# Signature Dictionary — the measurable visual signature behind every name

Date: 2026-08-26. Charter: for every GANg, Type, and SubType value with at least three
members, define the measurable visual signature the name promises, state it in one plain
sentence, give the supporting statistics, and list every member that violates its own name.
Instruments: unified per-GAN feature table `sig_feats.json` (palette warm/cool/saturation,
hue spread and hue cycling, hue drift direction, motion energy and profile, luminance
flicker, foreground-area trajectory, loop seam, PC1 morph direction and monotonicity,
frame delays) computed from the pipeB 32×32 frame cache; per-value separation stats in
`sig_stats.json` (robust median/MAD z within family, rank AUC against the rest of the
corpus); cyclic-palindrome scores in `sig_cycpal.json`. Machine-readable output:
`signature_violations.json`. Traits keyed from `theories/family_tables.json` (the
properties CSV's Number column mixes octal and binary display-numbers and must not be
used as a key).

Method guardrails, honoring the false-positives-are-the-enemy rule: each signature is
calibrated on the family median (robust to its own violators); a violator needs robust
z beyond 2.5 toward the rest of the corpus AND has to cross the corpus median, or carry
independent corroboration; frame count and frame delay are DROP conventions and were
excluded from violator calling (the 43 "skullGAN frame-count violators" of the first pass
were just skullGANs living in later drops); every candidate was cross-checked against
`identity_corruption_hunt.md` and `confirmed_glitch_registry.md`; ambiguous cases went to
visual inspection, and the survivors of that went to the collector's eye (manifest at bottom).

---

## Axis 1 — GANg

- **skullGAN (n=167)** — skullGANs are warm-toned skull portraits whose look is set by
  drop and SubType, not by the GANg word itself; no GANg-level color or motion signature
  survives drop control. No violator.
- **bitGAN (n=76)** — bitGANs are mildly desaturated pixel-figures with no signature
  beyond drop style; every apparent outlier (sat AUC 0.81 tail) is a SubType=Inverted
  piece, explained. No violator.
- **ghostGAN (n=64)** — ghostGANs are the palest family in the corpus (median value
  0.812, AUC 0.81); the only decisively dark member is #511 (val 0.767, light-on-dark),
  the the collector-confirmed corrupt Inverted. Violator: #511 (strong, same finding as the
  Inverted axis).
- **testGAN (n=39)** — testGANs are ordinary ghost-drop figures over a TEST-pattern
  backdrop; the "testing" is in the background, not the behavior. Tester flicker
  outlier #311 was cleared visually: Frozen-palette testers all swing through blues.
- **monsterGAN (n=30)** — monsterGANs are heavy solid bodies (colored fraction median
  0.56 vs 0.40, AUC 0.86); the pale tail {470, 471, 472, 498, 268} is review/solstice
  drop styling shared by every neighbor, not corruption. No violator.
- **cryptoGAN (n=21)** — no signature of its own; the word marks robo- and review-drop
  variety pieces whose look follows their SubType.
- **spookyGAN (n=18)** — wide-hue Halloween figures (hue spread AUC 0.81: clowns,
  pirates, pumpkins); the narrow-palette members are exactly the Flaming-labelled ones,
  so the exception is lawful. No violator.
- **spaceGAN (n=16)** — no measurable signature against its ghost-drop siblings.
- **nagaGAN (n=8)** — **the palindrome family, and the corpus's one clean TEMPORAL
  GANg signature**: palindromic names (Dr Awkward, Taco Cat…), the eight palindromic
  6-bit codes, palindromic SubType words (Deified, Dixid, Dioid, Sus, Tenet), and
  palindromic ANIMATIONS. Cyclic-palindrome score (best mirror axis over all circular
  shifts, normalized): #129 RoboboR, #141 Dr Awkward, #174 Evil Olive, #192 CalcuclaC
  = 1.000 (perfect, pixel-exact out-and-back loops); #162 Star Rats 0.67, #180
  Flow Wolf 0.63, #147 Taco Cat 0.55, #159 Dog God 0.53 (partial); the highest
  non-naga score in all 512 is 0.34. The 4-perfect/4-partial split does not follow
  any obvious code property; recorded as an open sub-pattern. No violator — every
  naga leads the corpus.
- **glitchGAN (n=8)** — glitchGANs carry violent frame-level glitch events: motion
  spike z 14–38 and flicker bursts in 7 of 8 members (median motion 0.13 vs 0.06).
  Violator: **#281 Entangled glitchGAN (ess `book`)** — motion 0.055, spike z 3.3,
  the only glitchGAN whose animation contains no glitch event; its glitchiness is a
  static scanline texture. NEW, medium, escalated (Q3).
- **roboGAN (n=40)** — no unified signature; the robo drop is a zoo of bespoke
  per-piece animation programs (speed11's five robo members are a delay-channel
  phenomenon, already canon).
- **cronenGAN (n=6)** — dense saturated flesh masses (sat 0.68–0.73); #507 sits mildly
  low on sat and colored — its inversion candidacy was already killed by the strict
  trio-style test; weak note only.
- **cryptoArtGAN (n=8)** — the artist tributes; the only family defined by frame count
  (60/120 vs 55) and tribute pacing — all deliberate design. No violator.
- **cyberGAN (n=3)** — the three b0rgs: sparse warm line-figures, lowest colored
  fraction in the robo drop, Eyes=0 blind puns. Family too small to grade violators.

## Axis 2 — Type

- **Skull (n=75)** — skull-drop portrait heads; their high front-half motion is the
  41-frame drop pacing, not a Type behavior. No violator.
- **Ghost (n=105)** — pale and slow (value AUC 0.72, motion AUC 0.28); the corpus's
  quiet figures. No violator.
- **Glitch (n=67)** — moves hard and flickers (motion AUC 0.92, flicker AUC 0.90, the
  crispest Type signature). The motionless minority (#102, #115, #211, #25, #9…) are
  the skull-drop's authored "lost the ability to move" glitch mode from the journal's
  four failure classes — stillness there is lore-consistent, NOT a violation. The #47
  reclassification (glitches in a cut dialect) from visual_spotting_B stands.
- **Bit (n=53)** — mildly desaturated; outliers all SubType-explained. No violator.
- **Tester (n=39)** — steady, low-flicker (AUC 0.19); swings belong to Frozen palettes.
- **Bot (n=27)** — no unified signature; bespoke robo programs.
- **Monster (n=27)** — big saturated solid bodies (colored AUC 0.80) that hold their
  mass through the loop. No violator.
- **Shimmer (n=18)** — **a temporal signature: the tremble.** One dominant principal
  component (pc1_var AUC 0.82) with tiny front-motion — oscillation in place rather
  than a journey. The two robo-drop "Shimmering cryptoGANs" (#143, #168) do not share
  the skull-drop tremble; that is drop styling, recorded not graded.
- **Cloud (n=18)** — full-width soft masses (aspect AUC 0.71) drifting bright-ward
  (pc1 slope AUC 0.65). No violator.
- **Smoke (n=17)** — thins out: low colored fraction, deep area minimum; fading is the
  design. Weak note: #254 Glowing Smoke bitGAN never thins (area floor 0.556,
  corpus-normal persistence) — the least smoke-like Smoke, below violation threshold.
- **Cronenberg (n=16)** — dense flesh masses, slightly darker/more saturated; soft.
- **Invader (n=16)** — saturated podGAN figures (sat AUC 0.73), steady hue. Clean.
- **Artist (n=7)** — the tributes: bespoke frame counts and the corpus's slowest
  motion; #228's 120 frames and #446's 88 are designed. #282's extra motion is a
  video-remix-artist tribute (Eclectic Method), authored. No violator.
- **Cyborg (n=6), Twins (n=3), Reifier (n=3), Zombie (n=4)** — small families:
  Cyborg no signature; Twins paired figures on darker grounds; Reifier pulses its
  mass into being (area_osc AUC 0.89); Zombie wide-palette decay. No violators graded
  at these sizes.

## Axis 3 — SubType (the trait the names put in front of the buyer)

- **Rainbow (n=82)** — "Rainbow pieces spread their hue within a frame or cycle it
  across frames" (rainbowness = max(hue_spread, hue_cycle); family median 0.50, MAD
  0.157; AUC 0.76 both components). A Rainbow with neither is wearing the wrong art —
  the instrument independently recovers confirmed-corrupt #14 at z −2.5.
  Violators, worst first: **#179 (z −2.8, strong, known)**, **#14 (z −2.5, strong,
  CONFIRMED byte-identical to Flaming #75)**, **#423 (z −2.0, monochrome blue, strong,
  known)**, #92 (z −2.0, flame art, medium, known), #166 (z −1.9, medium, known),
  **#140 (z −1.8, solid red robot, NEW, Q1)**, **#66 (z −1.8, sepia/flame skull, NEW,
  Q1 — and #66 already carries the chain-vs-Arweave SubType discrepancy)**, #96
  (z −1.7, weak, known), **#56 (z −1.7, flame-toned, NEW, Q1)**.
- **Flaming (n=58)** — "Flaming pieces are warm-palette: fire colors dominate the
  colored pixels" (median warm 0.855, AUC 0.88; flicker common but not required).
  Violators: **#204 Angry Clown (warm 0.436, rainbow clown, strong)**, #150 Firebot
  (0.468, body cycles red→green→blue→purple, medium, known label-swap partner of
  #179), #490 (0.494, rainbow skull, medium, known), **#468 Flaming testGAN (0.590,
  z −2.8, multicolor art, PLUS a speed11 delay break — NEW, medium, Q2)**. #24 was
  cleared on inspection (flame head; a blue shirt drags the metric); #116, #186 weak.
- **Frozen (n=31)** — "Frozen pieces are cool-palette and calm" (median cool 0.84,
  AUC 0.95 — the crispest color signature in the corpus). Violators: **#231 (cool
  0.472 + 3× family motion + mutating body — triple contradiction, strong, known)**,
  #491 (motion z +4.3, medium, known); weak: #232, #326, #343.
- **Inverted (n=17)** — "Inverted pieces sit on dark grounds, light-on-dark — a
  photographic negative of the corpus norm" (median bg lightness 0.25 vs 0.84, AUC
  0.96). Violators: **#511 (bg 0.847, essence `negative`, strong, confirmed)**,
  **#509 (bg 0.751, strong, confirmed)**, #479 (bg 0.587, medium — already a
  confirmed timing-glitch piece), #505 (weak).
- **Glowing (n=32)** — "Glowing pieces are highly saturated with a large colored
  fraction" (sat AUC 0.80). Warmth is a tendency (29/32) but NOT the signature: the
  three cool-metric members #429/#380/#371 are saturated magenta/green glows —
  inspected and cleared (the warm/cool bands simply exclude magenta). No violator.
- **Slender (n=36)** — thin-and-fade tendencies (low colored fraction, deep area
  minimum, narrow aspect) with wide variance; the only decisive member is **#478**
  (Rainbow bitGAN titled, colorful art, Slender label — pure metadata contradiction,
  medium, known).
- **Silver (n=11)** — "Silver pieces are near-colorless" (colored fraction ≤0.33 for
  every member, AUC 0.06; low sat). Weak: #81 (most saturated Silver, known flag).
- **Ultra (n=11)** — "Ultra pieces are solid, saturated, permanent: the highest area
  floor in the corpus — they never thin, fade, or vanish" (median min-area 0.75, AUC
  0.85). #430's warm phase inspected and cleared (still solid and saturated). Clean.
- **Vivid (n=26)** — modestly elevated saturation (AUC 0.72), nothing else; the
  weakest color name. No violator.
- **Primordial (n=29)** — slightly darker, higher-motion ancients; the name marks lore
  seniority more than look. No violator.
- **Ethereal (n=20)** — saturated (AUC 0.76) with a mild forward hue-rotation habit
  (median +0.22°/frame); #233 rotates backward (z −2.8, weak note).
- **Nebulous (n=19)** — elevated back-half motion (AUC 0.71), loose family; #276
  direction outlier (weak note).
- **Cyber (n=22), Binary (n=14), Invasive (n=8)** — no measurable visual signature;
  these words track drop styling or number-face designs, not animation behavior.
- **Crypto (n=14)** — the calm tributes: lowest motion of any SubType (median 0.044);
  #282's motion is authored (Eclectic Method). No violator.
- **Dancing (n=15)** — **negative finding: Dancing does not dance.** Dancing pieces
  move slightly LESS than the corpus (motion AUC ~0.33) and lean against their drop's
  direction (11/15 negative drop-relative). The name is not a motion grammar.
- **Entangled (n=16)** — full-width tendency (aspect 1.0), no color/motion/direction
  grammar. No decisive signature, no violator.
- **Twin (n=10)** — **a temporal signature: Twins animate against the current.** Wide
  two-figure compositions (aspect AUC 0.86) whose morph runs AGAINST the drop's
  consensus direction: 9 of 10 have negative drop-relative direction (−0.36 to −1.27).
  Violator: **#395 Twin ghostGAN (ess `position`) at +0.17 (family mono z +3.8) — the
  lone Twin animating WITH its drop. NEW, medium, escalated (Q4).** Note the essence:
  a reversed-`position` piece.
- **Waking (n=9)** — two motion programs (grow-while-dimming: #305/#366/#443; vs
  shrink-while-brightening: the other six) and no unified grammar; whether one program
  is "waking" and the other "asleep" is escalated (Q5).
- **Small families (Bloody 5, Shy 4, Deified 4, Shifty 3, Amorphous 3)** — signatures
  stated in `signature_violations.json`; no violators graded at these sizes (weak
  notes: #381 least-red Bloody, #340 deepest-fading Shy).

---

## Cross-axis section — names whose signature is TEMPORAL (direction / loop / timing), not color

1. **nagaGAN = loop shape.** The one name whose entire identity is a loop grammar:
   palindrome in name, code, SubType word, and animation (4 perfect / 4 partial;
   nothing else in the corpus above 0.34). New instrument: cyclic-palindrome score
   (mirror axis searched over all circular shifts), `sig_cycpal.json`.
2. **Twin = counter-current morph direction** (9/10 negative drop-relative direction;
   #395 the reversal). Direction here is part of the name's signature — the collector's
   question answered YES for Twin.
3. **Shimmer = oscillation shape** (tremble in place: one dominant PC, low net
   travel) — a motion-profile signature, skull-drop dialect.
4. **Crypto = stillness** (tribute pacing, lowest motion of any SubType).
5. **Drop-level direction conventions (new structural finding):** animation direction
   is a DROP signature before it is a name signature — ghost drop emerges (median
   direction-monotonicity +0.51), solstice descends (−0.47), spooky +0.35, review
   +0.34 but bimodal, skull/robo no consensus. The Review drop splits roughly in half
   (36 forward / 31 reverse by sign) — this widens D11's "reversal cluster"
   {457,458,459,462,463,497} into a drop-wide two-current structure; which half is
   "in review" is a metadata question, not a violation.
6. **Dancing, Waking, Entangled = NO motion grammar** (the honest negatives to
   the collector's direct question; Waking's two-program split escalated).
7. **Timing/delay channel:** unchanged from canon — speed11 (uniform per-drop delay
   breakers) and the held-frame pair #479/#480 remain the only delay signatures;
   delay is otherwise a drop convention (skull 0ms, ghost/solstice/review 100ms
   modal, robo mixed).

## Violator registry (graded, deduplicated across axes)

STRONG — the piece decisively contradicts its own name (7):
| # | piece | ess | axis broken | evidence |
|---|---|---|---|---|
| 14 | Rainbow skullGAN | accuse | Rainbow | no rainbow; byte-identical to Flaming #75 (CONFIRMED) |
| 179 | Ghostbot roboGAN | under | Rainbow | colorless ghost art, rainbowness z −2.8 |
| 423 | Rainbow skullGAN (123/144) | cruel | Rainbow | monochrome blue-teal every frame |
| 204 | Angry Clown spookyGAN | grab | Flaming | rainbow-wigged multicolor clown |
| 231 | Frozen glitchGAN | base | Frozen | warmest Frozen + 3× motion + mutating body |
| 511 | Inverted ghostGAN | negative | Inverted (+ghostGAN) | near-white ground (confirmed corrupt) |
| 509 | Dark Rainbow | network | Inverted | light ground (confirmed corrupt) |

MEDIUM (10): #92 `arena` (Rainbow, flame art), #166 `pink` (Rainbow, bicolor),
#150 `fire` (Flaming, color-cycling), #490 `knee` (Flaming, rainbow skull),
#468 `random` (Flaming, multicolor + speed11 — NEW), #491 `weapon` (Frozen, motion),
#479 `song` (Inverted, mid-gray ground; confirmed timing-glitch), #478 `recycle`
(Slender label on a Rainbow piece), #395 `position` (Twin, reversed direction — NEW,
temporal), #281 `book` (glitchGAN, no glitch — NEW).

WEAK / recorded (11): #140, #66, #56, #96 (Rainbow tail — first three NEW, escalated),
#116, #186 (Flaming), #232, #326, #343 (Frozen), #505 (Inverted), #81 (Silver); plus
notes #233, #276, #254, #381, #340.

Count check the collector asked for: strong = 7, strong+medium = 17. The strong seven do NOT
make eleven; the eleven-shaped cut is strong + the four label-contradiction mediums
{92 or 490, 150, 478, 468} — i.e. the same neighborhood identity_corruption_hunt
reached, now with #468/#395/#281 as live alternates to its #390 (whose double-mint
violation is structural, not visual — #390 is a perfectly conforming Frozen to the eye).
Mint-order essences of the strong seven: `accuse under grab base cruel network negative`
(14, 179, 204, 231, 423, 509, 511). All BIP-39 valid.

---

## FOR collector'S EYE (files in `intruders/signature_review/`)

- **Q1_rainbow_tail_56_66_140.png** — Do #56, #66, #140 read as Rainbow to you, or as
  flame/monochrome art wearing a Rainbow label? (references: confirmed #14, graded
  #92/#96 in the same strip)
- **Q2_flaming468_multicolor.png** + **Q2_gif_468_flaming_testgan.gif** — Does #468
  read as Flaming? It is the coldest unflagged Flaming AND a speed11 delay-breaker.
- **Q3_glitchgan281_no_glitch.png** + **Q3_gif_281_entangled_glitchgan.gif** (+
  reference **Q3_gif_244**) — Does #281 visibly glitch when playing, or is it a
  glitchGAN that does not glitch (its scanlines are a static texture)?
- **Q4_twin395_reversed.png** + **Q4_gif_395_twin_ghostGAN_direction.gif** (+
  reference **Q4_gif_379**) — 9 of 10 Twins animate against their drop's direction;
  does #395 play its morph the opposite way to the other Twins?
- **Q5_waking_grammar.png** — Waking splits into grow-while-dimming (#305/#366/#443)
  vs shrink-while-brightening (the other six). Is one of these the "waking" grammar,
  or does Waking carry no motion signature?

## collector'S VERDICTS ON THE FIVE ESCALATIONS (2026-08-26)
- Q1 Rainbow tail #56/#66/#140: CONFIRMED Rainbow (not violators). Remove from violator list.
- Q2 #468 Flaming: CONFIRMED acceptable — multicolor but enough red that Flaming holds. NOT a violation. (Note: #468's essence is `random`.)
- Q3 #281 glitchGAN: CONFIRMED does NOT glitch to the eye. A glitchGAN with no glitch — strong anomaly, the mirror of the #47 case (a non-glitch-labeled piece that DOES glitch). #281 is a live selection candidate.
- Q4 #395 Twin direction: UNRESOLVED — the collector can't evaluate direction without a comparison. Needs a side-by-side against a canonical Twin. Held.
- Q5 Waking grammar: UNRESOLVED by eye. the collector's guidance: (a) look to the FIRST instance of the Waking naming convention to establish canonical behavior; (b) "Waking" likely references the film "Waking Life" (rotoscoped, dreamlike), which Pindar cites. Re-frame the Waking signature around that reference, first-instance-canonical.

## TWIN SIGNATURE REFINED (the collector, 2026-08-26)
Twin = TWO FACES / GANs in one frame, Siamese-attached (not the counter-current motion the
cartographer measured — that was a secondary correlate). The real signature is compositional:
two heads in the frame. This makes the interesting violators the pieces that ARE two-faced but
are NOT named Twin.
- **#192 CalcuclaC (111111), ess `zero`:** confirmed two-faced (two eye-pairs side by side) but
  not called Twin. Instead its twinning is encoded by a PALINDROME NAME (CalcuclaC reads the
  same both ways), SubType "Tenet" (the palindrome film), Type Reifier, binary 111111=63, the
  LAST robo. A twin that hides its twinning in wordplay rather than the label.
TODO (cheap): a two-face census across all 512 (bilateral/two-blob detection) to find every
unnamed twin — candidates for the same name-vs-reality class.

## THE UNIFYING THEME (name vs reality mismatches, both directions)
Today's eye-confirmed anomalies all fit one frame: a piece IS one thing but is CALLED another.
- #14 named Rainbow, wears Flaming art (wrong art)
- #281 named glitchGAN, does not glitch (name over-claims)
- #47 not named glitchGAN, DOES glitch (name under-claims)
- #192 IS a twin, not named Twin (name hides it in a palindrome)
"Looking for glitches all wrong" may mean: the glitch is the MISMATCH between what a piece is
and what its metadata says, in either direction — not a visual event in the animation.

## THE 8 nagaGANs ARE TRIPLE PALINDROMES (confirmed 2026-08-26)
All 8 palindrome-named robos are palindromic on THREE axes at once:
  #129 RoboboR/000000/alpha, #141 Dr Awkward/001100/awkward, #147 Taco Cat/010010/cat,
  #159 Dog God/011110/dog, #162 Star Rats/100001/TWIN, #174 Evil Olive/101101/olive,
  #180 Flow Wolf/110011/wolf, #192 CalcuclaC/111111/zero.
- palindrome NAME, palindrome 6-BIT BINARY code (exactly the 8 six-bit palindromes exist, and
  these are them), palindrome MIRROR-LOOP animation (nagaGAN = anagram of… reversed).
- #162 Star Rats' essence is literally `twin`. #180 Flow Wolf is also a 0ms wrong-speed piece.
- Refines the per-drop medium map: the robo drop's glitch medium is PALINDROME/MIRROR, deeper
  than "born leet." 8 pieces, not 11 — a clean contributor set, not the whole answer.

## EMERGING SELECTION PRINCIPLE (from the collector's eye-findings, 2026-08-26)
The glitch may be a NAME-TRANSFORMATION: the piece's true nature is hidden in its name via a
reversible wordplay operation, one operation per piece:
  - PALINDROME: CalcuclaC, the 8 nagaGANs
  - LEET substitution: G1itch1ng, err0r (#191)
  - TRANSPOSITION typo: Primoridal (#16/#100), Cronenber (#123)
  - the wrong-art / wrong-behavior mismatches (#14, #281, #47)
"Looking all wrong" = read the NAME's transformation, not the animation. This unifies the leet
eleven, the palindromes, and the chain-name corruptions under one grammar. Untested as a single
selection rule (the pieces span drops — good for breadth).

## Q4/Q5 RESOLVED (measured, 2026-08-26)
- Q4 #395: CONFIRMED displaced, per the collector's eye. Loop restart is normal (frame 54, like all
  ghostGANs) but the MAX motion discontinuity is at frame 34 with jump 13.5 — nearly double any
  Twin sibling (#323: 6.9 @ f41). Its interleaving/motion event sits mid-file in a different
  place than its twins. Real structural anomaly. #395 stays a live Twin-direction candidate.
- Q5 Waking: no localized speedup above noise (all peaks 1.5-2.2x mean vs 3x+ for a real
  stutter). The perceived speed-up is the color illusion the collector suspected. Waking signature
  stays unresolved as a temporal grammar; not a glitch axis.

## DISCONTINUITY MAP — all 512 (seammap.json, 2026-08-26, the collector's request)
Per piece: where its largest frame-to-frame motion jump sits (as a fraction of the loop), its
magnitude, the wrap-seam smoothness, and spike count. DETECTOR VALIDATED — it independently
recovers the known glitchers: #47 (z=3.2, confirms the cut-dialect reclassification), splice
skulls #73/#97/#103/#113, the negative trio #231/256/274 (z=5-6, wrap distance ~110 = the
inverted frame), #479 (z=8.0). Per-drop median maxjump POSITION: skull 0.68, robo 0.66,
spooky 0.46, sol 0.58, ghost 0.56, review 0.51 (each drop has a characteristic loop structure).
Displaced/outsized outliers by drop are listed in seammap.log. NEW candidates not in any prior
set:
  - **#409 (ghost): z=6.2, jump displaced to frac 0.76** — the strongest new discontinuity find.
  - #311 (ghost, z=2.6), and among skulls #29/#41/#39 (large jumps at off-median positions).
#395 (the Q4 Twin) is mildly displaced but NOT a collection-wide outlier — real but modest.
TODO: the collector's eye on #409's animation; add seam-position as a temporal signature axis.

## collector'S EYE ON THE SEAM CANDIDATES + TWO NEW EFFECTS (2026-08-26)
- #409 skullGAN: dismissed — just a color splotch that explodes on the right before the loop
  ends. Not a glitch. (The seam detector fired on a large-but-authored color event.)
- #311 testGAN: NOT a discontinuity glitch but an instance of a RECURRING EFFECT worth mapping:
  the WHOLE GAN + BACKGROUND flash one uniform color at once (here light blue), then jump to
  another (purple). Seen elsewhere — some roboGANs do it as fast rainbow-cycling; the Cyber
  cryptoGAN (robo drop) does it. Detector built: flashmap.json (whole-frame color-uniformity
  collapse + fast hue-cycling). Results pending in flashmap.log.

## NEW SINGLETON CANDIDATE — #161 Cyber skullGAN (100000), ess `news` (the collector, never discussed)
Claim: possibly the ONLY GAN whose EYES never move. STRONGLY CORROBORATED by the motion heatmap
(per-pixel temporal std): the two eyes are the darkest (most static) regions in the frame,
holding rectangular eye-shape while the entire face and background churn. binary 100000 = 32
(the promise-piece number that recurs: reneGANs "Cyber ... 32", the #32 HiddenType thread).
Uniqueness test (static-region ranking across all 512) running. If unique, this is a clean new
singleton in the same name-vs-reality family: a piece defined by what DOESN'T move.

## FLASH DETECTOR v1 FAILED — recalibrating from #311 (the collector's validation, 2026-08-26)
the collector tagged all 24 candidates: ZERO true whole-frame flashes. 14 rainbow-cycling
(27,63,74,77,97,109,140,146,147,172,180,190,458,474), 9 neither
(25,121,179,198,210,256,257,259,479). The v1 metric (spatial-variance collapse) caught
rainbow-cycling and partial desaturation, NOT the specific effect. #311 (the motivating piece)
didn't even rank. TRUE effect per the collector: the ENTIRE GAN + background become ONE uniform color
for a frame, then JUMP to another uniform color (blue -> purple). Recalibrating v2 from #311's
actual uniform frames.

## FLASH DETECTOR — PARKED as eye-only (2026-08-26)
Four metrics tried to isolate the #311 whole-frame-flash effect: (1) spatial-variance collapse
= caught rainbow-cycling, not the effect (the collector tagged all 24 v1 candidates as rainbow/neither,
zero flashes); (2) #311 never goes spatially uniform (std 55-73); (3) background-color variance:
#311=8.3 barely above rainbows ~6.8; (4) fg-bg color SYNC: #311=0.39, LOWER than rainbows (0.8+).
None separate it. Conclusion: the effect is either perceptually subtle beyond global color stats,
or a sibling of the rainbow-cycling family. PARKED — flag pieces by eye if needed, no detector.
the collector's validated rainbow-cyclers (a real class, for the record): 27,63,74,77,97,109,140,146,
147,172,180,190,458,474. Confirmed non-effect: 25,121,179,198,210,256,257,259,479.

## nagaGAN CONVENTION-BREAKS (the collector's question, 2026-08-26)
The 8 nagaGANs are palindromes on FIVE axes: Name, binary code, SubType (Deified/Dioid/Dixid/
Sus/Tenet — all palindromes), Type (Reifier/Kook/Rotator/Pup/Reviver — all palindromes), AND
usually the ANIMATION (mirror-loop: frame k = frame N-1-k). The animation axis is where two break:
  - PERFECT mirror-loops: #141 Dr Awkward (err 0.0), #159 Dog God (0.0), #129 RoboboR (6.7)
  - degrading: #180 (14.1), #174 (19.8), #162 Star Rats (22.8, ess 'twin')
  - BREAK — no mirror at any center: #147 Taco Cat (30.2) and #192 CalcuclaC (30.6), verified
    against offset/nested mirror tests (both ~30, vs control #141 at 0.0). The "Taco Cat nested
    palindrome" claim does NOT hold at 32x32.
FINDING: #147 and #192 are palindrome in every metadata way but their animations do NOT mirror —
a naga-that-isn't-naga, a name-vs-reality mismatch. #192 CalcuclaC now carries THREE anomalies:
twin-not-named-Twin + binary 111111=63 (last robo) + broken mirror. Strong watch piece.

## SHIMMER SIGNATURE CORRECTED (the collector, 2026-08-26) + glitcher link
"Shimmer" is a middle-word title descriptor (Primordial/Flaming/Nebulous/Rainbow Shimmer), 14
pieces. OLD signature "trembles in place" was crude. TRUE signature (measured): SHAPES BARELY
MOVE while COLORS SHIFT — shape-motion 2.7-4.9 for pure Shimmers vs 7.5-12.5 for Rainbow-Glitch
skulls; color/shape ratio ~2.0-2.7. Light playing over a still surface. the collector: "colors blend
and move qualitatively differently from other Rainbow or Glitch skulls" — confirmed.
KEY LINK: all four GROUND-TRUTH splice glitchers {7,43,67,71} are Shimmer-titled, and they sit
at HIGH shape-motion for Shimmers (6.7-8.8) — the splice frame jolts the shape that should stay
still. So the skull glitch = A SHIMMER WHOSE SHAPE-STILLNESS IS BROKEN BY A SPLICE. Cleaner
definition than we had. Purest Shimmers (cleanest still-shape): #24,35,68,70,104,117,118,128.
Note #143 "Shimmering cryptoGAN" (ess 'problem', a montage no-match) and #168 (ess 'hidden').

## nagaGAN CORRECTION (2026-08-26, retracts the "convention-break" above)
The earlier "#147/#192 break the mirror" was a BUGGED test (mirror-center search didn't cover
off-center axes). Full search: BOTH mirror, both run forward-then-reverse (the collector's eye correct).
  - #147 Taco Cat: SINGLE mirror-loop, err 0.0, reversal OFF-CENTER at ~frame 35 of 60.
  - #192 CalcuclaC: DOUBLE mirror-loop — two reversal axes (~frames 9 and 31), so forward-reverse
    TWICE per cycle (partner pairs f0-f18, f20-f42 both err 0.0 about the two centers). N=44.
NEITHER breaks the convention. CalcuclaC is the MAXIMAL naga: palindrome name + binary 111111
(all ones) + SubType + Type + DOUBLE-palindrome animation + twin (two faces), ess `zero`, last
robo. It's the exemplar, not a defect. The 8 nagaGANs = the robo drop's palindrome/mirror
glitch-medium (per the per-drop map); robo's slot in any combined-drop eleven comes from them.

## collector'S COMPLETE 512-PIECE VISUAL NOTES (2026-08-26) — michael_visual_notes.json
The full human eye across the collection, categorized in collector_categories.json. Key classes:
- color_flash (6, DEFINITIVE - the effect no detector could isolate): 130,133,462,463,464,497.
  Mechanism per the collector: GAN sits BEHIND the binary bg, binary morphs to SINE WAVES during flash.
  #130 qubitGAN + #133 = "close cousins".
- color_speedup (16): 147,192,462,463,464,485,486,487,488,505,506,507,508,509,510,511.
- inverted_glitch (4, = our negative-frame set exactly): 231,256,274,479. #231 first inverted,
  #256 first DOUBLE-inverted (piece already black, glitch frame white), #479 first beige/black morph.
- sits_behind_binary (18, robo class): 129,130,133,138,141,143,148,153,158,161,163,168,173,174,178,183,185,188.
- larger_eyes (13): 41,43,48,66,89,97,109,449,450,451,452,455,456.
- morai_echo (~10-11, COUNT SOFT per the collector - faint echoes may be unstated): 11,12,43,44,45,73,78,100,117,123,454.
  #1 Moirai = the triplet template; echoes = ghost extra-heads. #454 = "first FOUR not three" (review capstone).
- ghost-drop BACKGROUND taxonomy: binary(65)/test-written(38)/space-invader(34)/skull(28)/podgan(16).
- HIDDEN NUMBERS in art (no scan catches these): #165 "8941 visible sequentially", #207 "#4 on its chest".
- unique_eyes: #147 ONLY Noun eyes, #161 static green-rimmed (=our still-eyes CONFIRMED), #187 first ghost-eye
  structure, #467 oval w/ large pupils. Eye-type mixing: #187/#191/#206 blend skull/robo/ghost eyes.
- FIRSTS (anchor markers): #217 first Entangled(=twin), #231/#256 inverted firsts, #454 four-not-three,
  #479 first beige, #207 precursor-to-solstice (#4 on chest), #501-504 prelude-to-QuantumGlitch(#512).
- #151 essence non-BIP39, a Bitcoin reference (satoshi). #357 color-gradient-off (CONFIRMED his original obs).
- #209 Solar / #300 Lunar = day-night positive-negative PAIR. #175 = roboGAN/skullGAN feature mixture.

## COLOR-FLASH CORRECTION (the collector, 2026-08-26)
The sine-wave binary morph is NOT universal to color-flash — it's a SUB-TYPE:
- SINE-WAVE color-flash (GAN behind binary, binary warps to waves): #130 qubitGAN, #133 (cousins).
- PLAIN color-flash (whole frame flashes color, binary pattern stays intact): #462,463,464,497.
So "color flash" (6: 130,133,462,463,464,497) splits into wave-morph {130,133} vs plain {462,463,464,497}.
The earlier "binary→sine waves" mechanism applies only to the {130,133} sub-type.

## GHOST COLOR-GRADIENT — CLOSED as too noisy (the collector's review, 2026-08-26)
the collector reviewed the full-size gallery: the automated hue dots frequently DON'T match the images
(GAN art too messy for reliable hue), and his own eye finds most flags borderline. Verdict: the
gradient exists loosely but is NOT a clean readable channel; no reliable eleven. Conviction
standouts (noise-level, not a set): #303 (too purple, should be orange/red), #347, #357, #371
(labeled green but looks fully orange). #309/#310 possible swap; #320/#319 possible order issue.
NOT pursuing further. The color-flash class (from the collector's notes, definitive) remains separate/valid.

## AMENDMENT (the collector, 2026-08-26): ghost gradient NOT closed
Reopened. Pindar said the puzzle is "hard even knowing how to solve it" — so the gradient being
noisy/hard-to-read is EXPECTED, not disqualifying. Pursuing with MEDIAN COMPOSITES (stable image
per GAN, no animation flicker) and the collector's eye leading; automated hue dots dropped as unreliable.

## FULL INTEGRATION OF collector'S 512 NOTES (2026-08-26) — catalog + all axes
Full catalog: `theories/glitch_type_catalog.md`; structured analysis: `notes_analysis.md`;
machine-readable: `glitch_types.json` (`glitch_types`=29 types + `axes`=7 cross-drop/identity axes).
Re-parsed line-by-line from `michael_visual_notes.json` (first-pass `collector_categories.json` corrected).
CSV mapping pinned: the collector ID N = `bitgan_properties.csv` row index N (row 0 = piece #0 Pandimensional).

CORRECTIONS to earlier notes:
- standard_glitch is 46 (added #7 splice+shimmer, #11 frame-insertion).
- larger_eyes is 15 not 13 (added #18, #37).
- no_glitch is 19 (added #2 "doesn't actually glitch").
- NEW clean class **naga_palindrome** (8): {129,141,147,159,162,174,180,192} = palindrome-titled
  roboGANs (RoboboR, Taco Cat, Dog God, Evil Olive...). "naga" is itself a palindrome.
- lines split: **line_background** {485,486,487,488} vs **scanline_removal** {489,491}.
- NEW: merge_convergence(#50 two GANs merge into one), robo_skull_mixture(#175 Rainbot skullGan),
  day_night_pair(#209 Solar/#300 Lunar), beige_background(#479/#480 = confirmed timing pair),
  nonbip_bitcoin_essence(#151 ess=satoshi, CSV-confirmed), color_gradient_off(#357 anchor),
  quantum_glitch_capstone(#512 anchor), spooky_ghost_eyes(#447), twin_entangled, figure_resemblance.

NEAR-11 classes: morai_echo=11 (FUZZY, soft) and costume_spooky=10 (CLEAN). All others miss 11.

NEW AXES (beyond glitch mechanics):
- WRONG-DROP/misplaced identity: #175,#187,#191,#206,#193,#200,#207,#447,#506,#302.
- EYE-STRUCTURE: skull(empty,cut-corner)/robo(square+pupil)/ghost(square,no-pupil)/noun(#147 only)/
  oval(#467,473-476). Mixers #187/#191/#206. Wrong-drop eyes #447(spooky ghost on artist),#506(robo on bitGAN).
- BACKGROUND firsts: beige #479, lines #485; scanline #489/#491.
- RELATION EDGES: 207→Solstice(#4-chest); 501-504→512(prelude); 209↔300(day-night pos/neg);
  130↔133(cousins); 205↔208; 194↔203; 217→512(Entangled=twin/quantum start).
- HIDDEN NUMBERS #165 "8941", #207 "#4"; NON-BIP #151 satoshi.

CONFIRMED-SET ties: the collector's inverted_glitch {231,256,274,479} = our negative-frame class;
beige pair {479,480} = confirmed timing pair; #357 & #512 anchors intact; #281 in his no-glitch list
(= confirmed 1-of-8 non-glitcher); #47 reads as glitching (consistent with hard-cut re-class).

## LENS COMPLEMENTARITY + #281 DUAL VERDICT (2026-08-26, w/ Theory Iterator)
Two axes, keep separate: collector_categories/catalog = VISUAL-EFFECT axis; confirmed_glitch_registry
= STRUCTURAL/NAME axis. Canon structural/name glitches (8 nagas, #14/#75 identity, spooky case/count,
hidden numbers) mostly do NOT appear as visual categories; catalog clean types naga_palindrome,
color_gradient_off(#357), quantum_glitch_capstone(#512) bridge both. A solver must UNION them.
#281 DUAL VERDICT: no_glitch(clean=True) on the visual axis, but IS the 1-of-8 name-lens anomaly
(glitchGAN that doesn't glitch) — do not read its clean flag as "uninteresting", do not double-count.
Iterator reconciliation: ZERO dim conflicts (glitch_types.json vs family_tables.json); all flagged
essences BIP-39 except #151(satoshi); Dim7 slot {487,507} RESOLVED = both color_speedup.
Mint-index xref: theories/collector_categories_xref.{md,json}; #512=mint 662=last (confirms capstone);
498/512 have mint_index, 14 missing.