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.