bitGAN Act 1 · "32 as the building block" · brainstorm status

The 32 Decoder Board

Seven idea-generators proposed 49 ways the number 32 could crack the puzzle. Deduped to 24 distinct ideas, each tagged by exactly where it stands: run to a result, ruled out for another reason, or still open.

17
Tested & dead
7
Killed — other
0
Remaining
1
Kept as fact
Complete. Every actionable idea on this board has now been run to a result. The 32-based word-registers and pixel reads are all dead; the selection rules don't isolate eleven; the base-system analysis explains the two "32" corruptions but yields no selector. One durable fact survives: the {#32,#95} same-transaction burn pair. The consistent lesson — 32 organizes the collection's structure but never names its content.

Tested & dead

8

Actually run — swept through the plant-validated harness or checked directly against the data — and came back negative.

Pixel-diff → 32×32 grid cell → 11-bit word index
swept dead
Uses 32: the artwork is a 32×32 grid; the most-changed cell's row+column (10 bits) plus a brightness bit make exactly 11 bits = one wordlist word.
The first idea to read actual pixel data. The math was clean, but the glitch on the Fibonacci eleven is a re-inserted whole animation frame — the entire figure redraws, so the "most-changed cell" is diffuse (it beats the runner-up by under 7%), not a spike. 16 variant word-sets (two reference frames × row/col-major × 0/1-based × max-cell/centroid) swept, plant-validated, no hit. Forensics and sweep agree: no single cell carries the message.
The ±32 "hop" across glitch classes
killed by control
Uses 32: add/subtract 32 from a glitch piece's number to find a partner piece (five of seven generators proposed this).
Debunked by the control. A +32 shift lands on another flagged piece only 20.0% of the time — lower than a shift of 16 (30.6%) and no better than 7 or 11. 32 is not a special distance; the hits are just the robo drop being densely flagged at binary intervals. (The one real survivor: #129↔#161, both still-eyes, differ by exactly the 32-bit — a structural note, not a mechanism.)
The doubling eleven {0,1,2,4,8,16,32,64,128,256,512}
swept dead
Uses 32: 32 is the 6th term of the powers-of-two list — the multiplication twin of Pindar's own Fibonacci HODL list.
Essence + piece-number + mint registers × 4 orderings, plant-validated, no hit. (The powers-of-2 skulls carry bip39[2^k−1] as essence by design.)
Collection-wide mint-position ÷ 32 filter
wrong count
Uses 32: keep every piece whose mint number is a multiple of 32.
Produces 17 pieces, not 11. Clean miss.
The 32nd glitch in mint order
null
Uses 32: walk the glitches in mint order and take the 32nd as a pointer.
Lands on #126, a member of no ratified class. Nothing there.
Mint-rank window [97, 243] as a class filter
too coarse
Uses 32: a 32-based band of mint ranks used to bracket candidates.
30+ pieces fall inside it — far too broad to select anything.
#32 / #161 as adjacency slot-markers
chance-level
Uses 32: treat the two "32" pieces as markers of which seed slots need a hop.
Their adjacency to existing classes is only chance-level. No signal.
Eyes as a 5-bit register / eye-count sums to 32
disproven by data
Uses 32: 5 bits = values 0–31; or a class whose eye-counts total 32.
Eye counts only ever run 0–10, nowhere near 32; class sums miss 32 by 2–3. This closes the literal version of the still-open "eyes-as-binary" idea without resolving that idea itself.

Killed for other reasons

5

Ruled out by logic, by an existing ruling, or because the premise itself failed on investigation.

bitGANg 1001's 32-piece roster
premise fails
Uses 32: a Pindar-curated set of exactly 32 pieces — intersect with the glitch classes, hope for an overlap of 11.
Investigated fully. bitGANg 1001 is a single auction NFT (22-for-2022) whose art is "an animated GIF of 32 unique bitGANs" — an 8×4 grid of rainbow ghost faces. Two blocks: shape-matching the 32 tiles to the 512 deck gave only weak, ambiguous scores (no reliable numbers), and — decisively — the lore frames these rainbow ghosts as a sneak-peek at the next drop's style, a separate generation that isn't part of the 512 the glitch classes index. So the intersection isn't even well-defined. No roster fabricated.
GAN number as a wordlist index (skull pieces)
re-derives dead register
Uses 32: look up bip39[GAN−1] — for #32 that gives word 32.
For the whole skull drop, bip39[GAN−1] already equals the piece's Essence by design. So on skulls this is just the already-dead Essence register in disguise. Stays alive only for non-skull pieces (listed under Remaining).
The piece at position 32 of each drop
dead per prior ruling
Uses 32: pick slot 32 inside every drop.
Already ruled out — position 32 is not a systematic anomaly slot (only 1 of 5 such pieces carries anything).
Split the Twin class 13→11 by a grid read
conflicts with a ruling
Uses 32: use the 32×32 split to prune two Twins.
Conflicts with the standing decision that Twin is context, not a glitch source right now. Parked unless that ruling changes.
Sort chosen words by wordlist-index mod 32
low-value ordering
Uses 32: a 32-based reordering of already-chosen words.
Re-orders by the words' own list positions; the generator itself rated it low, and essence orderings are already broadly dead. Not worth a slot on its own.

Resolved this session

9 now swept dead · 2 cannot stand alone

The former "remaining" list, run to completion. All nine testable ideas were swept plant-validated or computed directly, all negative. Two are un-testable in isolation by their own definition.

{#32, #95} as the missing 10th pair
pair verified · word-read dead
Uses 32: #32 is a "32" piece, burned in the same transaction as #95.
The burn pair is confirmed real and kept as a fact (D2's 10th-pair candidate). Its essence read, substituted as the 10th pair in the pairs-plus-anchor construction (first/last-speaks × two anchors × two orders), swept dead. The pair stands; its word-reading does not.
Mint-index pointer, base-32 radix, 64-page split, distance-from-32
swept dead
Uses 32: four number-registers — read bip39[mint-1]; two base-32 digits recombined; a page×slot split; the distance of each piece from 32.
Applied to the Fibonacci-eleven and HODL-eleven, both orderings, plant-validated. All dead.
32-row bitstring & direct pixel-coordinate read
swept dead
Uses 32: read one of the 32 rows as a 32-bit string, or use a piece number as an (row,col) address on its own 32×32 grid.
488 candidate word-sets across both ideas swept plant-validated, no hit. The reads are dominated by blank-border rows and flat-fill pixels — artifacts, not signal, exactly as the forensics predicted.
Base32-text alphabet decode
negative
Uses 32: map each number mod 32 through Base32's 32 symbols and read the string.
Fibonacci-eleven → "UY6AGKMQW67"; HODL → "ABCDFINVCXZ" (just Fibonacci-mod-32). No readable words.
Sixteen blocks of 32 — one member per block
fails
Uses 32: cut the 512 into 16 chunks of 32; hope a class puts one member per chunk.
Computed for every class: the glitches cluster in the robo drop (blocks 4-5 hold ~33 of 73), nowhere near one-per-block. No clean selector.
New 32-based reading orders
no standalone test
Uses 32: mod-32 sort, serpentine, bit-reversal, rotate-at-#32.
Pure re-orderings — and every register they would reorder (essence and the number-registers) is already dead, so a re-sort can't resurrect anything.
Literal "go to cell 32" self-read
not standalone
Uses 32: read cell 32 of #32's and #161's own art.
Two data points only — a tie-breaker for the (now-dead) pixel-diff idea, never an eleven.
The one structural keeper: across all 49 ideas, the cleanest real intersection is #129 ↔ #161 — both already in the Still-Eyes class, sitting exactly 32 apart, differing by exactly one binary digit (the 32-bit). It doesn't act as a selector on its own, but it's the tidiest evidence that 32 is a genuine structural axis in the robo drop, not just a theme.