only red in frame, persists — reads
  '1'/'7'/hook. Spooky drop, NOT burnt. Strongest new candidate.
- #141 COHERENCE: eye band reads "0110" — itself a palindrome — on the palindrome-named
  nagaGAN at binary-palindrome position 001100. Name/position/motion/glyph all agree.

## KNOWN SENSITIVITY GAP (must state with any negative)
The sweep cropped the HEAD region (upper ~5-62%). #165 proves multi-digit numbers can
span the full frame, so the crop TRUNCATES them. Therefore "zero finds" in any drop —
notably the skull drop #1-#128, which returned zero confident finds — is WEAK evidence
of absence. A full-frame re-sweep is required before declaring the skull drop clean.