# Curated Elevens — the "one of each kind" hypothesis

Worker brief: test the hypothesis that the eleven glitches are **not** a similarity class but a
**curated list** — one representative of each distinct kind of glitch, held together by a thread
rather than by a shared attribute.

Everything below is derived from `master_dataset.json` (513 rows, 26 `G:` anomaly columns) and from
the 109 plain-text lore pages in `site/txt/`. Where I quote Pindar I name the file.

---

## 0. What the lore actually says, and what it constrains

The clue itself is in `site/txt/news__cryptoartgans.txt` (5 February), and it is worth quoting in
full because the surrounding sentences are load-bearing:

> "As far as cracking the wallet holding the Magic Internet moneyGAN. This one will be extremely
> hard, probably harder than all previous bitGAN puzzles."
> "Only 1 clue will ever be given, even if this takes over a year to solve."
> "**The Act 1 Puzzle: The seed phrase is 11 glitches ending with a random word.**"

Four things follow from the corpus that materially change how a candidate should be judged.

**First, both solved seed-phrase puzzles ordered the words by mint order, and Pindar says so in
plain words.** `news__the-puzzle.txt`: "The second step was to use the minting order of the bitGANs
to determine the order of these 12 words." `news__pattern-revealed.txt`: "And finally, their order
was according to mint order." Mint order is therefore the default ordering for every candidate here,
and any candidate that requires a different ordering owes an argument.

**Second, Pindar has publicly promised that two specific anomalies are unredeemed puzzle material.**
In `news__hidden-types.txt` and `journal__page002.txt` he lists "(32/128) is Type 32 (typo or
important?)" and "(119/128) is only Odd number of Type E (typo or important?)", and then answers his
own question: "**Hint: These are not 'Hidden Typos', but part of the puzzle that will only be
solvable after some more drops. Hold on to this info for later.**" That promise is never redeemed
anywhere in the 109 files. The Act 1 Puzzle is the only puzzle left to redeem it in. This is the
single strongest piece of artist-stated evidence in the entire corpus, and it argues that **#32 and
#119 belong in the eleven**. I have weighted candidates accordingly.

**Third, the puzzle is called "The Act 1 Puzzle", and Act 1 is bitGANs 0–512 in their entirety.**
`journal__page016.txt`: "With this drop bitGANs Act 1 is closing with 512 bitGANs." So the whole
collection is in scope, not only the Review drop. However, the sheet column `M_DropPuzzle` records
that the Ghosts in the Machine drop (144 pieces) and the Act 1 - Review drop (68 pieces) share a
single unresolved puzzle — "No puzzle/combined with Ghosts in the Machine" and "these dropped the
same time as Act 1 Review and introduced the CryptoArtGANs puzzle". That is a real argument for
weighting GAN 301–512 higher, and it is why the Act-1-Review-only pass in section 4 is worth doing.

**Fourth, a warning from the previous puzzle that applies here.** Official clue 3 of the Solstice
puzzle (`news__pattern-search.txt`) was "Their essence is just a distraction, they hide their true
BIP39 nature elsewhere." In that puzzle the words were in the *date* property, not the essence. I
have used the essence throughout below because it is the only BIP-39-valued column that covers all
512 pieces, but every candidate here should be understood as having a shadow version in which the
word comes from the piece's **Name** instead (16 pieces carry a BIP-39 word in their Name; the flag
is `G: Name contains BIP39 word`). If a candidate's set of eleven pieces is right and the words are
wrong, the shadow version is the cheap next thing to try.

**A cost note that should shape how these are tested.** Eleven fixed words supply 121 bits; the
twelfth word supplies 7 bits of entropy plus the 4 checksum bits. So for any fixed ordering of
eleven words, **exactly 128 of the 2048 possible twelfth words produce a valid checksum**. Testing
one ordering therefore costs 128 address derivations — nothing. The expensive dimension is *order*
(11! = 39,916,800). This is why "which order" deserves as much thought as "which eleven", and why a
candidate that comes with its own intrinsic order is worth more than one that does not.

---

## 1. Catalogue of distinct kinds of glitch, with their representatives

I have organised the catalogue by **where in the object the glitch lives**, because that is the
axis on which the collection actually varies, and because it is the axis a collector who has
handled the files — not just the OpenSea page — would use. A kind marked **[SINGLETON]** has
exactly one clean representative in all 513 rows.

### The art itself

| Kind | Representative(s) | Essence |
|---|---|---|
| Made by different hardware — the only bitGAN from a quantum computer | **#512 Quantum glitchGAN** **[SINGLETON]** | final |
| Wholly new abstract types at the end of Act 1 (Arweave folder 13) | #505–511 (seven) | elder, wild, cancel, south, network, capital, negative |
| Ghost with a black background instead of white | #501–504 (four) | project, source, broken, weird |
| Statistical visual outlier within its own SubType cohort | #97, #228, #251, #287, #362, #396, #482, #494, #507 (nine, each a different cohort) | — |
| Art contradicts its own SubType (Flaming but cold, Frozen but warm, Inverted but light) | #81, #150, #231, #326, #343, #490, #509, #511 (eight) | — |
| Visual Space Invader hidden in the art | 34 pieces | — |

### The animation

| Kind | Representative(s) | Essence |
|---|---|---|
| Recorded frame count disagrees with the measured frame count | **#129 RoboboR** **[SINGLETON]** (`FrameCount_disagrees` is set on exactly one row) | alpha |
| Frame count off the drop norm of 55/41 | 29 pieces; the extremes are #228 (120 frames), #129 (108), #445/#446 (88) | — |
| Pindar's "fibonacci-glitch" — non-Fibonacci frames inverted | #0 Pandimensional skullGAN, per thatbloom in `news__hidden-types.txt` | universe |
| The prime "misstep" glitch — jumping back and forth in time on a Fibonacci timing cycle | Pindar's named example is #73, `journal__page004.txt` | animal |

### The file

| Kind | Representative(s) | Essence |
|---|---|---|
| The only missing skullGAN GIF in the Arweave archive | **#57** **[SINGLETON]** | mistake |
| The only missing roboGAN GIF | **#134** **[SINGLETON]** | rain |
| Missing from the Act1Review folder | #446, #472, #477 (three) | rose, snake, december |
| Malformed filename — 5-digit binary where six digits are required (`14_00110.gif`) | **#143** **[SINGLETON]** | problem |
| Byte-identical duplicate pair — the only two identical files in 512 | #14 and #75 (a pair, so it cannot supply "one representative") | accuse, announce |

### The filename (Arweave)

| Kind | Representative(s) | Essence |
|---|---|---|
| "gltich" for "glitch" in the stored filename | #9, #73 (two) | absurd, animal |
| "raindow" for "rainbow" in the stored filename | **#128** **[SINGLETON]** | avocado |
| Arweave *folder* name misspelled — "lines" where every neighbour is "line" | **#486** **[SINGLETON]** | remember |
| Arweave folder 7 does not exist — folders run 0–13 with 7 absent | *no representative; this is a hole, not a piece* | — |

### The title

| Kind | Representative(s) | Essence |
|---|---|---|
| Truncated title — "Cronenber", missing its final g | **#123** **[SINGLETON]** | aunt |
| Missing space — "RainbowShimmer" | **#128** **[SINGLETON]** | avocado |
| Case error — "skullGan" / "spookyGan" | #175, #205 (two) | skull, grace |
| Leetspeak title — 1 and 0 for i and o ("G1itch1ng", "B1-b0rg") | 13 pieces | — |
| Title contradicts the GANg it is filed under | #39, #154, #210, #211, #420, #474, #498, #502, #503, #512 (ten) | — |
| Named after an HTTP error | **#404 "404 Not Found"** **[SINGLETON]** | crash |

### The Type field

| Kind | Representative(s) | Essence |
|---|---|---|
| Misspelled Type — "Gltich" for "Glitch", and a Type value with n=1 | **#73** **[SINGLETON]** | animal |
| A Type that occurs exactly once in 513 rows | #130 Mechanic, #141 Kook, #147 Rotator, #162 Reviver, #194 Sprite, #445 Collector (six, plus #73) | quantum, awkward, cat, twin, goat, collect |

### The SubType field

| Kind | Representative(s) | Essence |
|---|---|---|
| The value written backwards — on-chain SubType "wobniaR" | **#48** **[SINGLETON]**; the only reversed value anywhere in the data | album |
| Misspelled SubType — "Raibow" | #170, #175 (two) | quiz, skull |
| A SubType that occurs exactly once | #0 Pandimensional, #162 Dixid, #174 Dioid, #180 Sus, #192 Tenet, #195 Anonymous, #206 Skull, #328 Invading | — |

### The Essence field — the BIP-39 register

This is the most structured finding in the dataset and I want to state it precisely, because it has
not been written down anywhere in the working directory.

**The essence column is a register, not a decoration.** In the skullGAN drop the rule is exact:
`bip39_index(essence) = GAN − 1`, with no exceptions across 128 pieces except two. The spookyGAN
drop (193–208) is likewise perfectly sequential. The Solstice drop (209–300) is sequential over
BIP-39 indices 128→223 with three skips and two transpositions. There are no duplicate essences
anywhere in 512.

| Kind | Representative(s) | Essence |
|---|---|---|
| Not a BIP-39 word at all — zero substituted for the letter o | **#191 "err0r"** **[SINGLETON]**; to use this piece you must repair the glitch, and the repaired word "error" is BIP-39 #576 | error |
| An intruder from far up the register — #21 holds index 156 where index 20 belongs | **#21 "bean"** **[SINGLETON]**. The displaced word, **"action" (index 20), is the only BIP-39 word in the whole consumed low band that is never assigned to any bitGAN** | bean |
| An intruder from very far up the register — #57 holds index 1135 where index 56 belongs | **#57 "mistake"** **[SINGLETON]**. The word it displaced, "alpha" (index 56), reappears as the essence of #129, the first roboGAN and the first piece of the next drop | mistake |
| Transposed inside a sequential run | #259 "bless" and #268 "bird" (two) | bless, bird |
| Cast to a spreadsheet boolean, destroying the word | #363 "TRUE", #473 "FALSE" (two) | true, false |

So the essence register contains exactly seven damaged slots: #21, #57, #191, #259, #268, #363,
#473 — plus one empty slot, "action", that no artwork holds. That is a beautiful thread and it does
not reach eleven. I say so plainly in section 3 rather than padding it.

### The number and the numbering system

| Kind | Representative(s) | Essence |
|---|---|---|
| HiddenType is the literal string "32" where a letter belongs — Pindar flagged it | **#32** **[SINGLETON]** | advance |
| The only odd-numbered Type E — Pindar flagged it | **#119** **[SINGLETON]** (note: our copy of the sheet records #119's HiddenType as blank, so this rests on Pindar's statement, not on our data) | attract |
| HiddenType contradicts the number rule | #187 (C where the rule demands E) | mirror |
| An undocumented sixth HiddenType, F | #129, #141, #147, #159, #162, #174, #180, #192, #193 (nine, all nagaGAN palindromes) | — |
| Numbered in binary instead of decimal | 56 pieces in the roboGAN drop; inside Act 1 Review, #457–464 and #497–500 (twelve) | — |
| Numbered in octal | #193–208 (sixteen, "0o0" through "0o17") | — |
| Numbered in hexadecimal | **#465, #466** (two, "0x20" and "0x21" — the only hex numbers in 513) | tragic, galaxy |
| Prime-glitch rule violations | #5 glitches when the rule says it should not; #47 is the one non-Fibonacci prime in the first drop that does not glitch. A complementary pair, a false positive and a false negative | above, alarm |
| Uniquely named but not on the Fibonacci sequence — Pindar: "Not sure where that fits in" | **#39 Flaming cyclopsGAN** **[SINGLETON]** (`journal__page003.txt`) | age |

### The chain

| Kind | Representative(s) | Essence |
|---|---|---|
| Chain asserts a different SubType than the artist's ledger, changing what the piece *is* | **#66** (sheet Rainbow, chain Flaming) **[SINGLETON]** | amused |
| Chain and sheet disagree on some field | 20 pieces | — |
| The chain is the corrupt side | 11 pieces: #16, #17, #44, #48, #58, #62, #66, #81, #173, #281, #309 — **already eliminated as `chaincorr` in `closeout_sets.json`** | — |
| Chain and sheet disagree on the GANg itself | **#281** (sheet glitchGAN, chain bitGAN) **[SINGLETON]** | book |
| No on-chain traits at all | 120 pieces, but the sheet itself notes this is a Blockscout cache artifact spread evenly across drops — not a property of the artwork. Do not use | — |

### The schema

| Kind | Representative(s) | Essence |
|---|---|---|
| The trait *key* is misspelled, not the value — "Subtype" with a lowercase t, unique in 393 on-chain records | **#383 Space-clops** **[SINGLETON]** | style |

### Lore and identity

| Kind | Representative(s) | Essence |
|---|---|---|
| Given a proper Name, not just a title | 38 pieces | — |
| Given personhood and its own chapter in the lore | #0 Pandimensional, #1 The Moirai, #2 The Artist, #512 Quantum Glitch. #3 The Geometer is named in the Fibonacci list and **never mentioned again anywhere in 109 files** — a genuine hole | universe, abandon, ability, able, final |
| Named after a past puzzle solver | #210 noohpGAN, #211 mazzoneGAN (two) | awake, aware |
| Named after a living cryptoartist | #228 Abosch, #246 Hackatao, #264 Killer Acid, #282 Eclectic Method, #445 Colborn, #446 SuperRareRoses, #447 Brinkman, #448 StellaBelle | — |
| Unique Type "Collector" — the prize itself | **#445 Magic Internet moneyGAN** **[SINGLETON]** | collect |
| Burnt in the forge to make a SuperRare | 30 pieces | — |

---

## 2. The artworks that are the only example of their kind

Consolidated, because this is the list the brief asked for. Twenty-one pieces are, on the evidence
available, the sole clean representative of a kind:

| GAN | Essence | The kind it alone represents |
|---|---|---|
| #21 | bean | The essence register's near intruder; displaces "action", the one BIP-39 word never used |
| #32 | advance | The only numeric HiddenType — and Pindar flagged it as unredeemed puzzle material |
| #39 | age | The only uniquely named bitGAN that is not on the Fibonacci sequence |
| #48 | album | The only value in the collection written backwards ("wobniaR") |
| #57 | mistake | The only missing skullGAN file; the essence register's far intruder |
| #66 | amused | The only piece whose chain asserts a different SubType than the ledger |
| #73 | animal | The only misspelled Type ("Gltich"); Pindar's named example of the prime glitch |
| #119 | attract | The only odd-numbered Type E — Pindar flagged it as unredeemed puzzle material |
| #123 | aunt | The only truncated title ("Cronenber") |
| #128 | avocado | The only "raindow" filename; the only title missing a space |
| #129 | alpha | The only piece whose measured frames disagree with its recorded frames; holds the word #57 displaced |
| #134 | rain | The only missing roboGAN file |
| #143 | problem | The only malformed filename (5-digit binary where six are required) |
| #191 | error | The only non-BIP-39 essence in 512 |
| #281 | book | The only chain/sheet disagreement on the GANg itself |
| #383 | style | The only record whose trait *key* is misspelled |
| #404 | crash | The only piece named after an HTTP error |
| #445 | collect | The only Type "Collector"; the prize |
| #486 | remember | The only misspelled Arweave folder name ("lines") |
| #512 | final | The only piece from a quantum computer; the only xxxxxGAN; the only Dimensions = 11 |
| #14/#75 | accuse/announce | The only byte-identical pair — but it is a pair, so it cannot contribute one representative |

---

## 3. Candidate elevens over the whole collection

### Candidate 1 — "The Stack" *(strongest; this is the one I would run first)*

**The thread, in a sentence a collector would nod at:** *A bitGAN is not one object but a stack of
eleven representations — the art, the animation, the file, the filename, the title, the Type, the
SubType, the essence, the number, the number's rule, and the chain — and Act 1 contains exactly one
piece that is broken at each layer and nowhere else.*

| # | Layer | GAN | Word | Why it is the only clean representative |
|---|---|---|---|---|
| 1 | the art | 512 | final | The only bitGAN from a quantum computer |
| 2 | the animation | 129 | alpha | The only piece whose measured and recorded frame counts disagree |
| 3 | the file | 57 | mistake | The only missing skullGAN GIF |
| 4 | the filename | 143 | problem | The only malformed filename |
| 5 | the title | 123 | aunt | The only truncated title |
| 6 | the Type | 73 | animal | The only misspelled Type |
| 7 | the SubType | 48 | album | The only reversed value |
| 8 | the essence | 191 | error | The only non-BIP-39 essence |
| 9 | the number | 32 | advance | The only numeric HiddenType — Pindar-flagged |
| 10 | the number's rule | 119 | attract | The only odd-numbered Type E — Pindar-flagged |
| 11 | the chain | 66 | amused | The only chain that asserts a different SubType |

**Mint order** (identical to GAN order for this set, since only #512 sits outside the first drop):
`advance album mistake amused animal attract aunt alpha problem error final`

**Why I rank it first.** It is the only candidate that contains both pieces Pindar personally
flagged and personally promised would matter later, and it contains them for a reason internal to
the thread rather than because I wanted them. Every member is a documented singleton. The phrase
ends on "final", which is the last-minted bitGAN in the entire collection — a natural place for a
seed phrase about the close of Act 1 to end. And the ordering question resolves cleanly: mint order
and GAN order agree, so there is only one order to try.

**Where I am reaching.** The layer list is elastic. I could have named nine layers or thirteen; I
named eleven partly because eleven is the target. That is exactly the failure the calibration rule
warns about, and I will not pretend otherwise. What partly rescues it is that each layer is a real
seam in how these objects exist — you can point at the GIF, the filename, the sheet row, and the
chain record as four separate artefacts — and that the representative at each seam was singular
*before* I started counting.

### Candidate 2 — "Eleven ways a computer says it is broken"

**The thread:** *Every one of the eleven is a different named failure mode from computing — not-found,
character substitution, string reversal, transposition, missing asset, type coercion, case
sensitivity, field truncation, type mismatch, dropped character, and nondeterminism — so the list
reads as a glossary that only someone who writes software would assemble.*

| # | Failure mode | GAN | Word |
|---|---|---|---|
| 1 | integer where an enum letter belongs (type mismatch) | 32 | advance |
| 2 | string reversal | 48 | album |
| 3 | missing asset | 57 | mistake |
| 4 | character transposition | 73 | animal |
| 5 | dropped character | 123 | aunt |
| 6 | fixed-width field truncated | 143 | problem |
| 7 | leetspeak character substitution | 191 | error |
| 8 | string coerced to boolean | 363 | true |
| 9 | case-sensitivity fault in a key | 383 | style |
| 10 | HTTP 404, resource not found | 404 | crash |
| 11 | nondeterminism / quantum indeterminacy | 512 | final |

**Mint order** (again identical to GAN order): `advance album mistake animal aunt problem error true
style crash final`

**Why second.** The thread is genuinely stateable and genuinely collector-shaped; the taxonomy is
computing's own, not one I invented. But it overlaps heavily with `curated11` in
`closeout_sets.json` (#9, #57, #73, #123, #128, #134, #143, #170, #175, #205, #404 — the
all-typos-and-missing-files set), which has already been eliminated. This candidate differs by
swapping four members and by reaching outside the typo class for #32, #363, #383 and #512, so it is
worth running, but the family it belongs to has been probed.

### Candidate 3 — "The Fibonacci names" *(not a curated eleven; a similarity class — but the thread is Pindar's own words, and it has not been tested)*

I am including this against the letter of my brief because it is the strongest artist-stated thread
in the corpus and it does not appear in `closeout_sets.json`.

**The thread, in Pindar's words** (`journal__page003.txt`): *"So if the skullGAN's number is a
Fibonacci, it has a unique name."* He then lists exactly eleven: 0 Pandimensional skullGAN, 1 The
Moirai, 2 The Artist, 3 The Geometer, 5 Primordial whaleGAN, 8 Flaming blobGAN, 13 Flaming
spiritGAN, 21 Beanie skullGAN, 34 Heart skullGAN, 55 Silver Warrior skullGAN, 89 Primordial
sharkGAN. And immediately afterwards: *"Only one one other bitGAN has a unique name but is not on
Fibonacci sequence — #39 the cyclopsGAN. **Not sure where that fits in.**"*

Eleven pieces held together by a rule Pindar states himself, plus exactly one piece he says he
cannot place. "Eleven glitches ending with a random word" reads, on this thread, as *the eleven
Fibonacci-named pieces, ending with the one that is there at random*. The word "glitch" is not
strained either: "fibonacci-glitch" is Pindar's own adopted term for the frame-inversion pattern
(`news__hidden-types.txt`, quoting thatbloom), and `journal__page004.txt` says the prime-numbered
skullGANs — which include #2, #3, #5, #13 and #89 — "have a visible glitch".

**GAN 0 has no mint index** (it is a SuperRare 1/1 on a different contract), and Pindar describes it
in `news__the-drop.txt` as "the 0th bitGAN, before and after all other bitGANs". So it goes first,
and the rest follow in mint order, which for this set is also GAN order.

`universe abandon ability able above abstract account bean aerobic almost arch` + twelfth word `age`

**Also try** the same eleven with GAN 0 placed *last* rather than first, which is what "before and
after all other bitGANs" literally licenses, and with "age" as the twelfth word instead of a
brute-forced one. Testing all four combinations costs almost nothing.

### Candidate 4 — "The essence register errata" *(does not reach eleven; reported honestly)*

**The thread:** *The essence column is a 2048-slot BIP-39 register in which word N is issued to
bitGAN N, and a handful of pieces sit in the wrong slot.* Members: #21 bean, #57 mistake, #191
error, #259 bless, #268 bird, #363 true, #473 false. **Seven.** Plus one empty slot — "action",
index 20, the only BIP-39 word in the consumed low band that no artwork holds.

I could reach eleven by adding four transposed or skipped words, but every way I tried to do it
required a rule I had invented five minutes earlier. Per the calibration rule I am reporting this as
a seven-member kind rather than dressing it as an eleven. It is still, in my judgement, the most
*Pindar-shaped* structure I found in the data, and if another worker can find a principled fourth
rule that adds exactly four more, that set should jump to the top of the list.

---

## 4. Candidate elevens restricted to the Act 1 Review

Scope note before the candidates. The Act 1 Review block is bitGANs **445–512 (68 slots)**, and
`journal__page015.txt` confirms the Arweave archive boundary: "Act 1 - Review (bitGANs 445-512)".
But `journal__page016.txt` and `news__the-final-bitgan-drop.txt` both say the drop is **64** pieces.
The discrepancy resolves cleanly: 445–448 are the four cryptoArtGANs from the separate February drop
(`news__cryptoartgans.txt` — StellaBelle, Colborn, SuperRareRoses, Brinkman), and 449–512 is the
64-piece Act 1 Review proper. **#445 Magic Internet moneyGAN is the prize and I have excluded it from
every Act 1 Review candidate** — a prize should not be a key to its own wallet.

Mint order within the drop is not GAN order: #447 and #448 minted first (585, 586), and #445, #446
and #512 minted last (660, 661, 662). Fourteen pieces — #457–466 and #497–500, exactly the pieces
numbered in binary or hexadecimal rather than decimal — **have no mint index at all**. That
correlation is itself a finding: in the Act 1 Review, abandoning decimal numbering and never
reaching the chain are the same event.

### Candidate 5 — "The Review's broken seams" *(best Act-1-Review-only candidate)*

**The thread:** *The Act 1 Review is a curated retrospective that re-stages every earlier drop, and
eleven of its pieces are the ones where the re-staging visibly fails — each in a different way, and
each way happening exactly once.*

| # | The seam | GAN | Word |
|---|---|---|---|
| 1 | the file was never delivered to the folder | 446 | rose |
| 2 | filed in the wrong block — a testGAN among the seasonals | 468 | random |
| 3 | the essence was eaten by the spreadsheet and became a boolean | 473 | false |
| 4 | the title contradicts the GANg it is filed under | 474 | truth |
| 5 | it carries a seed word from the already-solved Solstice puzzle | 477 | december |
| 6 | it is the only member of its GANg in the whole collection | 480 | fossil |
| 7 | the drop's statistical visual outlier | 482 | call |
| 8 | its Arweave folder name is misspelled ("lines" for "line") | 486 | remember |
| 9 | the first of the seven wholly-new types that belong to no earlier drop | 505 | elder |
| 10 | the art contradicts its SubType — Inverted but light | 509 | network |
| 11 | not from the same machine at all | 512 | final |

**Mint order:** `random false truth december fossil call remember elder network rose final`
**GAN order:** `rose random false truth december fossil call remember elder network final`

Both are worth running. Mint order has Pindar's precedent behind it; GAN order ends on "final",
which is the more satisfying close.

**Where I am reaching.** Seams 5 and 7 are softer than the rest — "carries a Solstice word" and
"statistical outlier" are kinds I chose because I needed eleven, not kinds the collection announces.
Seams 1, 3, 4, 6, 8, 10 and 11 are each documented singletons and I stand behind them.

### Candidate 6 — "The Review's broken seams, hedged"

Identical thread, with the two softest members swapped out: #468 "random" moved to the twelfth
position (see below) and #496 "reward" — one of the four high-definition skullGANs, and the only
Act 1 Review piece whose Name contains a BIP-39 word other than the prize's — taking its slot.

**Mint order:** `false truth december fossil call remember reward elder network rose final`

**On "ending with a random word".** #468 Flaming testGAN has the essence **"random"**, is one of
only three pieces the sheet flags as filed in the wrong block, and carries a new background. If the
clue is a pun — and Pindar's puzzles are full of them, from the wobniaR reversal to the SLIP-39
substitution in the podGALLERY solution — then the twelfth word is not arbitrary at all. It is
literally "random", and Candidate 6 is Candidate 5 with that reading applied. This costs one
derivation to test per ordering and should be tried before any brute force.

### Candidate 7 — "The Review sheet-flag sampler"

**The thread:** *One representative of every anomaly the sheet records inside the Act 1 Review,
walking the drop from front to back and taking a different kind at each step.* Deliberately built
from different representatives than Candidate 5 so that the two hedge each other rather than
overlap.

`divide give snake false song deposit knee sort source cancel final` — GANs 449, 467, 472, 473, 479,
485, 490, 494, 502, 507, 512, in GAN order, which for these eleven is also mint order.

I believe this one least. It is a sampler, and a sampler is not a thread; I include it because the
brief asked for a second Act-1-Review pass and because it covers kinds the other candidates do not.

---

## 5. Ordering, for every candidate

**Try in this sequence.**

1. **Mint order.** Pindar says in plain words that both solved puzzles used it
   (`news__the-puzzle.txt`, `news__pattern-revealed.txt`). This is the default and nothing else
   should be tried until it has failed.
2. **GAN number order.** For Candidates 1, 2 and 3 this is identical to mint order, so those
   candidates cost only one ordering each. For the Act 1 Review candidates it differs, mainly
   because #445, #446 and #512 minted last, and it is the ordering that puts "final" at the end.
3. **The thread's own order.** A curated list carries an intrinsic sequence that a similarity class
   does not. For Candidate 1 that is stack depth — art, animation, file, filename, title, Type,
   SubType, essence, number, rule, chain — giving `final alpha mistake problem aunt animal album
   error advance attract amused`, and its reverse. This is the ordering that only works if the
   curated hypothesis is right, so it is also the sharpest test of it.
4. **Reverse mint order,** for the Act 1 Review candidates only. The drop is a *review* — a reprise
   that looks backwards — and reading it backwards is a defensible reading rather than a guess.

For each ordering, only 128 of the 2048 possible twelfth words satisfy the checksum, so each
ordering is 128 cheap derivations. Every candidate in this document, across all four orderings, is
under six thousand derivations in total. None of this needs a sweep.

---
*Note added 17 September 2026.* The "seven damaged slots" above (#21, #57, #191, #259, #268, #363, #473) were
confirmed and completed by Pindar's own bip-0039 index column, found in the 1 January 2022 capture of journal
page 015 (see FINDINGS-2026-09-16.md). His column shows #259 and #268 are a transposed pair rather than two
displacements, decodes #191 to `error`, and adds four robots that hold words from the skull–solstice block
(#129, #131, #141, #189). The eleven is now the selection SEL-SCHEME-EXCEPTIONS. Nothing in this file is
closed; the sweeps recorded here ran on data corrected since (D70).