# Hypotheses from the January table — RUN 16 Sep 2026, all negative

Written 16 September 2026 after Michael's ask to think expansively about the swapped and
missing words, the chain-error moves, and stacking, then ratified by him and run the same day.
Every statement below has a selection rule, a word rule and an ordering rule.

## The two facts everything rests on

The block of BIP-39 indices 1–224 has 224 slots. 223 are held. **One is vacant: 21, `action`.**
Exactly one piece from the two drops that fill the block sits outside it: **skull #57, holding
`mistake` (1136).** Four robots reach into the block and take `alpha`, `awkward`, `blind`, `blue`.
Skull #21 takes `bean` from the solstice range.

## Set A — the word-scheme exceptions (11 pieces)

Selection: every piece whose essence is not the plain wordlist word the allocation gives it.
Four reasons, one set:

| piece | why it is an exception |
|---|---|
| #21  | holds `bean` (157); its own slot 21 `action` is empty |
| #57  | holds `mistake` (1136); its own slot 57 `alpha` was given away |
| #129 | robot holding block word `alpha` (57) |
| #131 | robot holding block word `blind` (191) |
| #141 | robot holding block word `awkward` (135) |
| #189 | robot holding block word `blue` (195) |
| #191 | prints `err0r`; Pindar's index 615 says `error` |
| #259 | prints `bless`; index 181 says `bird` |
| #268 | prints `bird`; index 190 says `bless` |
| #363 | `TRUE` — not a wordlist word |
| #473 | `FALSE` — not a wordlist word |

Eleven, and the count was not chosen — it fell out of the rule.

Word rule, two readings: A-printed (the essence as the sheet prints it) or A-index (the word
Pindar's own index points at; differs on #191, #259, #268).
Ordering: piece order and mint order give the SAME sequence for this set.

    A-printed, piece/mint order: bean mistake alpha blind awkward blue err0r bless bird true false
    A-index,   piece/mint order: bean mistake alpha blind awkward blue error bird  bless true false
    A-index,   index order:      alpha awkward bean bird bless blind blue error mistake true false

TRUE and FALSE have no index; in index order they can only sit at the ends.

## Set B — the 13 chain-error pieces, read as moves

| piece | essence | move it teaches |
|---|---|---|
| #4 | about | change a capital |
| #16 | acid | swap two letters |
| #48 | album | reverse the word |
| #58 | already | empty a field, double another |
| #62 | amateur | swap two fields |
| #66 | amused | add a stray character |
| #73 | animal | swap two letters (sheet side) |
| #81 | anxiety | cut a word from inside |
| #88 | april | drop a letter |
| #97 | army | cut the suffix |
| #100 | arrest | swap two letters |
| #124 | author | add a letter |
| #309 | frost | cut the suffix |

Applying each move to that piece's own essence never yields another wordlist word. So Set B
does not carry the words. **Set A and Set B are disjoint** — no piece is in both. That is the
strongest argument yet for stacking: one set supplies the words, the other supplies what to do
with them.

## Stacked hypotheses

**S1 — Set A words, Set B moves.** The 13 chain errors are instructions. Read them in mint
order and apply each move, in turn, to the Set A words. "Swap two letters" un-swaps #259/#268.
"Drop a letter" turns `err0r`... into nothing; "change a capital" turns `TRUE` into `true`.
The open question is which move applies to which word — by position, by drop, or by matching
kind (a swap-move to the swapped pair).

**S2 — Fill the empty chair.** Selection: Set A. Word rule: A-index. Twelfth word: `action`,
the one word deliberately removed from a complete block. Ordering: piece order. This is the
only hypothesis where the twelfth word is *motivated* rather than enumerated over the 128
checksum-legal options.

**S3 — The displacement path.** Selection: the six pieces the block displacements touch, in
the order the indices climb (21 → 57 → 135 → 157 → 191 → 195), reading the word that SHOULD
have been there and the word that IS there: action/bean, alpha/mistake, awkward, bean,
blind, blue. Too few for eleven on its own; a stacking candidate for the front of a phrase.

**S4 — Mistake then action.** The two anomalies of the block are a word called `mistake`
brought in and a word called `action` sent out. If the puzzle's own vocabulary is speaking,
the instruction is "fix the mistake, then act": un-swap #259/#268, decode #191, restore #21
and #57 to `action` and `alpha`, and read the corrected block in order. That gives a
corrected wordlist run, not an eleven — a derivation rule for other selections to use.

## What was run, and what came back

Twenty eleven-word orderings, each against the 128 checksum-legal twelfth words and all 33
derivation paths: 2,560 seeds. A planted seed inside the queue was detected, so the run could
see a hit. **No hit.**

| id | selection | words | order | result |
|---|---|---|---|---|
| H1 | Set A | as the sheet prints them (`err0r` read as `error`) | piece = mint, fwd + rev | negative; `action` was a legal 12th and was tested |
| H2 | Set A | the word at Pindar's index (swap and leet corrected) | piece, fwd + rev | negative |
| H3 | Set A | index-corrected | ascending index, TRUE/FALSE at either end, fwd + rev | negative |
| H4 | Set A | the word that SHOULD be there (#21 `action`, #57 `alpha`) | piece, fwd + rev | negative |
| H6 | the nine indexed exceptions + #0 `universe` / #512 `final` | index-corrected | piece and index, fwd + rev | negative |
| H10 | Set A | printed words in piece-index order (keeps the swap) | fwd + rev | negative |
| H11 | the ten disturbed WORDS themselves + one bookend | — | index order, fwd + rev | negative |
| H7 | the ten disturbed words + both bookends, twelve fixed | — | 4 orderings | all four fail the BIP-39 checksum; dead by construction |

S1 (Set B moves applied to Set A words) was not run: the rule for which move applies to which
word is not yet stated in a sentence, so it cannot be enumerated honestly. S3 and S4 are
derivation ideas, not selections, and were folded into H4 and H11.

The sets are in `tested_sets_no_hit.txt`; the run scripts and worker logs are in
`runs_2026-09-16/`; both runs are on the Runs tab.

## H12 — the glitches are edits at positions (run later the same day, negative)

Michael's question: is there a version where correcting the glitches gives the order? Yes.
Every length-preserving chain glitch is an edit at a letter position, and an 11-word phrase has
positions too:

| glitch | what it did | applied to the phrase |
|---|---|---|
| `Primoridal` (#16, #100) | letters 7 and 8 swapped | swap slots 7 and 8 |
| `Gltich` (#73) | letters 3 and 4 swapped | swap slots 3 and 4 |
| `wobniaR` (#48) | reversed | reverse the phrase |
| #62 | sheet columns 5 and 6 exchanged (SubType, Type) | swap slots 5 and 6 |
| #58 (variant) | SubType emptied, Type doubled | slot 5 := slot 6, a repeat |

Applied in mint order to scramble a clean phrase, and in reverse order to unscramble one, on
four bases (Set A printed, index-corrected, should-be, and in index order): 16 orderings, 2,048
seeds, plant seen, **no hit**.

Not run: the length-changing glitches — a dropped letter (#88, #123, #175), an added one
(#124, #66), a cut suffix (#81, #97, #309). Each turns eleven words into ten or twelve, and the
rule for the freed or added slot has to be said aloud first (the obvious candidate: the vacated
word `action` fills an insertion).