[{"id":1,"num":1,"title":"The word 'Glitch' in the label","picked":"We took every artwork whose label literally says Glitch. Three ways of counting: 67 have a Type field reading 'Glitch', 8 belong to the group 'glitchGAN', and 10 have 'glitchGAN' in the title.","tested":"Where a group already had 11 members we used it whole. Otherwise we tried every possible group of 11 picked from it. We ran each in artwork-number order, low to high and high to low.","happened":"Ruled out. None of these ways of counting gives 11. Michael later confirmed that when Pindar said we were 'looking in the wrong place', he almost certainly meant exactly this. RULED OUT BY LATER WORK: the rebuilt detector shows the label and the real glitching barely overlap at all.","status":"ruled-out","links":"[]","media":"[]","sort":0},{"id":2,"num":2,"title":"Numbers swapped in for letters","picked":"Titles that put a 1 where an i should be, or a 0 where an o should be: G1itch1ng, G1itchb0t, B1-b0rg, S1-b0rg. 13 artworks.","tested":"Every possible group of 11 out of the 13, in artwork-number order both ways.","happened":"Ruled out. This is a deliberate house style for the robot batch, not damage. Theory 62 later confirmed it: not one of those robots actually glitches.","status":"ruled-out","links":"[]","media":"[]","sort":1},{"id":3,"num":3,"title":"Prime-numbered robot glitches","picked":"Artworks that are both in the roboGAN group AND labelled Type=Glitch. Exactly 11 of them. Their binary numbers (7, 11, 17, 23, 31, 37, 41, 43, 47, 53, 61) are ALL prime numbers, which matches a rule Pindar wrote down in his own journal 002.","tested":"18 different orders: binary number up and down, position in the BIP-39 word list, bits reversed, count of 1-bits, number of eyes, dimensions, subtype, title, and alphabetical. For each order we tried all 2,048 possible 12th words in all 12 slots.","happened":"No match. Still the prettiest coincidence in the whole project. The odds of 11 randomly chosen numbers all being prime are about 1 in a million. Theory 3 pointed the way to the much stronger prime finding in Theory 66. [2026-08-25 correction: the Photoshop provenance channel showed the robo glitches are BORN glitched - present at original export, not planted afterward like the skull and solstice edits. Theories treating robo glitches as planted picks were wrong in kind.]","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}, {\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":2},{"id":4,"num":4,"title":"Real typing mistakes in Pindar's own spreadsheet","picked":"Actual misspellings in the official Arweave spreadsheet: Type='Gltich' on #73, SubType='Raibow' on #170 and #175, 'Cronenber' on #123, 'RainbowShimmer' on #128, 'spookyGan' on #205. Plus two filenames spelt wrong: 'gltich' on #9 and #73, and 'raindow' on #128.","tested":"We built them into a group of 11 and a group of 12. We tried every version with one member dropped. We ordered them by artwork number, alphabetically, and by position in the BIP-39 word list, and we put the 12th word in all 12 slots.","happened":"No match. One near-miss worth knowing: the group of 12 in artwork-number order does pass BIP-39's built-in error check. That sounds exciting but is not, because roughly 1 in 16 random word sequences passes that check by luck. We then tried 78 different wallet address recipes on it and none produced the target address. UPDATE (Sep 2026): the sheet's own errors are now exactly known — Cronenber #123, RainbowShimmer #128, Raibow #175 (the sheet introduces the error), plus Gltich #73 which is in Pindar's sheet AND filename. See theory 177.","status":"near-miss","links":"[]","media":"[]","sort":3},{"id":5,"num":5,"title":"Words that are not proper BIP-39 words","picked":"Only three of the 512 essences are not clean lowercase BIP-39 words. #191 is 'err0r' with a zero in place of the letter o. #363 is 'TRUE' and #473 is 'FALSE', both in capitals because a spreadsheet turned them into true/false switches. Michael confirms the TRUE and FALSE are deliberate.","tested":"We folded these three into the typo groups and the mixed-glitch-kind groups, and tested every group and every order those produced.","happened":"No match on their own, because three is not eleven. Still the hardest piece of evidence we have that Pindar planted deliberate corruption in the data.","status":"ruled-out","links":"[]","media":"[]","sort":4},{"id":6,"num":6,"title":"Artwork files that were missing or badly named","picked":"Artworks whose animation file we could not find in the Arweave archive (#57, the only missing skullGAN, whose word is 'mistake'; plus #134, #446, #472 and #477). Plus #143, whose filename has 5 binary digits where every other robot has 6.","tested":"Every possible group of 11 built from these plus other kinds of damage, in artwork-number and mint order, with the 12th word in every slot.","happened":"No match, and NOW KNOWN TO BE WRONG. Theory 63 went back to Pindar's original archives and found every one of those files. They were missing from our download, not from the collection. Only #143's short filename survives as a real defect.","status":"retracted","links":"[]","media":"[]","sort":5},{"id":7,"num":7,"title":"Named after a computer error","picked":"Artwork #404 is called 'Not Found' / '404 Not Found', and its word is 'crash'. 404 is the classic web error code.","tested":"We kept it as a fixed member in every corruption search and every mixed-glitch-kind search.","happened":"No match in any group it appeared in.","status":"ruled-out","links":"[]","media":"[]","sort":6},{"id":8,"num":8,"title":"Words that break their batch's alphabet run","picked":"Within each batch the words normally march through the BIP-39 list in order. Four break the run: #57 'mistake', #21 'bean', #259 'bless' and #268 'bird'. We found them by looking for the longest run that does stay in order and flagging what falls outside it.","tested":"Every group of 11 combining these with other kinds of damage, ordered by artwork number, alphabetically, and by BIP-39 list position.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":7},{"id":9,"num":9,"title":"Artworks that break Pindar's own hidden numbering rule","picked":"The blockchain carries a hidden trait that the Arweave spreadsheet does not. Pindar's journal 002 says the letter is set by the number type: Fibonacci gets A, other primes get B, square numbers get C, other odd numbers get D, everything else gets E. The rule holds for 163 of the 173 artworks that carry the trait. It breaks on #32 (whose trait is literally the text '32'), on #187 (marked C where the rule says E), and on 8 palindrome nagaGANs carrying an extra type 'F' that Pindar never documented.","tested":"All 12 rule-breakers, every version with one dropped, 6 orders each, and the 12th word in all 12 slots.","happened":"No match. Worth knowing that #32 and #119 are the two oddities Pindar himself flagged in journal 002 as 'typo or important?', and #32's is still sitting in the live data. CONFIRMED BY LATER WORK: Theory 66 shows the a-to-e rule is real and load-bearing, so these breaks are genuine breaks. UPDATE (Sep 2026): the code is fully decoded (theory 151); the breaks are the five misfiles of theory 153 (#32, #119, #161, #187, #193), and F is the robo-only palindrome class.","status":"ruled-out","links":"[{\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":8},{"id":10,"num":10,"title":"Artworks with no blockchain traits at all","picked":"22 skullGANs whose blockchain record carries no traits whatsoever, just a name, a description and a picture. All from batch one. The group includes #119 (Pindar's other flagged oddity) and #57.","tested":"One representative was used in the mixed-glitch-kind searches, and the earliest one (#4) in the first-of-each-kind search.","happened":"No match. We still do not know whether the blank traits are deliberate or just a gap in what OpenSea stored. Six of the artworks with no traits do glitch (see Theory 66), so the gap is not harmless.","status":"ruled-out","links":"[]","media":"[]","sort":9},{"id":11,"num":11,"title":"Artworks whose colours contradict their own label","picked":"We measured the actual colours and compared them with the SubType label. Eight disagree, for example an 'Inverted' one with a light background, a 'Frozen' one that is not cool, a 'Silver' one that is strongly coloured. Crucially we measured across the WHOLE animation.","tested":"Every group of 11 built from the 8 offenders plus each of the other 513 artworks in turn as the eleventh member, both orders, 128 possible 12th words each.","happened":"No match. IMPORTANT WARNING: an earlier 10-artwork version of this group was junk. It came from looking at only the first few pictures of each animation, and these animations cycle through colours, so the early pictures lie. Michael caught the error using #19.","status":"ruled-out","links":"[]","media":"[]","sort":10},{"id":12,"num":12,"title":"Artworks that look unusual for their own subtype","picked":"Artworks sitting more than 2.8 standard deviations away from the normal look of their own subtype, measured on colour, brightness and background. In plain terms: the ones that look most out of place next to their own siblings.","tested":"Every group of 11 from the offenders, plus each other artwork in turn as the eleventh, both orders, 128 possible 12th words each.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":11},{"id":13,"num":13,"title":"Animations with an odd number of pictures","picked":"Artworks whose animation is a different length from the rest of its batch. 29 across the collection. Length is otherwise perfectly uniform inside a batch. In the Solstice batch the only 4 odd ones are exactly the four artist tribute pieces that mark the season boundaries, which proves Pindar uses animation length as a deliberate marker.","tested":"All 21,470,000 possible groups of 11 from the 28 candidates with the word 'random' stuck on the end, both orders. Separately, the 10 odd ones in Act 1 Review, each paired with one of 58 other reprises as the eleventh.","happened":"No match.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":12},{"id":14,"num":14,"title":"One strange picture in the animation (first attempt)","picked":"Pindar's own words: 'noohp discovered a glitch in the portals... one frame of each testnet bitGAN reveals mysterious figures.' So we read a glitch as one odd picture hiding something.","tested":"We scanned all 483 animations we had for pictures that break the smooth flow, and scored each animation.","happened":"Ruled out as formulated. The scores form a smooth slope with no natural break (average 2.67, highest 5.57), so there is no clean set of 11. NOTE: the idea was right but the measurement was wrong. Theory 64 finds the real signal by comparing every picture to every other picture instead.","status":"ruled-out","links":"[]","media":"[]","sort":13},{"id":15,"num":15,"title":"Title names one creature, the group field names another","picked":"Artworks whose title says one creature but whose group field says a different one, such as 'Invasive ghostGAN' filed as a bitGAN, or 'cyclopsGAN' filed as a skullGAN. 10 of them once you remove the named originals, the palindromes and the artist tributes.","tested":"Every group of 11 from those, plus each other artwork as the eleventh, both orders, 128 possible 12th words each.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":14},{"id":16,"num":16,"title":"Named after real people","picked":"Artworks named after a real person or organisation. That includes #210 'noohp' and #211 'mazzone', who are people who SOLVED Pindar's earlier puzzles, plus the artists he paid tribute to (Abosch, Hackatao, Killer Acid, Eclectic Method, StellaBelle, Brinkman) and the prize artwork itself. Exactly 11.","tested":"More than 24 orders: artwork number both ways, alphabetical both ways, BIP-39 list position both ways, word length, and every way of shuffling the solver / artist / Act 1 blocks. The 12th word went in all 12 slots. We also tried swapping in The Moirai, Quantum Glitch and Four twentyGAN, and dropping the prize.","happened":"No match. This is still the cleanest single rule that lands on exactly 11 people. CLOSED BY LATER WORK: Theories 38 and 72 have now tried all 39,916,800 orders of this set and five variants of it. Order can never explain its failure again.","status":"ruled-out","links":"[]","media":"[]","sort":15},{"id":17,"num":17,"title":"BIP-39 words hiding in the Name field","picked":"Artworks whose Name field contains a BIP-39 word. 16 of them. Michael's reasoning was that the answer should need no metadata at all - you should be able to read the word straight off the picture.","tested":"All 4,368 possible groups of 11 out of the 16. Where a name held more than one word we expanded it (evil/olive, eye/swarm, cloud/skull). 4 orders each. 10,583,040 phrases tested.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":16},{"id":18,"num":18,"title":"BIP-39 words hiding in the Title","picked":"Words you can pull out of titles once you strip the GAN suffix, so skullGAN becomes skull. Inside Act 1 Review there are exactly 12 different ones: skull, monster, test, slender, ghost, bean, frozen, magic, quantum, rose, vibrant, vivid.","tested":"All 12 versions with one word dropped, times 8 orders (first artwork, alphabetical, BIP-39 list position, how often the word appears, and each of those reversed), times all 12 slots for the 12th word. Plus the true mint order.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":17},{"id":19,"num":19,"title":"Words that MEAN something broke","picked":"Artworks whose word describes a malfunction: error, crash, broken, virus, chaos, destroy, erase, problem, mistake, cancel, escape, random, false, true. 26 candidates.","tested":"All 4,460,000 groups of 11 out of 25 with 'random' on the end, then again with 'random' at the front, then the same 4,460,000 with 'random' in ALL 12 slots.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":18},{"id":20,"num":20,"title":"Words about puzzles","picked":"Artworks whose word is itself a puzzle word: answer, puzzle, hidden, unlock, valid, code, true, question, correct, verify, key, solution, solve, false, truth. 15 of them. Read together they sound like a deliberate message.","tested":"All 1,365 groups of 11 out of the 15, times 128 possible 12th words, both orders, plus 'random' inserted at every slot.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":19},{"id":21,"num":21,"title":"Act 1 Review artworks that look like nothing before them","picked":"Michael's five groups that resemble nothing in any earlier batch: the TEST-stamped ones (#467-469), the rainbow scan-line ones (#485-488), the vertical-bar smeared ones (#497-500), the black-background ones (#501-504), and the corrupted or abstract ones (#505-511).","tested":"COMPLETE SEARCH. All 705,432 groups of 11 out of the 22, in artwork-number order, with all 128 12th words that pass BIP-39's built-in error check. About 90 million phrases over 8.7 hours. Plus all 6 whole-group combinations across 48 orders.","happened":"No match, and this one is FINAL for that way of framing it.","status":"ruled-out","links":"[]","media":"[]","sort":20},{"id":22,"num":22,"title":"Two artworks sharing one identical file","picked":"We took a digital fingerprint of every animation file. #14 'Rainbow skullGAN' (word: accuse) and #75 'Flaming Glitch skullGAN' (word: announce) are the same file byte for byte. The shared picture measures 0.916 on warmth, which matches Flaming (0.768), not Rainbow (0.464). So #14 is carrying the WRONG picture.","tested":"Both members were used as a kind of damage across the mixed-kind searches.","happened":"No match as part of any group. The duplicate itself is real and the community already knew about it. CONFIRMED BY LATER WORK: Theory 75 re-ran the fingerprinting on all 512 files (we only had 483 before) and #14/#75 is still the only identical pair in the whole collection.","status":"ruled-out","links":"[]","media":"[]","sort":21},{"id":23,"num":23,"title":"An Act 1 Review artwork too similar to an older one","picked":"Michael's idea that the MATCH itself is the glitch. We ranked all 64 Act 1 Review repeats by how closely they resemble any earlier artwork.","tested":"The top 11 by similarity (score above 0.796) were tested, and so were the older artworks they matched, across 8 orders each including mint order.","happened":"No match. The similarity scores form a smooth slope with no natural break, so repeats resemble their originals on a sliding scale rather than in a clean set.","status":"ruled-out","links":"[]","media":"[]","sort":22},{"id":24,"num":24,"title":"Eleven DIFFERENT kinds of glitch","picked":"Michael's reframe: maybe each of the 11 is a different kind of glitch. We found 12 kinds with exactly one clear member, and 18 kinds in total once kinds with several members are counted.","tested":"COMPLETE SEARCH. All 714,935 ways of picking 11 different kinds with one representative each, in artwork-number order, with 128 possible 12th words. 9 hours. Plus every version of the 12 single-member kinds with one dropped, across 6 orders and all 12 slots.","happened":"No match, and FINAL for that way of framing it.","status":"ruled-out","links":"[]","media":"[]","sort":23},{"id":25,"num":25,"title":"The FIRST example of each kind","picked":"Michael's refinement: take only the first time each kind appears, and define the kinds consistently, so a typo counts as a typo wherever it turns up. That gives 14 kinds and 14 first-example artworks.","tested":"All 364 groups of 11 out of the 14, times 6 orders, times 128 possible 12th words. 279,552 phrases.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":24},{"id":26,"num":26,"title":"Words from the OTHER wordlist","picked":"Pindar's podGallery puzzle had a twist: the words were from SLIP-39, not BIP-39, and you had to swap each one for the BIP-39 word at the same position in the list. We tried that trick on our candidate groups.","tested":"We checked every candidate group of 11 to see whether all its words exist in SLIP-39.","happened":"Does not apply. 140 of the bitGAN words do exist in SLIP-39, but not a single one is SLIP-39-only, so no group could ever be fully converted. The trick is real but it has no grip here.","status":"tested","links":"[]","media":"[]","sort":25},{"id":27,"num":27,"title":"Doing sums on the word positions","picked":"Every BIP-39 word has a number from 0 to 2047. We tried shifting each artwork's number by things Pindar likes: add or subtract the artwork number, add or subtract a Fibonacci number based on its position, set the number equal to the artwork number, or flip it (2047 minus the number). Inspired by Michael's own Fibonacci-Caesar portal puzzle.","tested":"All of the above, plus a complete sweep shifting every word by the same amount k for every k from 0 to 2047 on the two strongest groups. 533,504 phrases.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":26},{"id":28,"num":28,"title":"The words are hidden in the clue text","picked":"Pindar's FIRST puzzle hid the 12 words as 'the only BIP-39 words contained in the clue'. So we pulled every BIP-39 word out of the glitch clue, the announcement tweet and the full cryptoArtGANs blog post.","tested":"Straight extraction, then ordered by where each word appears in the text.","happened":"Ruled out. The clue itself yields only 4 BIP-39 words (seed, phrase, random, word). The tweet yields 19 and the announcement 45. Nothing gives a clean 11.","status":"ruled-out","links":"[]","media":"[]","sort":27},{"id":29,"num":29,"title":"The answer is hidden inside the prize artwork","picked":"We asked whether the prize animation itself hides the answer, in hidden comment blocks, in the pixels, in extra pictures, or in the colour of its two eye-squares across all 88 pictures.","tested":"We pulled out all 88 pictures, read the raw file structure, and sampled the eye colours picture by picture.","happened":"Ruled out. The file holds nothing but ordinary Photoshop metadata, there are no hidden pictures, and the eye colours slide smoothly from one to the next. That smoothness is how these animations are generated, so it cannot be a code.","status":"ruled-out","links":"[]","media":"[]","sort":28},{"id":30,"num":30,"title":"The blockchain holds traits the spreadsheet does not","picked":"We discovered the official Arweave spreadsheet is INCOMPLETE. The blockchain record carries extra traits that vary by batch: a 'Date' trait (which is what solved the Solstice puzzle) and a 'HiddenType' trait.","tested":"We recovered blockchain records for 498 of the 512 artworks by walking the token numbers, then listed every trait name used in every batch.","happened":"This is a structural finding rather than a solution. Ghosts in the Machine and Act 1 Review carry NO hidden trait in 160 records checked. That is either the answer or a genuine dead end. LATER WORK MAKES IT LOOK LIKE A DEAD END: Theory 66 shows those two batches contain no glitches at all.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":29},{"id":31,"num":31,"title":"We had been using the wrong order all along","picked":"Pindar used MINT ORDER (the order the artworks were actually created on the blockchain) in two of his three solved puzzles. We had been using the artwork number as a stand-in, but the two disagree at 150 of 498 positions.","tested":"We re-ran every strong candidate group in TRUE mint order, up and down, with the 12th word in all 12 slots: the 11 people, the 11 prime robot glitches, the 11 typos, the odd-length animations, the 11 ghosts and both groups of 12 with one dropped.","happened":"No match. The flaw was real and is now fixed, but it was not what was blocking us.","status":"ruled-out","links":"[]","media":"[]","sort":30},{"id":32,"num":32,"title":"Fibonacci blink (first attempt at Pindar's real meaning)","picked":"Michael's account of what glitch originally meant: these animations are smooth loops, and a glitch is one that was not fully smoothed, with extra pictures inserted on the Fibonacci sequence so it blinks faster and faster before restarting. We built a detector and tuned it against the artworks already labelled Type=Glitch: those average 0.685 on picture-to-picture change versus 0.365 for everything else, and hold about 11 times as many near-frozen pictures. We could see it too: #231 freezes for pictures 1 to 28 then bursts, with still periods shrinking 28, 4, 2, 1. #274 shrinks 30, 2, 2, 1.","tested":"We scored all 487 animations for freeze runs that shorten towards the end. 12 score above 1 (angle, aim, board, base, bike, cat, bottom, analyst, alarm, august, avocado, around). All 12 versions with one dropped, times 8 orders, times all 12 slots.","happened":"No match, and the detector was WRONG. It looked for FREEZING, meaning neighbouring pictures barely changing. Michael corrected us: a glitch is one picture pasted back in several times with other pictures in between. Theory 64 rebuilds it properly and the answer changes completely.","status":"ruled-out","links":"[]","media":"[]","sort":31},{"id":33,"num":33,"title":"Michael's cluster map","picked":"Column D of the main properties sheet sorts all 68 Act 1 Review artworks into 25 clusters, which are really Pindar's own Arweave filename endings. They include 13.0=corrupted (#505-511), test (7 members), aa=high-definition robots, line/lines, bs=black background, highdef, skullall, dgans, and the reprise groups 0 to 13. EXACTLY 11 CLUSTERS HAVE 3 OR MORE MEMBERS. Pindar's numbering skips 7 - there is no reprise07.","tested":"First-of-each-cluster and last-of-each-cluster as 11-word groups across several orders. Then a COMPLETE search of every way of taking one member from each cluster: 5,419,008 groups on 8 processor cores.","happened":"No match. The complete one-per-cluster search finished with zero hits.","status":"ruled-out","links":"[]","media":"[]","sort":32},{"id":34,"num":34,"title":"Artworks Michael flagged as being in the wrong place","picked":"Michael's own notes in column R flag artworks as misfiled: the 'aa' four ('high definition robogans from the 2021.9.29.1202 drop'), the 'line' four ('Pindar playing with colour'), the 'test' three ('should be with the other Ghost in the Machine GANs'), and two spookyGAN invaders. The first three groups come to EXACTLY 11 and include the word 'random'.","tested":"466 orders: the group of 11 and every version of the group of 13 with two dropped, across artwork number, TRUE mint order, alphabetical, BIP-39 list position and cluster-block order, each reversed, with the 12th word in all 12 slots.","happened":"No match. CLOSED BY LATER WORK: Theory 38 has since tried all 39,916,800 orders of this group.","status":"ruled-out","links":"[]","media":"[]","sort":33},{"id":35,"num":35,"title":"The 8x8 grid picture","picked":"We asked whether the 8-by-8 grid animation on the 'Final bitGAN Drop' post disagrees with Arweave or with the live collection, and whether anything is buried in the file itself.","tested":"We cut the 512-by-512 grid into 64 cells, averaged each cell over time, and matched each against every Act 1 Review artwork. We also read the raw file for hidden comment blocks and trailing data.","happened":"Ruled out. The grid is simply #449 to #512 in order and it matches the live collection (54 of 64 matched exactly; the 10 flagged by a weak measure were false alarms and we confirmed that by eye). No hidden data, only ordinary Photoshop metadata, zero trailing bytes. Worth noting Pindar deliberately LEAVES OUT #445 to #448, the prize and the three artist pieces.","status":"ruled-out","links":"[]","media":"[]","sort":34},{"id":36,"num":36,"title":"Whole clusters added together","picked":"If the eleven are simply whole clusters stacked up, how many ways can cluster sizes add to exactly 11? Michael's cluster map puts all 68 Act 1 Review artworks into 25 codes, and there are 15,086 ways: 12 using two clusters, 129 using three, and thousands more using the small reprise groups and singletons.","tested":"COMPLETE SEARCH. All 15,086 stacks, with the 12th word at the front, at the back, and at every join between clusters, trying all 2,048 words at each spot, in artwork-number and mint order, forwards and backwards. 11.3 hours.","happened":"No match, and FINAL for whole-cluster stacks.","status":"ruled-out","links":"[]","media":"[]","sort":35},{"id":37,"num":37,"title":"The seven-plus-four stacks","picked":"Eight Act 1 Review clusters have either 7 or 4 members (13.0, test, 9.0, 11.0, aa, skullall, highdef, bs). Only 7 plus 4 makes 11, which gives 12 combinations.","tested":"All 12 stacks, times 128 orders, using both the word list and a mixed title-word list, times all 12 slots including the join, times all 2,048 words.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":36},{"id":38,"num":38,"title":"Trying EVERY possible order","picked":"Closing the order gap for good. For any group of 11 words we had only ever tried 4 to 6 of the 39,916,800 possible orders. If you fix the 12th word to one specific word, each order only costs a single cheap check instead of 2,048 expensive ones, which makes trying every order affordable.","tested":"ALL 39,916,800 orders for the out-of-place 11 and for the people 11, each with 9 candidate 12th words, on the end and on the front. 6 hours each.","happened":"No match, and FINAL. A wrong order can never again explain why those two groups failed. This trick is the foundation of Theory 72, which scaled it up to 18 groups.","status":"ruled-out","links":"[]","media":"[]","sort":37},{"id":39,"num":39,"title":"Trying all 2,048 possible last words","picked":"Michael asked whether every possible 12th word could be tested. Across every order that would take 28.6 days for one group, measured at 2,066 phrases per second. So instead we shrank the order space to orders that keep the three out-of-place blocks (test, line, aa) unbroken, which is 20,736 orders. That makes full coverage of the 12th word affordable.","tested":"20,736 block-respecting orders, times ALL 2,048 possible 12th words, on the end and on the front. 45 minutes.","happened":"No match. No word anywhere in the BIP-39 list completes that group in any block-respecting order.","status":"ruled-out","links":"[]","media":"[]","sort":38},{"id":40,"num":40,"title":"One word repeated eleven times","picked":"Michael's idea that the answer might be a trick of simplicity. BIP-39 allows the same word more than once, so the phrase could be a single word repeated eleven times plus one other word. That removes the ordering problem entirely.","tested":"COMPLETE SEARCH. All 2,048 words repeated 11 times, times all 2,048 possible 12th words, times all 12 slots. 50,331,648 phrases. 25 minutes.","happened":"No match, and FINAL. This kills an entire SHAPE of answer for every word in the list, not just the likely ones.","status":"ruled-out","links":"[]","media":"[]","sort":39},{"id":41,"num":41,"title":"Artworks sitting at Fibonacci positions","picked":"Michael's idea of treating the whole collection as one big glitch and taking the artworks at Fibonacci positions, anchored either on #0 Pandimensional skullGAN or on #1 The Moirai. Fibonacci numbers up to 512 give 14 values, or 15 if you count 1 twice as the true sequence does, so there are 4 or 5 sliding windows of eleven.","tested":"Every window for four versions of the sequence (with and without 0, with 1 once and 1 twice), across the essence list, a mixed list and a title-word-first list, expanding every combination where an artwork yields more than one title word, in Fibonacci, mint and alphabetical order, with the 12th word in all 12 slots and all 2,048 words tried.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":40},{"id":42,"num":42,"title":"The word and the title agree","picked":"Exactly 10 artworks where the official word and the word you can read off the title are the SAME: awkward, cat, dog, slender, skull, wolf, bird, space, vivid, rose. Six come from the robot batch and four of those are palindrome nagaGANs. That concentration is not chance.","tested":"The 10 held fixed, plus ANY of the 2,048 words as the eleventh inserted at ANY of 11 positions, plus ANY 12th word, across four base orders. 8,192 separate jobs.","happened":"Was in progress when this was written. No hit reported.","status":"tested","links":"[]","media":"[]","sort":41},{"id":43,"num":43,"title":"Michael's own hand counts of animation length","picked":"Michael counted the pictures in all 512 animations by hand. Most batches are PERFECTLY uniform: every skullGAN has 41, every spookyGAN and every Ghost in the Machine has 55, with zero exceptions. So any deviation is deliberate. There are 15 in the robot batch, 10 in Act 1 Review, and 4 in the Solstice batch which are EXACTLY the four artist tribute pieces marking the season boundaries.","tested":"Every group of 11 from the 15 robot oddities; the Act 1 Review ten plus each other oddity in turn; and the Solstice four plus every group of 7 from the robot oddities. 7,819 candidate groups, 4 orders each, 128 possible 12th words.","happened":"Was in progress when this was written. No hit reported. NOTE: 55, 89 and 21 are all Fibonacci numbers, which fits Pindar inserting pictures on the Fibonacci sequence.","status":"tested","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":42},{"id":44,"num":44,"title":"Adding one of Pindar's bookend artworks","picked":"Michael's idea that any group of 10 might become eleven by adding one of Pindar's boundary pieces: #0 Pandimensional ('before the first, after the last'), #1 The Moirai (the first bitGAN), or #512 Quantum Glitch (the last).","tested":"14 different groups of 10, found by searching a library of 53 rules for combinations that land on exactly 10, times 3 anchor artworks, on the front and the back, times 4 orders, times 128 possible 12th words.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":43},{"id":45,"num":45,"title":"Let the computer find the rules","picked":"Rather than guessing at definitions, we built a library of 53 yes/no rules covering batch, cluster, title word, Michael's flags, blockchain traits, blink measurements and glitch categories, then searched every combination for groups of exactly eleven.","tested":"All single rules, all pairs, and all title-word-anchored triples. 15 different groups of exactly 11 turned up, each tested using both the word list and the title-word list across 5 orders and all 12 slots.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":44},{"id":46,"num":46,"title":"The burn wallet and the Forge","picked":"Every bitGAN that was burned to forge a SuperRare piece. 32 NFTs sit in wallet 0xa7c2...1a0c, and 30 of them are collection bitGANs.","tested":"We listed them from the blockchain, paired them by who sent them, and swept their words.","happened":"No hit. We did confirm 16 Flaming plus 16 non-Flaming, meaning 16 forges out of the 24 Pindar planned.","status":"tested","links":"[]","media":"[]","sort":45},{"id":47,"num":47,"title":"The 11 burnt Flaming skullGANs","picked":"Narrow the burnt Flaming pieces down to just the skullGANs. That gives exactly eleven, which fits Pindar's rule that every forge consumes a Flaming bitGAN.","tested":"Artwork-number, burn, mint, alphabetical and BIP-39-position orders, times 12 slots, times 2,048 possible 12th words, times 33 different wallet address recipes.","happened":"No hit. KEY FINDING: #66 is listed as SubType=Rainbow in the spreadsheet but Flaming on the blockchain. So the spreadsheet gives you 10 and the blockchain gives you 11. CLOSED BY LATER WORK: Theory 72 has now tried all 39,916,800 orders of this group.","status":"tested","links":"[]","media":"[]","sort":46},{"id":48,"num":48,"title":"The burnt forge partners","picked":"The 11 non-Flaming bitGANs that were burned alongside the Flaming ones. Their words include ghost, only, ice, puzzle and hidden.","tested":"The same battery of orders, plus orders that keep each burnt pair together.","happened":"No hit. CLOSED BY LATER WORK: Theory 72 has now tried all 39,916,800 orders of this group.","status":"tested","links":"[]","media":"[]","sort":47},{"id":49,"num":49,"title":"Forged traits add up","picked":"Pindar says forged traits combine: '2 eyes and 3 eyes gives 5 eyes'. So we treated each pair of artwork numbers as adding up to a position in the BIP-39 word list.","tested":"Sums and differences of each forge pair, in every order.","happened":"No hit. We also disproved the idea that the forged piece numbers (20,051 and so on) are combined traits, because the Dimensions trait only ever runs from 2 to 5.","status":"tested","links":"[]","media":"[]","sort":48},{"id":50,"num":50,"title":"Anything to do with aliens or invaders","picked":"Any artwork whose Type, SubType, group, Title, Name or word mentions an alien or an invader.","tested":"We listed every rule-based group we could build.","happened":"No group of 11 exists in the metadata alone. The group sizes come out as 2, 3, 8, 16, 18, 19 and 20. [2026-08-25: the ghost-drop Photoshop re-save batch of 16 turns out to be exactly the complete spaceGAN family - the alien set was a family re-export, housekeeping rather than anomaly picks.]","status":"tested","links":"[]","media":"[]","sort":49},{"id":51,"num":51,"title":"Space Invaders you can SEE in the art","picked":"Michael flagged 34 artworks whose PICTURE shows a Space Invader, as a T-shirt design, as a head, or in the background. This is about the art, not the labels.","tested":"We cross-referenced those 34 against the metadata.","happened":"This produced an eleven: the artworks that show an invader where the metadata never mentions one.","status":"tested","links":"[]","media":"[]","sort":50},{"id":52,"num":52,"title":"The visible-invader eleven","picked":"#314, #320, #349, #377, #379, #398, #401, #412, #431, #473, #475. In each one you can see an invader and the metadata says nothing about it.","tested":"13 orders. Then the #473 slot alone was swept across all 2,048 words in all 12 positions, split across 8 processes. 6,290,000 phrases.","happened":"No hit.","status":"tested","links":"[]","media":"[]","sort":51},{"id":53,"num":53,"title":"The one word that is not a real BIP-39 word","picked":"#191 'err0r' is the ONLY word in all 512 that is not a valid BIP-39 word. If the puzzle needs a slot you cannot read off the art, this must be it.","tested":"We took the other 11 hand-picked glitches, put all 2,048 candidate words into the #191 slot at all 12 positions, across 28 orders and 33 wallet address recipes.","happened":"No hit. The uniqueness finding is still true and does not depend on the sweep.","status":"tested","links":"[]","media":"[]","sort":52},{"id":54,"num":54,"title":"Rebuild the definition of a glitch from scratch","picked":"We spell-checked every word in every title against the collection's own vocabulary, hunted near-duplicates column by column, and checked number-for-letter swaps, odd capitals, missing spaces and invalid words.","tested":"We threw out anything that turned out to be a systematic house style (0o for octal, 0x for hex, the G1itch1ng and b0rg spellings).","happened":"This independently reproduced Michael's own hand-picked twelve: 7 spreadsheet typos plus 5 file or asset problems. NOW PARTLY WRONG: Theory 63 disproved the 5 file problems. The 7 typos still stand.","status":"tested","links":"[]","media":"[]","sort":53},{"id":55,"num":55,"title":"Where the blockchain and the spreadsheet disagree","picked":"We compared every blockchain trait against the Arweave spreadsheet, ignoring the '(n/128)' suffix.","tested":"20 real disagreements, split by which side is the corrupted one.","happened":"This produced an eleven where the BLOCKCHAIN is the corrupt side. Best single find: #48's SubType reads 'wobniaR', which is 'Rainbow' spelled backwards. 12 orders, no hit. PROVISIONAL, because only 393 of 513 artworks carry blockchain traits at all. CLOSED FOR ORDER BY LATER WORK: Theory 72 has tried all 39,916,800 orders of this group.","status":"tested","links":"[]","media":"[]","sort":54},{"id":56,"num":56,"title":"Brute-force all 39,916,800 orders (the naive way)","picked":"If the words are right but we cannot work out the order, just try every order.","tested":"We measured the speed first. The password-stretching step alone costs 1.4 thousandths of a second per phrase and cannot be sped up.","happened":"NOT RUN this way. It works out at roughly 2,255 processor-hours, or 9 days on 10 cores, for ONE group. SUPERSEDED: Theory 38 found a trick that does the same job in minutes by fixing the 12th word first.","status":"tested","links":"[]","media":"[]","sort":55},{"id":57,"num":57,"title":"The seventeen liars, exhausted","picked":"The 17 skullGANs Michael checked by hand and confirmed are named as a Glitch but do not actually glitch: #9, #15, #25, #35, #45, #46, #51, #56, #58, #60, #72, #75, #84, #86, #90, #102 and #115. If the puzzle rewards spotting the lie in the labels, eleven of these are the answer.","tested":"Every one of the 12,376 possible groups of 11, each in artwork-number order and reversed, with all 2,048 words tried at each of the 12 positions, across 3 wallet address recipes. 57,045,696 phrases.","happened":"Zero hits, and this is CLOSED rather than sampled. For skullGANs the word is simply BIP-39 word number [artwork number], so artwork order IS alphabetical order IS mint order. There is no fourth order left to try. The complete space for this idea has been searched.","status":"open","links":"[]","media":"[]","sort":56},{"id":58,"num":58,"title":"Sort into groups first, then sort inside each group","picked":"If the eleven words are right but the order comes from a property rather than the number, then the order is a two-stage sort: first sort the artworks into groups, then order the members inside each group.","tested":"10 candidate groups times 7 grouping properties (batch, group name, Type, SubType, first letter of the word, animation length, mint order), every way of arranging the groups, with members going up and down inside each group. 11,129,380 phrases.","happened":"Zero hits.","status":"tested","links":"[]","media":"[]","sort":57},{"id":59,"num":59,"title":"The podGAN symbol hidden in monsterGAN backgrounds","picked":"Michael spotted 16 monsterGANs with a podGAN symbol in the background: #310, #314, #320, #328, #349, #355, #370, #372, #375, #377, #379, #398, #401, #412, #421, #431. The metadata names only 5 of them Invader, Invading or Invasive (#310, #328, #370, #372, #375) and says nothing about the other 11. That is the same shape as the Space Invader eleven, but drawn from a symbol the collection uses elsewhere as a seed reference.","tested":"The silent eleven (#314, #320, #349, #355, #377, #379, #398, #401, #412, #421, #431 = supreme fox exile off drip please expire dolphin lucky then mystery) plus every group of 11 out of the full 16, across 16 orders.","happened":"Zero hits. The eleven is clean and the metadata silence is real, but it is not the answer. PARTLY SUPPORTED BY LATER WORK: Theory 68's automatic look-alike detector independently ranked 9 of these 16 in the top 10% of their own batch. CLOSED FOR ORDER: Theory 72 has tried all 39,916,800 orders of the silent eleven.","status":"tested","links":"[]","media":"[]","sort":58},{"id":60,"num":60,"title":"The words the forge pairs COMBINE into","picked":"The Forge burns two bitGANs to mint one new piece. If the forge is the operation the puzzle wants, the answer is not the burnt words but the words their pairs turn into.","tested":"11 forge pairs, each pair collapsed into one new BIP-39 position by adding, XOR-ing, multiplying, averaging, and both subtractions, all wrapped at 2,048. That gives 11 new words per method and 54 resulting word groups. 82,727 phrases.","happened":"Zero hits.","status":"tested","links":"[]","media":"[]","sort":59},{"id":61,"num":61,"title":"The podGAN SLIP-39 mapping","picked":"podGANs are project 2 on BrainDrops and each carries a 'seed' trait. That trait is the SLIP-39 word at position (edition number minus 1). We confirmed this on all 22 podGANs we checked.","tested":"Once confirmed, we pushed the same idea at the bitGANs: SLIP-39 words taken by edition number and by artwork number, then converted to BIP-39 by position, across 8 orders.","happened":"Zero hits for the bitGANs. The mapping itself is a confirmed fact about the podGANs and now sits in the PodGAN_SLIP39_seed column on the master sheet. CONFIRMED IN FULL BY LATER WORK: Theory 69 checked all 1,024 podGANs, not just 22.","status":"tested","links":"[]","media":"[]","sort":60},{"id":62,"num":62,"title":"Do the robots named G1itch actually glitch?","picked":"Six roboGANs carry 'G1itch' in their title (#170, #172, #176, #182, #188, #190). We rescanned their animations with the tuned freeze rule to find out.","tested":"We recounted held pictures across all 64 robot animations in the 2021.9.29.1202 batch.","happened":"The whole batch scored ZERO true glitches by that rule. The highest count anywhere in it was 6 (#182) against a floor of 12. So all six G1itch-named robots looked like liars, and the leet spelling looked like a house style rather than real damage. NOW OVERTURNED: the freeze rule was the wrong test. Theory 64's correct detector finds 16 real glitchers in this very batch, and #170, #172, #176, #182 and #190 are among the 11 with the complete backwards-Fibonacci run. Theory 2's conclusion about the leet spelling still stands on its own evidence, but this row's evidence for it does not. [2026-08-25: provenance confirms - the G1itch robos were authored glitched in the original Sep 2021 session; no re-save event touched them.]","status":"tested","links":"[]","media":"[]","sort":61},{"id":63,"num":63,"title":"Going back to Pindar's original archives","picked":"Six artworks were being treated as file glitches because their animations were missing: #57, #134, #143, #446, #472 and #477. We tested that claim by going back to the original Arweave archives instead of our local folder.","tested":"We downloaded and unpacked the source archives (01_skullgans with 128 files, 02_robogans with 64, 05_ghostgans with 18, 06_act1review with 68) and matched every flagged artwork against them.","happened":"THE MISSING-FILE GLITCH CLASS IS DEAD. Every one of them exists: #57 is 057_flaming_ghost.gif, #134 is 05_000101.gif, #446 is 446_roses.gif, #472 is 472_reprise11.gif, #477 is 477_reprise11.gif. The absences were holes in our download, not in the collection. Only #143 survives, and only as a filename defect: '14_00110.gif' has five binary digits where every other robot filename has six. This retracts the hand-picked twelve, which was 7 typos plus 5 of these phantom file problems. The 7 typos still stand.","status":"retracted","links":"[]","media":"[]","sort":62},{"id":64,"num":64,"title":"THE CORRECTED GLITCH DETECTOR","picked":"Michael corrected the definition and it changed everything. A glitch is NOT the animation freezing. A glitch is ONE picture that gets pasted back into the animation several times, with other pictures in between, and the gaps between those repeats shrink following the Fibonacci sequence run BACKWARDS. The old detector only compared each picture with the one next to it, so it could only ever see freezing. The new detector compares EVERY picture with EVERY other picture, finds the largest family of near-identical pictures, and records the gaps between them.","tested":"We scanned all 512 animations. For each one we build a full picture-versus-picture difference table, set the threshold at 30% of the typical neighbour-to-neighbour change, and keep any family of 3 or more matching pictures that has at least one gap of 2 or more. Nothing is tuned by hand beyond that single 30% figure.","happened":"It works. It separates Michael's confirmed examples cleanly, and 46 of the 512 artworks glitch. To be exact about what the shrinking looks like: the repeats sit 6, 4, 3, 2 and 2 picture-positions apart, which means 5, 3, 2, 1 and 1 pictures in between. That is 5, 3, 2, 1, 1 - the Fibonacci sequence backwards. This detector replaces the one in Theory 32 and overturns the conclusion of Theory 62.","status":"tested","links":"[]","media":"[]","sort":63},{"id":65,"num":65,"title":"GAP-SIGNATURE CLASSES - and the eleven","picked":"We sorted all 46 glitching artworks by the exact pattern of their gaps. Five classes appear: 9 with an irregular pattern, 9 with a shortened 4-step backwards run, 15 that start backwards then break, 2 that run FORWARDS (1,1,2,3,5) instead of backwards, and 11 with the complete backwards-Fibonacci run.","tested":"Exactly 11 carry the complete run: #148, #152, #158, #160, #166, #170, #172, #176, #182, #190 and #287. Their words are: hint jump that slender pink quiz robot three virus primary bottom. Ten of the eleven are robots from the 2021.9.29.1202 batch and the eleventh, #287, is a Vivid glitchGAN. We swept this group hard: 2,944 different orders (4,522,177 phrases, 15 hours), then a re-check pass of another 496 orders (761,947 phrases), then ALL 39,916,800 possible orders under Theory 72.","happened":"No match yet, but this is the strongest candidate group in the project. It survived a re-run of the classifier with leading and trailing odd gaps ignored - the same eleven came out both times, with only the smaller classes shuffling. Two things make it striking: the group falls out of the art alone with no metadata needed, and it lands on exactly eleven without anybody choosing a cut-off.","status":"ruled-out","links":"[]","media":"[]","sort":64},{"id":66,"num":66,"title":"THE PRIME PATTERN","picked":"In the first batch (artworks #1 to #128) every artwork that glitches sits at a prime number that is NOT also a Fibonacci number. There are 26 non-Fibonacci primes in that range and 25 of them glitch. #47 is the single exception: it is a non-Fibonacci prime that does not glitch. One artwork glitches from outside the class, #5, and 5 happens to be both prime and Fibonacci.","tested":"Confirmed four independent ways. (1) The full picture-versus-picture detector from Theory 64. (2) EXACT FILE FINGERPRINTING with no threshold at all - we took a digital fingerprint of every single picture in every animation and looked for fingerprints that repeat at spaced-out positions. That method has no tuning knob whatsoever, and it flags exactly 25 artworks in batch one: the non-Fibonacci primes minus #47. Zero false alarms, zero misses. (3) Pindar's own hidden metadata. His OnChain_HiddenType field labels each artwork by the KIND of number it sits on: a for Fibonacci, b for other primes, c for square numbers, d for other odd numbers, e for everything else. The rule holds for 105 of the 106 labelled skullGANs; the only break is #32, whose label is literally the text '32'. Every single artwork he labelled 'b' glitches - all 20 of them. (4) The pattern also appears in the robot batch, where the numbering restarts at 0, and the a-to-e rule holds there too for 44 of 52 labelled artworks once you use the batch-relative number.","happened":"This is the most solid structural finding in the project. The label letters are Pindar's own, so he is telling us the numbers matter. BUT IT ONLY WORKS IN THE FIRST TWO BATCHES. Glitch counts by batch: skullGANs 26, 2021.9.29.1202 robots 16, 2021.10.17 spookyGANs 0, Solstice 4, Ghosts in the Machine 0, Act 1 - Review 0. So both of the last two batches contain no glitches at all. That kills any theory built on Act 1 Review being the glitchy batch, and it makes Theory 30's blank-trait finding look like a dead end rather than a clue.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":65},{"id":67,"num":67,"title":"MEDIAN COMPOSITES - averaging every picture together","picked":"If you average every picture of an animation into one still image, moving things blur away and anything that stays put comes out sharp. That reveals background patterns and flickering elements that no single picture shows on its own.","tested":"We built one composite for all 512 artworks by taking the middle value of each pixel across all its pictures, at 256 by 256. We also measured the colour of the outer border of each composite, where the background patterns sit, and saved those numbers.","happened":"512 composites built and saved as a browsable contact sheet (contact_sheet.html) so the whole collection can be eyeballed at once. This is a tool rather than an answer, but it is what made the look-alike detection in Theory 68 possible.","status":"tested","links":"[]","media":"[]","sort":66},{"id":68,"num":68,"title":"AUTOMATIC VISUAL ODD-ONE-OUT DETECTION","picked":"Rather than trusting our eyes, we asked the computer which artworks look least like their neighbours. We turned each composite into a set of numbers describing its colours, its overall brightness pattern and its coarse layout, then measured how far each one sits from the average look of its group.","tested":"We ran it two ways: within each batch, and across the whole collection at once.","happened":"It independently rediscovered things Michael had already spotted by eye, which is a real check on both. The black-background group #501 to #504 all land in the global top 15 most-unusual artworks. In Ghosts in the Machine, 9 of Michael's 16 podGAN-symbol artworks from Theory 59 land in the batch's top 14 out of 144, which is the top 10%: #349 is 3rd, #314 4th, #377 5th, #412 6th, #431 8th, #310 10th, #320 12th, #372 13th, #398 14th. CAVEAT, and it matters: the global ranking is dominated by one label. 15 of the global top 30 are SubType=Inverted, and inverted colours are extreme by definition. So the global list partly just rediscovers a metadata field. The within-batch rankings are the ones worth trusting.","status":"tested","links":"[]","media":"[]","sort":67},{"id":69,"num":69,"title":"THE PODGAN COLLECTION - ruled out","picked":"podGANs are project 2 on BrainDrops, 1,024 of them. Each carries a 'seed' trait, which made them look like a possible source of words. Theory 61 had confirmed the mapping on 22 of them. We fetched the metadata for all 1,024 and checked properly.","tested":"We compared every podGAN's seed trait against the SLIP-39 word list. We then swept the podGAN words at the bitGANs: 8 different groupings and orderings across the whole collection, 835,619 phrases in total.","happened":"RULED OUT. A podGAN's seed trait is just the SLIP-39 word at its own number, with #0 wrapping round to the last word ('zero'). It holds for 1,022 of 1,024 exactly, and the only two breaks are #329 and #941, whose words are 'false' and 'true' and got turned into FALSE and TRUE by a spreadsheet - the same coercion we saw in Theory 5. So a podGAN's word carries no information at all beyond its number. Roughly 836,000 phrases tested, no match. This closes a whole avenue.","status":"ruled-out","links":"[]","media":"[]","sort":68},{"id":70,"num":70,"title":"MICHAEL'S CONDENSED-LIST IDEA","picked":"Michael's idea: 553 of the 1,024 podGAN words are also valid BIP-39 words. Filter down to just those 553, renumber them 1 to 553 in order, and then look up each bitGAN's number in that shortened list. That would be a real puzzle mechanism rather than a plain lookup.","tested":"We built the 553-word list and tested it both ways: 1-indexed (bitGAN #1 gets the first word) and 0-indexed (bitGAN #1 gets the second word). We looked bitGANs up both by artwork number and by mint order. The main sweep ran 32,256 different arrangements across three word sources and eight sort keys, and tried all 12 slots for the 12th word: 4,130,991 phrases over 5 hours.","happened":"No match. Two of the 553 needed a fix first: podGANs #329 and #941 carry FALSE and TRUE in capitals because of a spreadsheet, and had to be read back as 'false' and 'true' to count. NOTE: parts of the pod553 and mintpod runs were stopped and restarted, and those log files have no final DONE line, so we cannot claim complete coverage for every arrangement they queued. The arrangements that did run reported zero hits.","status":"ruled-out","links":"[]","media":"[]","sort":69},{"id":71,"num":71,"title":"THE ELEVEN NAMED AFTER REAL PEOPLE","picked":"Michael's reasoning: the prize artwork is called the Collector, so the answer should be something only a genuine bitGAN collector would recognise. Ten artworks are unmistakably named after real people or real organisations: #210 noohp and #211 mazzone (both past puzzle solvers), #228 Kevin aboschGAN, #246 Hacka taoGAN, #264 Killer acidGAN, #282 Eclectic methodGAN, #445 Magic Internet moneyGAN (the prize itself), #446 SuperRare roseGAN, #447 Brinkman nimBudGAN and #448 Crypto stellaGAN (StellaBelle). Their words are: awake aware barely belt blame boost collect rose cloud legend.","tested":"Ten is not eleven, so we tried five different eleventh members: #1 The Moirai (abandon), #2 The Artist (ability), #3 The Geometer (able), #512 Quantum Glitch (final) and #482 Testy McTestGAN (call). First we ran 10 sensible orders across 15,410 phrases. Then, under Theory 72, we ran ALL 39,916,800 orders for every single one of the five variants: 12,481,771 phrases, zero hits each.","happened":"No match. This is now a CLOSED question for order. For all five versions of this group, every possible arrangement of the eleven words has been tried with 'random' as the twelfth. If the people idea is right, then either the twelfth word is not 'random', or the eleventh member is someone we have not thought of, or one of the ten is wrong.","status":"ruled-out","links":"[]","media":"[]","sort":70},{"id":72,"num":72,"title":"THE BIG ONE - THE FIXED-TWELFTH-WORD TECHNIQUE","picked":"This is a method, not a theory, and it is what unlocked the last two days. BIP-39's last word carries a built-in error check, so only about 1 word in 16 can legally finish any given eleven. Normally you must try all 2,048 words at every position of every order, which is 39,916,800 orders x 12 positions x 128 legal words = about 61.3 billion phrases per group - completely out of reach. But if you ASSUME the twelfth word (we use 'random', straight from Pindar's clue) and pin it to the last slot, each order costs one cheap arithmetic check instead. The error check then throws away 15 out of every 16 orders before you ever do the expensive wallet maths. That leaves 39,916,800 / 16 = about 2.5 million real tests per group.","tested":"We ran it on all 13 candidate groups plus the 5 people-variants from Theory 71, which is 18 complete order-spaces. Each group is split across 10 processor cores. 13 sets x 39,916,800 = 518,918,400 orders, plus 5 variants = 718,502,400 orders in total. Measured phrases actually derived: 32,428,053 for the 13 sets and 12,481,771 for the 5 variants, which is 44,909,824 in total. Each phrase was checked against 3 wallet address recipes. Wall time was 9 minutes per group when the machine was quiet and about 17.5 minutes when other jobs were running.","happened":"ZERO MATCHES across all 18. The groups covered were: the exact-Fibonacci eleven, the podGAN-symbol eleven, the burnt Flaming skullGANs, the forge partners, the chain-corrupt eleven, the ce11 liars, the puzzle-word eleven, Sol/Luna plus artists, Moirai/Artist/Geometer plus artists, the Pythagorean primes, the contains-a-7 primes, the named eleven, and the inverted-Solstice group. WHAT THIS BUYS US: for those 18 groups the ORDER no longer matters. Every possible order was tried. If any of them is the right set of words, then the twelfth word is not 'random' at the end. That is a genuine narrowing, and it is the first time we have been able to say it.","status":"tested","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":71},{"id":73,"num":73,"title":"PRINCIPLED ORDERINGS (running now)","picked":"The other half of the trade-off. Theory 72 covers every order but only one twelfth word. This covers every twelfth word, but only orders that follow a rule you could actually state out loud. The reasoning is Pindar's own habit: he does not like brute force, so the order must be derivable from something.","tested":"For each candidate group we generate every ordering that comes from a real, stateable rule. Three shapes of rule: sort all eleven by one property; group them by a property, arrange the groups, then sort inside each group; or group by a property and sort inside by two properties in sequence. The properties are all real artwork facts: artwork number, mint order, the word itself, position in the BIP-39 list, eyes, dimensions, animation length, held pictures, unevenness, SubType, Type, group name, title, batch, hidden type, and number of repeated pictures. That produces 202,726 orderings across 31 groups. Every single ordering is then tried with ALL 2,048 possible twelfth words in ALL 12 slots, across 5 wallet address recipes.","happened":"STILL RUNNING as this was written. The work is split into 3,380 chunks and roughly 1,440 were finished at the time of writing, which is about 43%, on 10 processor cores after about 10 hours. 132,671,918 phrases tested so far, zero hits. The full job should land near 310 million phrases. The 31 groups are: burntflam, ce11, chaincorr, corrupt7_9, corrupt7_aa, corrupt7_bs, corrupt7_line, curated11, digit7, exactfib, glitch9_176_182, glitch9_97_147, invact8_trio, invader_old, invsol9_solQ, invsol9x, irregular9_Q, liar_flamprim, liar_rainprim, metaword, moiraiart, named, partners, people_moirai, people_quantum, podsymbol, pyth, rev4term9_x, revbreak11, shim4_glitch7 and sollunaart.","status":"tested","links":"[]","media":"[]","sort":72},{"id":74,"num":74,"title":"SOL/LUNA AND THE ARTIST TRIBUTES","picked":"Two more ways of landing on exactly eleven using Pindar's bookend habit. He marks boundaries with first-and-last pairs (sunrise/sunset, Solar/Luna, first and last square). Version one: #209 Sol and #300 Luna, plus the 8 artist tribute artworks (#228, #246, #264, #282, #445, #446, #447, #448), plus #512 Quantum Glitch, the very last artwork. Their words are: avoid barely belt blame boost brief collect rose cloud legend final. Version two: #1 The Moirai, #2 The Artist and #3 The Geometer - the first three artworks - plus the same 8 artist tributes. Their words are: abandon ability able barely belt blame boost collect rose cloud legend.","tested":"Both come to exactly eleven with no fudging. Version one got 158 different orders (243,047 phrases) and version two got 166 (255,672 phrases), including one-property sorts and group-then-sort arrangements. Both then went through the complete Theory 72 treatment: all 39,916,800 orders each.","happened":"No match on either. Both are now closed for order. They remain attractive because they are exactly the kind of first-and-last construction Pindar has used in every puzzle he has actually set.","status":"ruled-out","links":"[]","media":"[]","sort":73},{"id":75,"num":75,"title":"DUPLICATE ARTWORK - RE-CHECKED ON ALL 512","picked":"The original duplicate hunt in Theory 22 only had 483 of the 512 animation files. Theory 63 recovered the rest from Pindar's archives, so we could finally run the check on the complete collection.","tested":"We took a digital fingerprint of all 512 complete animation files and grouped them by fingerprint. Any two files with the same fingerprint are identical byte for byte.","happened":"#14 and #75 remain the ONLY byte-identical pair in the entire collection, now confirmed across all 512 files rather than 483. The finding from Theory 22 survives the wider check. Separately, six artworks (#83, #147, #159, #162, #174 and #180) contain repeated pictures INSIDE their own animation, which is a different thing and feeds the glitch work in Theory 64.","status":"tested","links":"[]","media":"[]","sort":74},{"id":76,"num":76,"title":"THE BIGGEST OPEN PROBLEM - OUR TWO GLITCH TESTS DISAGREE","picked":"We have two different tests for what counts as a glitch, and they point at different artworks. The OLD test looked for pictures that barely change from the one before them, in other words the animation slowing down or freezing. The NEW test looks for one single picture coming back again and again at gaps that keep shrinking. Both tests are careful, both were checked against examples, and neither has been proved wrong.","tested":"We compared the two lists side by side. The old test's clearest finds were #7, #43, #67, #71, #231, #256, #274, #474 and #479. The new test finds 46 artworks in total.","happened":"ONLY FOUR ARTWORKS ARE FOUND BY BOTH: #7, #43, #67 and #71. The old test's other five strong finds - #231, #256, #274, #474 and #479 - are not picked up by the new test at all. That is uncomfortable, because four of those five are literally labelled Type=Glitch and belong to the glitchGAN group (#231 Frozen glitchGAN, #256 Inverted glitchGAN, #274 Glowing glitchGAN, #479 Inverted glitchGAN). Meanwhile the four that both tests agree on are all Shimmer skullGANs and none of them is labelled a glitch at all. So the two tests are not just disagreeing at the edges - they disagree about the artworks Pindar himself named after glitches. WE ARE NOT PICKING A WINNER. Until we know which idea of a glitch Pindar actually meant, we cannot be confident that ANY candidate group of eleven is the right one, including the backwards-Fibonacci eleven in Theory 65. This is the single biggest unresolved question in the project. [2026-08-25 partial resolution: #474 is now CLEARED as a false positive of the old quiet-frame detector (clean on every modern test), and #5's recurrence is a loop-wraparound artifact. The disagreement between detectors shrinks accordingly.]","status":"tested","links":"[]","media":"[]","sort":75},{"id":77,"num":77,"title":"SPELLING TRICKS ON THE WORDS - CLOSED FOR GOOD","picked":"This is not a way of picking artworks. It is a way of changing the word an artwork gives you. The idea was that Pindar took each artwork's essence word and bent it - spelled it backwards, swapped each letter for its opposite (that is called an atbash), or slid every letter along the alphabet by a fixed number of steps (that is called a Caesar shift). We built 67 word-to-word changes in one library and measured each one against the whole 2,048-word BIP-39 list before using it on anything.","tested":"First we counted how often each change turns a BIP-39 word into another BIP-39 word. Only 15 of the 2,048 words spell a second BIP-39 word backwards. Only 6 survive an atbash. Fourteen of the 25 Caesar shifts map nothing at all. Then we ran the falsifiable half of the library over 1,232 candidate groups of eleven that were built independently of each other, which is 13,552 chances for the spelling family to work.","happened":"IT FIRED ZERO TIMES. Not once in 13,552 chances. The arithmetic said this would happen and the sweep proved it. A seed phrase needs eleven words, and there are not enough BIP-39 words that survive these changes for eleven of them to line up. Pindar COULD NOT have built the seed this way even if he wanted to. This family is closed permanently. Nobody should test it again.","status":"tested","links":"[]","media":"[]","sort":76},{"id":78,"num":78,"title":"DATA HIDDEN INSIDE THE GIF FILES - EXCLUDED ACROSS ALL 512","picked":"A GIF is a file, not just a picture, and a file has places to hide things. We treated all 512 animations as byte streams and looked in every one of those places. The last 108 files, the spookyGANs and solGANs, were downloaded from Pindar's own Arweave archive to close the gap, so this covers the whole collection with nothing left over.","tested":"We walked every file block by block. We checked for comment blocks, plain-text blocks, unknown block types, extra bytes stuck on the end, and per-frame colour tables. We compared each frame's packing against what a normal encoder must produce. We read the lowest bit of every palette colour and every picture pixel, forwards, backwards and column-first, and tested the result for readable text. We ran the Westfeld chi-square test, which is the standard detector for hidden data in low bits.","happened":"NOTHING, EVERYWHERE. No comment blocks. No plain-text blocks. No unknown block types. No trailing bytes. No per-frame colour tables. All 26,781 frames of 26,781 are packed correctly, so no file was opened and patched by hand. Palette order is random. Palette low bits are noise. The lowest Westfeld score in the whole collection is more than 100 times higher than hiding data would produce. The GIF container is clean.","status":"open","links":"[]","media":"[]","sort":77},{"id":79,"num":79,"title":"HANDWRITTEN TEXT PAINTED INTO THE ARTWORKS","picked":"The byteGAN puzzle was solved by finding seed words painted faintly into a GIF's frames. So we asked whether any bitGAN carries text the same way. First we calibrated on a known positive: Pindar's own roboGAN clue image, which really does carry his handwriting. The handwriting shows up clearly in a standard-deviation projection, which measures how much each pixel changes across the frames. It is almost erased by a median projection, which is what the project's old composites folder used.","tested":"We rebuilt eight projection types straight from the source GIFs at their full 512 by 512 size: minimum, 10th percentile, median, mean, 90th percentile, maximum, standard deviation and range, plus CLAHE local-contrast and high-pass versions. We tiled them into contact sheets and looked at all 512 artworks in both the standard-deviation set and the CLAHE set, with no sampling, and re-checked anything mark-like at full size.","happened":"NO ARTWORK CARRIES HIDDEN HANDWRITING. Nothing like a pen stroke, a written word or a faded note appears anywhere, in any projection, at any contrast. One word IS painted into the art, and it is 'TEST'. It is tiled across backgrounds and fitted into the subjects' teeth. It is detectable in well over a hundred artworks, mostly in the ghostGAN range. It is a watermark the generator copied from its training pictures, not a clue. One word shared by a third of the collection cannot pick out eleven artworks.","status":"tested","links":"[]","media":"[]","sort":78},{"id":80,"num":80,"title":"HIDDEN TEXT IN THE PUZZLE PAGE'S OWN IMAGES","picked":"The Act 1 Puzzle page shows four GIFs, one for each cryptoArtist: st.gif for StellaBelle, br.gif for Brinkman, ro.gif for SuperRareRoses and co.gif for Colborn. The page also carries a fifth image that a reader never sees, the social-card image gitm_3__1001.gif. That fifth one was the single most promising unopened thing on the page, because the byteGAN puzzle was solved twice over from exactly this kind of image.","tested":"We pulled all five files and built the same projections that DID reveal Pindar's handwriting on his roboGAN clue image, above all the standard-deviation projection, which measures how much each pixel changes across the frames. clue_subtle_888.gif was run through the identical pipeline as a positive control, and it read correctly: the artist's handwriting came out as white writing on a black field. The projection files are saved in clue_assets as page4_std.png for the four body GIFs and gitm_3__1001_std.png for the social card, so the check can be repeated.","happened":"VERIFIED NEGATIVE. No hidden text in st.gif, br.gif, ro.gif, co.gif or the gitm_3__1001.gif social card. The test is proven to work on this artist, because the positive control run alongside it made his handwriting blaze out. It found nothing on any of the five puzzle-page images. This closes the single most promising unopened thing on the page, since the byteGAN puzzle was solved twice over from exactly this kind of image.","status":"open","links":"[]","media":"[]","sort":79},{"id":81,"num":81,"title":"THE PODGALLERY WORD SWAP APPLIED TO ACT 1","picked":"Pindar's podGALLERY puzzle worked by a two-step hop. You take the word in a trait, find its position on the SLIP-39 word list, and read the BIP-39 word sitting at that same position. He has used the trick once, so it was fair to ask whether he used it again here.","tested":"We tested every artwork in both directions. Does any artwork's essence equal the hop applied to the artwork's own register word? Does any essence sit at the same position on both word lists, which would make it its own image? We also checked how many artworks the hop is even defined on.","happened":"ZERO ADHERENTS ANYWHERE IN ACT 1. Not one of the 513 artworks satisfies the hop in either direction. 140 artworks have an essence that happens to be a SLIP-39 word, so the hop CAN be computed on 140 pieces, but it is never the rule that filled the field. Pindar reused the vocabulary of that puzzle and did not reuse the mechanism. Separately, the hop reduces to 'the BIP-39 word at your own number minus one', which is exactly what the skullGAN essences already are - so applying it to a skullGAN gives nothing new.","status":"tested","links":"[]","media":"[]","sort":80},{"id":82,"num":82,"title":"A CLOSED LOOP OF ELEVEN ARTWORKS POINTING AT EACH OTHER","picked":"This was the prettiest idea anyone had. If eleven artworks point at each other in a ring, then the ring picks the eleven AND puts them in order at the same time, and the ordering problem disappears. Every essence is a BIP-39 word, every BIP-39 word has a number, and every number under 513 is an artwork number, so the collection can be read as artworks naming other artworks.","tested":"We built sixteen different artwork-to-artwork maps and ran a full loop count on every one. They include the essence-number map at both counting bases, the same map with the podGALLERY hop inserted, the map applied twice, the forge burn pairing, mint position read as an artwork number, frame count read as an artwork number, and the title code misread as decimal.","happened":"NO LOOP OF ANY LENGTH EXISTS, with one exception that does not count. The only loops anywhere are inside the Forge, where 'burnt with' is an involution - if A is burnt with B then B is burnt with A - so it can only ever make loops of two. There are 19 of those and they are loops by construction, not by design. There is no loop of three, no loop of eleven, and nothing in between. Whatever Pindar built, he did not build it out of the essence field, the numbering, the mint order, the frame counts or the SLIP-39 list.","status":"tested","links":"[]","media":"[]","sort":81},{"id":83,"num":83,"title":"CALIBRATION LESSON 1 - 'ELEVEN VALID BIP-39 WORDS' MEANS NOTHING","picked":"This is the most valuable thing produced today and it changes how every earlier row should be read. All 512 artworks carry an essence that is a valid BIP-39 word. 511 are literally on the list, and the 512th, #191, holds err0r, which reads as error under the leetspeak convention Pindar uses everywhere else. There is no artwork you can pick that fails to supply a word.","tested":"The consequence is arithmetic, not opinion. There are 1.4 x 10^22 ways to choose eleven artworks out of 512, and EVERY SINGLE ONE of them yields eleven valid BIP-39 words. To show this rather than assert it, we ran 48 pools crossed with 137 stated position rules. 222 of those combinations landed on exactly eleven members.","happened":"ALL 222 PRODUCED ELEVEN VALID BIP-39 WORDS, ALL DISTINCT. NOT ONE WAS REJECTED. The test 'does this set give eleven legal seed words' has exactly zero power to tell good sets from bad ones when the words come from the essence field. Any earlier row in this sheet that offers 'and all eleven words are valid BIP-39' as support is offering nothing at all. What is left is the motivation of the pool, the motivation of the rule, and a wallet test that either fires or does not. Nothing in between counts.","status":"tested","links":"[]","media":"[]","sort":82},{"id":84,"num":84,"title":"CALIBRATION LESSON 2 - HOW EASY IT IS TO LAND ON ELEVEN","picked":"Rather than picking mechanisms by inspiration, we enumerated the whole space of them. A mechanism here is a recipe: pick a scope, pick a property, pick how to pull words out of that property, pick a hop to apply, pick a filter, pick an order. Eight scopes, 27 properties, 5 extractors, 11 hops, 2 filters and 2 orderings.","tested":"That is 23,760 stated mechanisms, and we ran every one of them. Pindar's own roboGAN move - keep only the entries that are valid BIP-39 words - is applied inside all 23,760.","happened":"113 of the 23,760 land on exactly eleven, and those 113 collapse to just 22 different answers. All 22 were tested at up to ten orderings each. None matched. The instructive part is WHY 113 exist. Twenty of them exist only because the collection happens to use exactly eleven distinct Eyes values, so 'the Eyes value of every artwork read as a BIP-39 number, distinct values only' gives eleven. That is a fact about how many eye counts the artist happened to draw. It is not a fact about the puzzle.","status":"tested","links":"[]","media":"[]","sort":83},{"id":85,"num":85,"title":"CALIBRATION LESSON 3 - STATE THE RULE BEFORE YOU COUNT","picked":"The honest way to test whether eleven is special is to write a selection rule down BEFORE looking at what it returns, and then accept whatever number comes back. We did that fourteen times, across the Act 1 Review drop and the whole collection, and every one of the fourteen counts is printed in the source table.","tested":"The rules were ordinary and stateable. Examples: the title's leading adjective differs from the SubType field. The essence is on the SLIP-39 list as well. The essence number is above 1023. The essence number equals the artwork's own number. The title contains a leetspeak digit inside a word. Each was written down first, then run.","happened":"The counts came back as 0, 1, 4, 10, 13, 21, 24, 32, 41 and 47. NEVER ELEVEN. Not once in fourteen tries. Set that against the 113 eleven-landing combinations in the row above and the lesson is plain: eleven turns up readily when you sweep a big mechanised space, and hardly ever when you state a rule in advance and take its answer. Stating the rule before counting does all the work. Any candidate that arrived by sweeping should be discounted accordingly.","status":"tested","links":"[]","media":"[]","sort":84},{"id":86,"num":86,"title":"CALIBRATION LESSON 4 - CHANCE PRODUCES ELEVENS AT THE EXPECTED RATE","picked":"A direct measurement of how often eleven turns up by luck. We took every artwork sitting at a Fibonacci position inside its own drop, using the counting base Pindar prints on the artwork titles themselves. That gives a pool of 58 artworks. The pool was fixed before any of this was tried.","tested":"The master dataset tags each artwork with anomaly categories - filename typo, leet-corrupted title, burnt in the forge, HiddenType F, named after a real person, and so on. Sixteen categories apply inside the pool. We took every union of one, two or three of them, which is 696 combinations, and counted how many land on exactly eleven artworks.","happened":"48 OF 696 LAND ON EXACTLY ELEVEN. That is about 7%, which is roughly what chance predicts for a 58-item pool. It is not a discovery, it is a background rate. All 48 were tested anyway, at ten orderings each, which is 61,440 seed phrases against 33 wallet recipes. All clean. Five of the 48 turned out to have the same eleven members as an earlier union under a different name, so the 48 labels cover only 43 genuinely different sets.","status":"tested","links":"[]","media":"[]","sort":85},{"id":87,"num":87,"title":"CALIBRATION LESSON 5 - WE HAVE LESS BREADTH THAN WE THOUGHT","picked":"Every candidate group of eleven words this project has ever generated now goes into one registry file, theories/_registry.json, keyed by the eleven words themselves. Any group that arrives under a new label but with the same words as an old one lands on the same key.","tested":"We counted the labels and we counted the keys.","happened":"The registry holds 4,570 entries but only 3,786 DISTINCT word-sets. One single set of eleven words appears under 58 different labels. Roughly 17% of our apparent breadth is the same answers arriving again under new names. Anybody counting candidates from the JSON files should collapse duplicates before quoting a number, and anybody impressed by a large candidate count should knock about a sixth off it first.","status":"tested","links":"[]","media":"[]","sort":86},{"id":88,"num":88,"title":"CORRECTION TO THEORY 72 - 'ORDER IS CLOSED' COVERS 0.0488%","picked":"Theory 72 and the 'What We Have Ruled Out' sheet both say that for eighteen groups every one of the 39,916,800 orderings was tried, and that 'a wrong order can never again be the excuse'. That is the largest overstatement in the record and it needs saying plainly.","tested":"Here is the arithmetic. The program fixed12.py pinned the twelfth word to the literal BIP-39 word 'random' and kept only the orderings where that word satisfies the checksum, which is one in sixteen. So each group produced about 2.49 million derived phrases. The complete space for one group is 39,916,800 orderings x 128 legal twelfth words = 5,109,350,400 phrases.","happened":"EACH OF THE EIGHTEEN GROUPS WAS COVERED TO 0.0488% - one twentieth of one percent. The claim is true in a narrow technical sense and misleading in every practical sense. It closes order ONLY IF the twelfth word is the literal string 'random'. That reading is a pun on the clue, not a deduction from it. The clue says the phrase ends with 'a random word', which most naturally means an arbitrary word with no meaning. Those runs also used only 3 wallet recipes. Eighteen groups are not closed. They are barely scratched.","status":"tested","links":"[]","media":"[]","sort":87},{"id":89,"num":89,"title":"CORRECTION TO THEORY 70 - FOUR SWEEPS WERE NOT COMPLETE, AND NOW ARE","picked":"Four sweeps - pod553, pod553b, mintpod and meta11 - were recorded as complete. The Methodology sheet said only that they 'have no final DONE line'. That understated it badly. Their resume files record exactly how much was left.","tested":"pod553 had covered 3,654 of 12,382 tasks, which is 29.5%. pod553b had covered 3,634 of 12,396, which is 29.3%. mintpod had covered 5,310 of 12,624, which is 42.1%. meta11 had covered 2,974 of 4,373, which is 68.0%. Across all four that is 15,572 of 41,775 tasks, or 37.3%. Worse, the work queue holding the missing 26,203 tasks was then deleted, because queue_build.py starts by wiping its own directory.","happened":"THE MISSING WORK IS NOW DONE. All four resume files survived on disk and their lengths sum to exactly 26,203, matching the audit. We rebuilt the queue from them and ran all 26,203 remaining tasks: 3,353,984 seed phrases against 33 wallet recipes, which is 110,681,472 addresses. Zero hits. The four sweeps now stand at 100% of the 41,775 tasks they queued, against the 37.3% they actually had. Under the reading that the twelfth word goes last and only last, this family is genuinely closed at the orderings it queued.","status":"tested","links":"[]","media":"[]","sort":88},{"id":90,"num":90,"title":"CORRECTION - EXTRA WALLET RECIPES ARE NEARLY FREE AND WE WERE NOT USING THEM","picked":"A seed phrase does not point at one wallet address. It points at a whole tree of them, and which address you get depends on the recipe, called a derivation path. Every earlier sweep used 3 or 5 recipes. The program pathsweep.py has known 78 of them all along, and had only ever been pointed at hand-typed phrases, never used inside a sweep.","tested":"We measured the cost. Turning a phrase into a seed uses a deliberately slow step called PBKDF2, and that costs 1.42 milliseconds. Once you have paid it, each ADDITIONAL recipe costs only 0.167 milliseconds. So testing 33 recipes instead of 3 does not cost eleven times as much. It costs well under three times as much.","happened":"This is the best value in the whole project and we had been leaving it on the table. Every job run today used 33 recipes, and 18 of those 33 had never been tried by any earlier sweep - address numbers 3 to 9, Ledger account numbers 2 to 9, the change chain, and three of the bare old layouts. So today is the first time any of these candidate sets has been seen at an address number above one or a Ledger account above one. Still zero hits, but the coverage is real and it was cheap.","status":"tested","links":"[]","media":"[]","sort":89},{"id":91,"num":91,"title":"CORRECTION - LANDING ON ELEVEN WAS TREATED AS EVIDENCE THROUGHOUT","picked":"Read back through rows 1 to 76 and a habit shows up. Several sets were tested BECAUSE they reached eleven, and their negative results are then written up as if they eliminated something general. They do not. They eliminate one set of eleven words in a handful of orderings and nothing more.","tested":"Rows 83 to 87 above give the measurements. Every set of eleven artworks gives eleven valid words. 222 of 222 rule-and-pool combinations that reached eleven were accepted. 113 of 23,760 enumerated mechanisms reach eleven. 48 of 696 anomaly unions reach eleven, which is the chance rate.","happened":"LANDING ON ELEVEN IS NOT EVIDENCE. It is the background. This warning has been added to the Start Here sheet as the fifth row, so that anyone opening this workbook meets it before they meet any candidate. The counter-discipline the project already had - hapax_catalogue.md, gif_structure.md and missing_drops.md all refused to write a set list rather than offer an unmotivated eleven - is the right habit and should be kept.","status":"open","links":"[]","media":"[]","sort":90},{"id":92,"num":92,"title":"THE HIDDEN TYPE LETTER CLASSES","picked":"Pindar published a classification for every artwork based on what kind of number it sits on: Fibonacci gets 'a', remaining primes 'b', squares 'c', remaining odd numbers 'd', everything else 'e'. The letters are lowercase in the skullGAN drop and uppercase in the drops after it, and there is an undocumented sixth uppercase value 'F'. That is exactly eleven letter values - a b c d e A B C D E F - with no lowercase f anywhere. Eleven is how many values the field takes, not a number we chose.","tested":"We took the eleven classes as the source and read one artwork out of each. Two clean ways to do it: the lowest-numbered member of each class, which gives 'abandon absorb absurd achieve absent quantum dance math flame ghost alpha', and the highest-numbered, which gives 'arch asthma anxiety auto avocado grain grab govern grape grant glue'. Both were tested. Separately, the field has exactly two real defects - #32, which holds the literal string '32' where a class letter belongs, and #187, typed C when the number rule demands E - and combining those two with the nine F-carrying artworks also makes eleven.","happened":"No match on any reading. Two notes for the record. First, #55's blank Hidden Type is NOT a missing 'a'; it is a fetch artefact, the same one that leaves #119 blank. Second, we flag the defects-plus-F eleven as bent. The F class is not a defect at all, it is the palindrome marker - RoboboR, Dr Awkward, Taco Cat, Dog God, Star Rats, Evil Olive, Flow Wolf, CalcuclaC. Adding two genuine errors to nine correctly-classified palindromes reaches eleven by putting two different kinds of thing in one bag.","status":"ruled-out","links":"[]","media":"[]","sort":91},{"id":93,"num":93,"title":"THE ELEVEN FIBONACCI ARTWORKS PINDAR TOLD HOLDERS TO KEEP","picked":"In 'Hidden Types' Pindar named eleven skullGANs and told the community to hold them because they 'will unlock special things in upcoming drops'. They are the artworks at Fibonacci positions in his own numbering, which starts at zero: #0, #1, #2, #3, #5, #8, #13, #21, #34, #55, #89. This is the only group of exactly eleven the artist has ever named himself.","tested":"Their essences are 'universe abandon ability able above abstract account bean aerobic almost arch'. But #21 is one of only two breaks in the skullGAN register: the rule 'essence = BIP-39 word at your number minus one' demands 'action' there and the artwork carries 'bean' instead. 'action' is the one BIP-39 word that a stated rule of this collection demands and no artwork anywhere supplies. So we tested the repaired version with 'action' in the eighth slot as well as the raw version with 'bean'.","happened":"No match on either version, and no falsifiable word transform survives on either. One structural finding came out of it and is worth keeping. Under the change 'subtract one from the word number', the repaired set becomes 'unit zoo abandon ability about absorb accident act advice allow april'. Under the change 'read the word number as an artwork number and take that artwork's essence', ten of the eleven give the same answer. Those two agree because they are the SAME FACT about the skullGAN register, not because they converge on a solution.","status":"ruled-out","links":"[]","media":"[]","sort":92},{"id":94,"num":94,"title":"THE GHOSTS DROP AT FIBONACCI POSITIONS - THE STRONGEST NEW CANDIDATE","picked":"A pool contains exactly eleven Fibonacci positions only if it has the right size, and the size window is narrow. Counting from one, a pool of F(n) members holds exactly n-1 Fibonacci positions. The Ghosts in the Machine drop has 144 members, and 144 is F(12), so it holds exactly eleven. The drop size IS the count. Pindar could have made that drop 140 or 150 pieces. He made it 144.","tested":"The counting base is not our choice. All 144 Ghosts titles print (n/144) with n exactly equal to the artwork number minus 300, with no exceptions, and those titles are on the blockchain, not just in our spreadsheet. So the drop is 1-indexed by the artist's own hand. Positions 1, 2, 3, 5, 8, 13, 21, 34, 55, 89 and 144 give #301, #302, #303, #305, #308, #313, #321, #334, #355, #389 and #444, essences 'exact gravity hello joke sting example nothing emerge off correct danger'. Tested at sixteen stated orderings, all 33 wallet recipes.","happened":"No match at those sixteen orderings. What makes this the strongest new candidate is that the same rule REPRODUCES A PUBLISHED ANSWER: applied to the skullGAN drop, whose number space runs 0 to 128, it returns exactly the eleven artworks Pindar told people to hold. No other rule in the project passes that test. The set also opens on the drop's first piece and closes on its last. THE CASE AGAINST IT: the 144 ghostGANs contain not one splice. That drop does not glitch at all in the animation sense. This set is only right if 'glitch' means the Fibonacci pattern rather than the visible defect. That fork is unresolved and the whole puzzle turns on it.","status":"ruled-out","links":"[]","media":"[]","sort":93},{"id":95,"num":95,"title":"THE GHOSTS FIXED-POINT ELEVEN","picked":"The skullGAN drop follows a register: the essence of artwork g is the BIP-39 word at position g minus one. Later drops abandoned it. So we ran a census asking which artworks anywhere in the collection still satisfy it. Not a filter looking for eleven - a census, with eleven as the output.","tested":"The map has 137 fixed points. 126 of them are skullGANs, which are fixed trivially because that drop IS the register. Remove them and exactly eleven remain, and all eleven are in the Ghosts in the Machine drop: #307, #322, #348, #368, #372, #380, #389, #404, #407, #429, #444, essences 'chaos choice clock combine common control correct crash crazy culture danger'. That is eleven pieces from a later drop whose essence field was filled in by the EARLIER drop's rule, which is a data-corruption glitch in exactly the sense the clue uses the word.","happened":"No match at ten orderings against 33 wallet recipes, and no match under a bounded perturbation sweep of 919 orderings. Two honest problems. First, the eleven sit inside a band of fifteen: #354, #360, #423 and #440 miss the diagonal by one or two, and nothing else in the drop comes within eight. So the boundary is drawn by an equality test we chose. Second, this set has effectively ONE stateable ordering - artwork order, mint order, drop-local number, BIP-39 position and alphabetical order are all the same sequence, and every member has 55 frames. If it is the answer, the ordering cannot come from the collection at all. It was queued for a full ordering sweep. THAT QUEUE HAS SINCE BEEN CANCELLED - see the Methodology sheet - because every ordering anyone can state had already been tested on this set.","status":"ruled-out","links":"[]","media":"[]","sort":94},{"id":96,"num":96,"title":"THE PERFECT SQUARES, MARKED TWICE","picked":"Pindar's Hidden Type C class is 'squares'. Inside the skullGAN drop the squares at or below 128 are 1, 4, 9, 16, 25, 36, 49, 64, 81, 100 and 121 - eleven of them, with no tuning. The squares that are also Fibonacci get type 'a' instead, so they are not in this class.","tested":"The new finding is that the class is marked a SECOND time, in a place nobody had written down. Nine of the 128 skullGAN filenames carry an 's_' immediately after the number: 004, 009, 016, 025, 036, 49, 064, 081, 100. Those are exactly the perfect squares. The 's_' is a square marker living in the filename rather than in the trait table, and it is completely undocumented. Their essences are 'abandon about absurd acid adapt afford alcohol among anxiety arrest audit'. Tested as given, plus variants dropping #1 and adding #130.","happened":"No match. Two loose ends worth keeping. #121 is the only perfect square whose filename LACKS the s_ marker, and its Hidden Type was never fetched from the blockchain, so we cannot check whether the trait table fails it too. And #49's filename, 49_s_flaming_smoke.gif, is the only file in the 128-file archive whose number is not padded to three digits, so sorted alphabetically it lands after 128 at the end of the drop. It was queued for a full ordering sweep. THAT QUEUE HAS SINCE BEEN CANCELLED - see the Methodology sheet - because every ordering anyone can state had already been tested on this set.","status":"ruled-out","links":"[]","media":"[]","sort":95},{"id":97,"num":97,"title":"THE ELEVEN FORGED SUPERRARE PIECES THAT EXISTED ON THE DAY OF THE CLUE","picked":"Collectors could burn two bitGANs and get back one new SuperRare 1/1. Those forged pieces live on their own contract, 0x2D413006b09DC8953a468E662d81382942b4C0D2, whose token name is literally 'SuperRare bitGANs'. It holds 25 tokens today. But on 5 February 2022, the day Pindar posted the Act 1 clue, exactly ELEVEN of them existed. Token 11 was minted 8 January 2022. Token 12 was not minted until 23 May 2022, three and a half months after the clue.","tested":"Pindar confirmed the count himself two days later. His own Act 1 inventory in journal page 016 lists '12 SuperRare bitGANs' - the eleven numbered pieces plus The Cryptographer, which sits on a different contract and carries no attributes. All eleven carry a clean BIP-39 essence. In mint order the words are 'super rare series ritual level secret update switch symbol upgrade chief'. Tested at mint order, token id order, stroke count and their reverses.","happened":"No match. This eleven has a motivation nobody chose: a contract boundary that closed at exactly eleven members on exactly the announcement date. But it has a hard problem. Running the hops backwards shows the eleven words have NO PREIMAGE in Act 1 - not one is the essence of any of the 513 artworks. The forged pieces use a separate vocabulary. So if this is the answer, the ordering cannot come from the collection at all, and only the SuperRare contract can supply one. It was queued for a full ordering sweep. THAT QUEUE HAS SINCE BEEN CANCELLED - see the Methodology sheet - because every ordering anyone can state had already been tested on this set.","status":"ruled-out","links":"[{\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":96},{"id":98,"num":98,"title":"THE SHEET-CORRUPTED WORDS ELEVEN","picked":"The puzzle was announced on 5 February 2022, after every drop had shipped. So Pindar did not plant eleven glitches and build artworks round them. He looked over a year of finished work, found things that were already odd, and declared them the puzzle. That reframing changes which artefact matters: not the blockchain, but the properties file he published and had already pointed solvers at once before.","tested":"So we asked one question of bitgan_properties.csv: which cells hold text that is corrupted - a letter transposed, a letter dropped, a space dropped, a character substituted, or a word in the wrong case? Exactly eleven artworks answer, and no others: #73, #123, #128, #154, #170, #175, #191, #205, #363, #383, #473. Their essences are 'animal aunt avocado smoke quiz skull error grace true style false'. Eleven is the complete extension of one predicate, so no member was chosen by hand.","happened":"No match. Four related variants were also built and tested - a strict nine-member reading, the rows that break their own drop's rules, one instance of each way a string can be damaged, and a hybrid for a looser reading of 'the sheet'. All clean. It was queued for a full ordering sweep. THAT QUEUE HAS SINCE BEEN CANCELLED - see the Methodology sheet - because every ordering anyone can state had already been tested on this set. Honest caveat from the source document: whether #383 and #128 count as corrupted words is a judgement, and if both go, the set has nine members.","status":"ruled-out","links":"[]","media":"[]","sort":97},{"id":99,"num":99,"title":"THE TERMINAL ELEVEN, #501 TO #511","picked":"#512, the Quantum glitchGAN, carries the essence 'final' and is the last artwork in the collection. So the eleven artworks standing immediately before it, #501 to #511, form an eleven that the collection itself counts down to. Their essences are 'project source broken weird elder wild cancel south network capital negative', which is glitch vocabulary end to end.","tested":"This also solves a problem the workbook had left open. The sub-run #505 to #511 was recorded as a 'CORRUPTED seven' that was never swept, with a note that there was no principled way to find four more. The terminator gives one: take the eleven that end at the terminator. We checked the whole corpus of candidate sets and this exact eleven appears nowhere else. It was tested at ten orderings, under every falsifiable transform in the library, and then at 88 deep stateable orderings, in three phrase modes - free ending, 'final' pinned as word twelve, and as a complete twelve-word phrase.","happened":"No match in any mode. The set collapses to only 88 stateable orderings because #501 to #511 are consecutive, share a drop and agree on most columns, so many sort keys produce the same sequence. Its number order, mint order and artwork order are all the same order. When 'final' was pinned as word twelve, only 68 of 1,128 orderings were checksum-legal, which is 6.0% against the 6.25% predicted - a small check that the arithmetic is behaving.","status":"ruled-out","links":"[]","media":"[]","sort":98},{"id":100,"num":100,"title":"THE RARE-FRAME ELEVEN","picked":"Almost every artwork in a drop has the same number of frames - 41 for skullGANs, 55 for most of the rest. So an unusual frame count is a genuine oddity. The rule was stated before counting: take every artwork whose frame count is shared by three or fewer artworks in the whole collection, then remove the two that carry 'random' and 'final', because the clue treats those two words as belonging to the twelfth slot rather than the eleven.","tested":"That returns exactly eleven artworks: #129, #146, #159, #162, #163, #165, #228, #445, #446, #467, #469. Their essences are 'alpha hollow dog twin time spirit barely collect rose give calm'. The set includes #228, which five independent encoder fields single out as built unlike anything else in the collection, and #146, one of only three files whose frame timing varies within itself.","happened":"No match at any ordering tested. This was the first candidate in a master priority queue that was going to try every one of its 39,916,800 orderings against all 128 legal twelfth words. THAT QUEUE WAS CANCELLED after 212 of its 63,872 blocks - about 200 blocks of the 7,984 this set alone needs. The reasoning is on the Methodology sheet: every ordering anyone can state had already been tested, so the arrangements left are ones no rule in the collection can produce, and an answer found there could not be explained even if it hit. Nothing should be read into the partial result.","status":"ruled-out","links":"[]","media":"[]","sort":99},{"id":101,"num":101,"title":"THE 22 PUZZLE-PAGE WORDS THAT ARE ALSO ESSENCES","picked":"The Act 1 Puzzle page contains 45 BIP-39 word tokens, 34 of them distinct. Twenty-two of those 34 are also the essence of a bitGAN. That looks striking, and the honest test kills most of the excitement: the same measurement across 101 other archived pages gives a median rate of 0.40, and this page scores 0.63. It tops the list but sits inside the ordinary spread, because the essence list is full of short common English words.","tested":"We tested the 22 anyway, because 22 is exactly two elevens. Twelve different splits were run: the first eleven and last eleven in page order, the first and last eleven in artwork order, the odd and even positions in both orders, and the four 'drop random first' variants of the same. Separately we ran six inclusive and exclusive readings - all 22, all 35 with the title and byline, and versions with the clue words or the puzzle-meta words removed - each at six slices and five orderings.","happened":"No match on any of the twelve splits or any of the six readings. Total 18,436 seed phrases on the splits and 20,480 on the inclusive readings. The most seductive coincidence on that page is that the last two BIP-39 tokens of the body are 'random' and 'word' - which is a coincidence, because the clue sentence itself ends with the phrase 'a random word'. Three other pages on the site also yield exactly eleven BIP-39 words under some rule, and all three are coincidences of length. All were tested and all fail.","status":"ruled-out","links":"[]","media":"[]","sort":100},{"id":102,"num":102,"title":"THE DROP DESCRIPTIONS AS A WORD SOURCE","picked":"Every bitGAN carries a description on OpenSea and nobody in this project had ever read them. We recovered all 512 from the chain, using the metadata URL the contract itself publishes. No API key was needed. The sweep also identified fourteen artworks that the old token index had lost, #457 to #466 and #497 to #500, because they are titled in their reprised drop's numbering rather than with the usual (n/512) suffix.","tested":"We pulled every BIP-39 word out of each drop's description. Five of the six drops give nine or ten distinct words. The ghostGAN description gives exactly ELEVEN: 'series art there theme this color find more about van act'. We built and tested that eleven, plus several variants that swap one member for a marked artwork - #445 the prize, #191 err0r, and #32 - and the radix-notation eleven that came out of the same work.","happened":"No match. The ghostGAN eleven should be held loosely: the count is an artefact of that template being the longest, and the words are ordinary marketing vocabulary. What the descriptions DID produce is three real facts. #130 is the only artwork in all of Act 1 whose description says the collection is 1024 animations where all 63 of its drop-mates say 2048, and #130 is the qubitGAN, essence 'quantum'. The word 'image' appears in exactly six descriptions, #301 to #306, and nowhere else. The word 'final' appears in exactly one, #512. #130's '1024' leaves no trace in any word extraction, which is a caution against treating word extraction as the only lens.","status":"ruled-out","links":"[]","media":"[]","sort":101},{"id":103,"num":103,"title":"ELEVEN-STEP WALKS THAT END ON #468, WHOSE ESSENCE IS 'random'","picked":"Artwork #468, the Flaming testGAN, carries the essence 'random'. That is the clue's own word, sitting on an artwork. If 'eleven glitches ending with a random word' is literal rather than a pun, then a walk visits twelve artworks, the twelfth is #468 and supplies word twelve, and its eleven predecessors are the glitches. A walk is valuable because it fixes the eleven words AND their order in one act, so it costs 128 tests instead of 128 times some number of orderings.","tested":"We ran every step rule backwards from #468. ONLY ONE RULE REACHES IT IN ELEVEN STEPS, and that rule produces exactly two chains. The rule is 'the next artwork is the one whose number equals this artwork's mint position, modulo 513'. Chain A runs #432 token, #57 mistake, #124 author, #199 gorilla, #295 brave, #406 power, #26 add, #91 area, #160 slender, #242 behave, #345 original, #468 random. Chain B differs in its first two links.","happened":"No match. Chain A IS a checksum-legal twelve-word BIP-39 phrase, and it does not open the wallet at any of the 33 recipes, nor does its reverse, nor the version with 'random' first. Chain B is not checksum-legal at all. The construction was extended to six termini - #468, #512, #445, #191, #32 and #119 - giving 346 walks in total, all tested, all clean. One thing worth noticing: #57, the second link of Chain A, is an independently flagged glitch and its essence 'mistake' is one of only two breaks in the skullGAN register. Also worth noticing: none of the 346 walks lands on any of the 1,263 elevens stated by any other line of attack. Zero coincidences.","status":"ruled-out","links":"[]","media":"[]","sort":102},{"id":104,"num":104,"title":"ORDERINGS READ OFF THE OBJECTS INSTEAD OF SORTED FROM A COLUMN","picked":"Every ordering this project had ever tried was either a sort on a column of our own dataset or a random shuffle. Nobody had tried an order READ OFF AN ARTEFACT. This closes that gap and it is a whole class of ordering, not a single theory.","tested":"The sources are the objects themselves: where the spliced frames fall inside each animation (first splice, last splice, splice count, smallest gap, largest gap, first gap, total gap); where the inverted frames fall; repeated-frame counts; the order files appear inside Pindar's Arweave archives; the Arweave filename string; the upload order that produced the 8x8 review contact sheet; the order of the burn transactions; the row order of Pindar's published bitgan_properties.csv; mint order; and the GIF's own loop length, byte size and frame count. Each sequence was also laid into grids 2 to 6 wide and read down the columns, boustrophedon, and as an inward spiral. Each was also read backwards.","happened":"No match. 3,361 orderings across 79 candidate sets, 430,208 seed phrases, 33 wallet recipes, 14,196,864 addresses. One finding is worth carrying: MINT ORDER IS THE SAME SEQUENCE AS ARTWORK ORDER for large parts of this collection, and for some sets Pindar's published CSV row order is the same again. Any job that sweeps 'artwork order and mint order' on those sets is testing the same arrangement twice and buying nothing. We confirmed this on 300 random completions and rebuilt one queue because of it.","status":"ruled-out","links":"[]","media":"[]","sort":103},{"id":105,"num":105,"title":"THE CORRUPTED SEVEN, COMPLETED FROM THE 48-ARTWORK SPLICE POOL","picked":"Artworks #505 to #511 form the group tagged 13 in the Arweave codes, and their essences are 'elder wild cancel south network capital negative'. Seven is not eleven, and the workbook had marked the set NOT YET SWEPT because there was no principled way to pick four more. That blocker has now weakened: the 48-artwork splice pool is a defensible pool, fixed by detector_conflict.md across a ninety-point threshold sweep, and it contains all four of Michael's confirmed positives and none of his seventeen confirmed negatives.","tested":"None of the seven is a member of the pool, so every quadruple drawn from it is a valid completion. C(48,4) = 194,580 completions. Each was swept in artwork order and in reverse artwork order. The brief's second ordering was replaced because it was vacuous: mint order and the published CSV row order are both identical to artwork order across all 48 pool artworks and all seven corrupted ones, so sweeping 'artwork order and mint order' would have tested the same arrangement twice.","happened":"No match. 389,160 orderings, all distinct, 49,812,480 seed phrases, 33 wallet recipes, 1,643,811,840 addresses. This is the single largest job of the day. Two independent counts agree exactly on the total: counting the orderings inside the 390 completed chunk files, and summing what the eight workers reported themselves. Both give 49,812,480, so no chunk was skipped and none was counted twice. The planted test seed was found. The same treatment was given to the true-glitcher nine, where the completion count is 780 rather than the 1,128 first stated, because eight of the nine are already in the pool.","status":"ruled-out","links":"[]","media":"[]","sort":104},{"id":106,"num":106,"title":"OPEN ITEM - FOUR RECORDS ARE BLANK BECAUSE NOBODY EVER FETCHED THEM","picked":"This is not a theory that failed. It is a job that takes minutes and has never been done, and three separate documents rest a top-ranked candidate on it. #119, #47, #121 and #192 have no on-chain traits in our files. Their records are EMPTY, not absent - bitgans_full_meta.json holds #119 as a name with an empty traits object, and #121 the same. Anyone filtering our data sees a blank and reads it as 'no value'. The truth is 'never asked'.","tested":"Why each one matters. #119 IS ONE OF ONLY TWO ARTWORKS PINDAR PUBLICLY PROMISED WAS FUTURE PUZZLE MATERIAL. On the Hidden Types page he wrote that (32/128) is Type 32 and that (119/128) is the only odd number of Type E, asked 'typo or important?', and then added: 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 archived pages, and the Act 1 Puzzle is the only puzzle left that could redeem it. We HAVE read the other one: #32's Hidden Type is the literal string '32'. So half the hint is confirmed and the other half is unread. #47 is the single artwork that breaks the prime-glitch rule, and the Prime Rule sheet records its Hidden Type as '(never fetched)'. #121 is the only perfect square whose filename lacks the s_ square marker, so we cannot check whether the trait table fails it too. #192 is one of the eight robot Hidden-Type breaks.","happened":"STILL OPEN. Cost is one API call per token and zero compute. It sits at number two on the audit's own priority list and it has still not been done. The wider gap is worse: 120 of 513 rows carry no on-chain traits - 105 that came back empty from Blockscout and 15 that were never in the token index at all (#0, #457 to #466, #497 to #500). Every conclusion in word_sources.md that the trait schema is 'too poor to carry the puzzle' was computed on 296 fetched tokens out of 512. Those conclusions are floors, not verdicts. The chain-corrupt eleven is the clearest casualty: it is marked ruled out, but its membership is a function of how many tokens we fetched, so it cannot be called ruled out until the fetch is complete.","status":"ruled-out","links":"[]","media":"[]","sort":105},{"id":107,"num":107,"title":"THE MONTAGE SIX-SETS COMPLETED TO ELEVEN","picked":"The montage work left two verified six-piece sets: the six frozen on a splice frame (7 19 31 37 73 244) and the six whose tiles seemed to show wrong content (143 155 187 223 294 357). We completed each to eleven using every five-piece subset of the ten disagreement-pool pieces, plus every leave-one-out eleven from the union of the two sixes.","tested":"Essence words and all four number-index readings (wordlist at GAN, GAN minus one, mint index, mint minus one), forward and reverse, 128 twelfth words, 33 wallet recipes. 660,480 seed phrases across the two queues. The plant seed was recovered in both runs.","happened":"No match. The montage sets do not complete to the eleven under any derivable word reading. See theory 108 for why the no-match six then dissolved entirely.","status":"ruled-out","links":"[]","media":"[{\"caption\": \"Pindar\\u2019s official montage image (all 512 pieces)\", \"url\": \"https://api.bitganspuzzlebook.com/media/theory/montage_all.jpg\"}]","sort":106},{"id":108,"num":108,"title":"THE MONTAGE NO-MATCH SIX WAS A WRONG-FRAME ARTIFACT - CHANNEL CLOSED","picked":"Michael hand-graded all 512 montage tiles and six tiles matched none of the frames we showed him. The open question was whether those tiles showed a DIFFERENT artwork's face - which would have been a planted identity glitch.","tested":"Every one of the six tiles was searched against all 26,781 frames of all 512 animations with a colour-aware metric, plus mirrored, negative, and mirrored-negative transforms, calibrated with 22 control tiles.","happened":"RESOLVED, and the answer is deflating: every tile's best match in the whole collection is its OWN animation, rank 1 of 512, at an in-between morph frame that our grading tool never sampled (we showed a handful of candidate frames, not all fifty). No montage tile wears another artwork's face. The montage channel is now fully explained: the splice six freeze on their glitch frame, the no-match six were an artifact of sparse sampling. Theories 60-ish montage entries and the 8/23 grading stand corrected by this.","status":"tested","links":"[]","media":"[{\"caption\": \"Pindar\\u2019s official montage image (all 512 pieces)\", \"url\": \"https://api.bitganspuzzlebook.com/media/theory/montage_all.jpg\"}]","sort":107},{"id":109,"num":109,"title":"THE PHOTOSHOP FINGERPRINT CHANNEL - A NEW EVIDENCE LAYER","picked":"Every GIF carries an embedded Photoshop XMP packet nobody had ever read. The packet's IDs are UUID-v1 values with decodable timestamps, so every file tells you exactly when Pindar's Photoshop saved it and which Photoshop version did it.","tested":"All 512 packets extracted and decoded. The corpus splits into 435 files saved once in their drop's original batch export, and 77 files Pindar RE-OPENED AND RE-SAVED later on newer Photoshop versions - fingerprints of post-export editing sessions. Cross-checked against every glitch definition we hold.","happened":"NEW EVIDENCE CHANNEL, the day's biggest find. Every confirmed splice glitcher and every negative-frame glitcher sits inside the 77, so the re-save class is a superset of the planted glitches. The prime-numbered skulls were authored in their own separate session (Photoshop 22.3, Aug 27 2021), confirming the prime rule from a second independent direction. The ghost-drop re-save batch is exactly the complete spaceGAN family (housekeeping, not anomaly picks). Three files re-saved with no detectable change remain unexplained: #203, #407, #454, #457, #460, #461. Three provenance-derived elevens (the final authoring session, the singleton saves, the dotted-and-glitched) were swept - final session to full stateable depth in five registers including save order, about 1.7M phrases - all negative.","status":"tested","links":"[]","media":"[]","sort":108},{"id":110,"num":110,"title":"THE RE-SAVED INVERSION TWELVE","picked":"All ten Inverted-subtype pieces in the re-save 77, plus the two non-Inverted negative-frame glitchers (231, 274) = twelve re-saved inversion pieces, one of which (#300 Lunar skullGAN, the only file in 512 with a layered-Photoshop lineage, saved alone on the winter solstice) reads as a capstone. Drop the capstone and eleven remain.","tested":"Leave-one-out across all twelve, plus every stateable ordering of the primary eleven, five word registers, 462,080 phrases, plant recovered.","happened":"No match. Also fails Michael's breadth criterion badly - ten of twelve are solstice pieces, and Pindar said the puzzle spans the whole collection.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":109},{"id":111,"num":111,"title":"THE RE-SAVED RAINBOW ELEVEN","picked":"SubType Rainbow intersected with the re-save 77 gives EXACTLY eleven pieces (29 53 61 71 83 203 291 407 457 460 466) spanning five of six drops, containing ground-truth glitcher #71 and absorbing both unexplained singleton saves (#203, #407). Two independent workers converged on this same set. #14, the one proven wrong-art piece, is itself a Rainbow.","tested":"Every stateable ordering (7,810) plus save-time order, five word registers, just under one million phrases, plant recovered.","happened":"No match. And a structural strike was found before the sweep finished: the robo drop has fourteen Rainbows (the Rainbot pun pieces) and Pindar re-saved none of them, so a Rainbow rule can never cover all six drops.","status":"ruled-out","links":"[]","media":"[]","sort":110},{"id":112,"num":112,"title":"CORRECTION - invsol9_solQ WAS NEVER PROPERLY CLOSED, NOW IT IS","picked":"The 77-audit showed the old inverted-solstice eleven {209 213 230 239 256 266 270 286 294 300 512} has ten of its eleven members inside the re-save 77, and its original closure used only the weak random-pinned sweep.","tested":"Re-swept at full stateable depth plus save-time order, five word registers, part of a 401,920-phrase queue, plant recovered.","happened":"No match. Properly closed this time. The revival logic was sound and the closure is now real.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":111},{"id":113,"num":113,"title":"THE TIMING GLITCHES - A NEW CONFIRMED GLITCH CLASS, AND THE TIMING ELEVEN","picked":"A frame-timing census found the collection is not uniform: 370 files run 100ms per frame, 138 run 0ms (all 128 skulls plus five robos and five review pieces at the wrong speed for their drops), one file (#228 Kevin aboschGAN) runs a unique 50ms, and two files carry ONE single held 100ms frame inside an otherwise 0ms animation: #479 holds frame 27, #480 holds frame 24. That held-frame pair is a deliberate mechanism no detector had defined - both are also in the re-save 77. The eleven timing-anomalous pieces outside the skull baseline: 155 176 180 182 190 228 446 467 468 469 512, whose essences include 'random' (#468) and end at 'final' (#512).","tested":"The eleven at full stateable depth plus every eleven from the fourteen-piece superset (adding 146, 479, 480), five registers, 21,190 orderings, 2.7 million phrases, plant recovered.","happened":"No match on the words, but the DISCOVERY stands: #479/#480 are confirmed glitches of a brand-new class (single held frame), now in the canonical registry. Loop counts and disposal methods were also censused - uniform across all 512, so the timing channel is fully harvested.","status":"ruled-out","links":"[]","media":"[]","sort":112},{"id":114,"num":114,"title":"THE IDENTITY-CORRUPTION ELEVEN - FIRST CANDIDATE EVER TO SPAN ALL SIX DROPS","picked":"The confirmed corruption pair gave the class: art that is not what its label says (#14 wears #75's flaming art; #390 was minted twice). A calibrated label-versus-art instrument - which independently re-found #14 as the number-two Rainbow outlier, validating itself - flagged strong contradictions (#179 a Rainbow with colourless ghost art, #423 a Rainbow in monochrome blue, #231 a warm mutating Frozen, #511 an Inverted on a white background whose essence is literally 'negative', #509 Dark Rainbow) and mediums (#204, #150, #490, #478, #92, #491). #150 and #179 are mutual label-swap partners - exchange their labels and both pieces' misfit collapses.","tested":"Core seven (2 confirmed + 5 strong) plus every choice of four from the eight mediums/alternates = 70 elevens at GAN order and reverse, and the primary eleven at full stateable depth, five registers, 15,010 orderings, 1.9 million phrases, plant recovered.","happened":"No match. The set remains the best-constructed eleven of the project (all six drops, byte-proven anchors) and the #150/#179 swap discovery stands on its own. If the word-selection rule is right but the reading is wrong, this membership is where to iterate.","status":"ruled-out","links":"[]","media":"[]","sort":113},{"id":115,"num":115,"title":"CORRECTION - THE TOKEN MAP HAD FOURTEEN HOLES AND MINT-ORDER SWEEPS ON REVIEW SETS WERE RUNNING WITH THEM","picked":"Fourteen Act 1 Review pieces (457-466, 497-500) had no entry in our GAN-to-mint map because their names use binary and hex numbering ('1000000', '0x21') the original matcher never handled. They sat mislabelled as 'gap' tokens. #457 was wrongly reported as never minted - it is mint 597.","tested":"All fourteen matched to their chain tokens by exact name and essence, map now complete for all 512.","happened":"FIXED. Every past mint-register negative on a Review-drop set that included any of these fourteen is suspect and should be re-run if such a set is ever revived. Sweeps from today onward use the repaired map.","status":"tested","links":"[]","media":"[]","sort":114},{"id":116,"num":116,"title":"THE LITERAL-RANDOM GAP - FIVE YEARS OF SWEEPS NEVER TRIED THE WORD ITSELF","picked":"The clue says 'ending with a random word'. Read literally, the twelfth word is the word 'random'. Seed derivation (PBKDF2) does not actually require a valid BIP-39 checksum, but every sweep ever run filtered to checksum-valid endings - so wherever 'random' failed checksum, it was silently never tried, on any set, ever.","tested":"All 81,662 historical sets in the master registry, each in alphabetical, reverse-alphabetical, mint, and reverse-mint order, twelfth word fixed to 'random', no checksum filter. 231,558 derivations, plant control recovered.","happened":"No match. An important negative: the whole project's history is now patched against this reading, in the orderings Pindar himself endorsed (both solved bitGAN puzzles say 'use the minting order').","status":"ruled-out","links":"[]","media":"[]","sort":115},{"id":117,"num":117,"title":"ONE GLITCH PER KIND - THE ELEVEN ARE ELEVEN DIFFERENT MECHANISMS","picked":"Our confirmed glitches refuse to be one mechanism (splice, negative frame, timing hold, duplicate file, double mint, label swap, corrupted essence, corrupted name, type typo, leet title, filename typo, case typo, unique speed - thirteen classes). Maybe 'looking all wrong' means expecting eleven of the same kind: the eleven could be one exemplar per mechanism. #191's corrupted essence err0r decodes to the valid word 'error'.","tested":"Primary eleven (one per class, all six drops) at full stateable depth, plus every eleven-of-thirteen class choice with exemplar alternates. 24,140 orderings, 3.1 million phrases, plant recovered. Mint orderings verified against the repaired token map.","happened":"No match.","status":"ruled-out","links":"[]","media":"[]","sort":116},{"id":118,"num":118,"title":"ONE VISIBLE GLITCH PER KIND - EVERYTHING SEEABLE ON THE OPENSEA LISTING","picked":"Michael's refinement: 'looking all wrong' = solvers dig in metadata, but every planted glitch is visible to a person on OpenSea - in the animation as it plays or in the official name as written. This re-sorts the classes: the timing holds become INVISIBLE (browsers clamp 0ms delays so the held frame renders normally), while chain-name corruptions become first-class ('Primoridal' #16/#100, 'Nebulous Glitch' #17, the wrong name on #44, the truncation on #97, the stray backtick on #66, the 0x radix break on #465/466 inside an octal-named drop).","tested":"Eleven visible classes, primary at full stateable depth plus all 3,840 exemplar combinations, 22,044 orderings, 2.8 million phrases, plant recovered.","happened":"No match. The twelve visible chain-name corruptions are newly catalogued evidence either way.","status":"ruled-out","links":"[]","media":"[]","sort":117},{"id":119,"num":119,"title":"PINDAR DEFINED OUR VOCABULARY HIMSELF - THE JOURNAL RE-READ","picked":"A full re-read of all 109 site pages against the newly confirmed mechanisms.","tested":"Close reading, no sweeps.","happened":"THE STRONGEST CONTEXT FIND OF THE PROJECT. (1) His testnet community called one-frame reveals 'the glitches' - 'one frame of each testnet bitGAN reveals mysterious figures'. (2) The article that names 'A Glitch' contains exactly two observations - 'it's not moving' and 'I see files being saved' - literally the 0ms-speed and re-save mechanisms confirmed at byte level this week. (3) The only time he decomposed a glitch for readers he decomposed it into frame delays. (4) On order, both solved puzzles say 'use the minting order' - no other ordering is supported anywhere. (5) Pre-clue he told holders to HODL exactly eleven Fibonacci pieces 'cause they will unlock special things'. Michael's testimony tempers (1): in years of viewing he has never seen figures - only fleeting single letters (o, e, s, t class) - so reveal-as-content is weak; reveal-as-pointer was tested in theory 120.","status":"tested","links":"[{\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":118},{"id":120,"num":120,"title":"THE REVEALED FRAME AS A POINTER","picked":"Every confirmed glitcher reveals one frame (splice frame, negative-run landing, held frame). Read the frame INDEX as a pointer: the word is the essence of the GAN at that index, or the wordlist entry at that index. The three solstice negatives share one exact signature (negatives at 37,43,47,50,51,52,53 - Fibonacci-backwards from the end) and #479 runs the mirrored run from the start, landing at 26 with frame 27 held - the hold marks where the run ends. Distinct pointer targets across all glitchers: only seven (16,19,24,26,27,30,37).","tested":"All eleven-piece subsets of the 13-14 visible glitchers, GAN and mint order both directions, three word readings (pointer essence, pointer wordlist, own essence), 1,284 orderings, 164k phrases, plant recovered.","happened":"No match. The shared negative-frame signature and the hold-marks-the-landing structure are recorded as real, unexplained design.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":119},{"id":121,"num":121,"title":"THE SILENT SAVES AS TRACE, THE BURN AS SEPARATE CHANNEL, THE MISSING RAINBOW","picked":"Three follow-ups run to ground. (a) H3-as-trace: the six no-change re-saves decoded - #203 alone at 23:24 closing spooky night; #407 at 02:49 on final-session day; #454/457/460/461 inside the final sitting itself. And the last two files Pindar ever touched (16:14, 16:17, Jan 30 2022) are #480 and #479 - the two timing glitches. The corpus's final act was planting the held frames. (b) The 16 burnt Flaming pieces overlap the re-save 77 at exactly one piece (#11) - the burn ritual and the edit sessions marked different pieces; independent channels. (c) The missing rainbow (#14's true art) appears in no held composite - the skullall review pieces are abstract GAN morphs, not grids; recovery would need external archives.","tested":"Timestamp decoding, set intersection, and visual inspection; no sweeps.","happened":"All three recorded. The final-session sitting is confirmed as THE deliberate puzzle session (already swept to full depth).","status":"tested","links":"[]","media":"[]","sort":120},{"id":122,"num":122,"title":"THE EXTRACTION PAIR'S FIVE - INCLUDING THE UNION-OF-CHANNELS ELEVEN","picked":"Two collaborating theorists (three critique rounds) produced five never-tested extraction methods. The primary: every timing/frame glitch MARKS skullGAN numbers (0ms robos mark their printed title numbers 26/47/51/53/61; the thrice-repeated negative trio marks its shared frames 37/43/47/50/51/52/53; held frames mark 27/24; marks outside 1-128 are void), selecting exactly eleven skulls {24,26,27,37,43,47,50,51,52,53,61}. The other four: the final-session eleven re-read, speed11 essences, the one-frame-reveal eleven, and pointer-indirection.","tested":"All five swept same-day: the union eleven at full stateable depth in ALL five word registers plus save/mint orders (essence, then wl0/wl1/mint/mint1 supplement), the other four at their stated orderings with 128 endings, plus literal 'random'/'final' endings checksum-free on everything. About 1.1 million phrases total, plants recovered throughout.","happened":"No match anywhere. The union eleven's construction (redundant double-marks, the void rule catching the decoy layer whose essences are rose/give/random/calm/final, the silently re-saved #47 marked) remains the most elegant dead theory in the log. Also killed by the pair's own censuses: corruption-as-edit-operation, sheet/chain delta-words, XMP-as-word-carrier, clue-page word counts.","status":"ruled-out","links":"[]","media":"[]","sort":121},{"id":123,"num":123,"title":"SPOTTER RESULTS - FOUR AXES CLOSED, TWO CANDIDATES KILLED, #47 RECLASSIFIED","picked":"Two collaborating visual spotters, constrained to 2022-solver evidence (playing animation + displayed name/traits + pre-clue website).","tested":"Detectors over all 512: frame-order anomalies, single-frame color swaps, letter census (glyph template matching, plant-calibrated), 11 sibling-comparison instruments (loop seams, motion profiles, season-vs-palette, Eyes audit, near-duplicates, morph direction).","happened":"Negatives, all clean: nothing plays out of order beyond known glitchers; no one-frame color swap exists; the letter census found only noise glyphs (Michael reviewed the 8-frame residue: not letters); no undiscovered duplicates. The soft 'inverted-flavoured' flags on #507 and #446 were KILLED by the strict negative-of-neighbors test (control #231 separates 143-vs-15; both candidates score normal). Standing corrections: #47 Twin Glitch skullGAN DOES glitch (hard-cut dialect the recurrence test misses) - the 'sole non-glitching prime' anomaly is retired; a 32-of-92 Solstice hue event at frames 33-37 is drop-wide authored behaviour; parts of the Review drop animate in reverse (#457/458/459/462/463/497 - thematically apt for a 'review', recorded not graded). Open candidates surviving: #280 (Solstice season-violator with corrupted frame), the Review reversal cluster.","status":"open","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}, {\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":122},{"id":124,"num":124,"title":"SPRITE-NAMING AND PAINTED WORDS - BOTH DEMOTED ON MICHAEL'S REVIEW","picked":"GAN-lore worker proposed: each glitch's reveal frame resolves toward a memorized 8-bit sprite; naming the sprite gives the word. Support cited: 'pulsating in and out of it is what looked like its training data', the 32x32 video-game training set, and the testnet 'one frame reveals mysterious figures, with tentacles'.","tested":"Michael's source review of both cited precedents.","happened":"DEMOTED. (1) The potato/elite/book/moon item is an unendorsed community observation on the explicitly 'not a clue' synopsis page of the SOLSTICE hunt - not authorial confirmation of painted words. (2) The testnet tentacle-figures resolve as reveal->identify-piece->numbers (testGANs 3/9/27, powers of three) - pointer semantics, not sprite naming; and Act 1's reveal frames contain no foreign art (montage full-search + swap-hunt) and no figures (Michael's eye). The 110-word BIP-39 8-bit inventory is banked in gan_lore_theories.md if naming ever revives.","status":"tested","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}, {\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":123},{"id":125,"num":125,"title":"THE SPOOKY LABORATORY - ONE GLITCH PER DROP, IN THAT DROP'S OWN MEDIUM","picked":"Michael's proposal: glitch solutions may differ drop by drop, with counts scaled to drop size - spooky (16 pieces) likely holds exactly one, making it the cleanest laboratory. Full spooky dossier compiled from every instrument.","tested":"All 16 spooky pieces crossed against every detector and forensic channel.","happened":"STRUCTURAL FINDING: spooky has ZERO animation-level glitches (no splice/negative/timing; uniform 55 frames at 100ms). Its glitch lives in names/metadata. Three candidates, one per evidence channel: #200 Count spookyGAN (Type=Glitch visible in traits; 'Count' puns on the drop's octal counting stunt; ess gospel), #205 'Flaming spookyGan' (lowercase-g case typo in the displayed name; ess grace), #203 Friendly Clown (the only re-saved spooky, silent and alone; hidden Rainbow; invisible to a 2022 solver; ess gown). The per-drop medium map: skulls=frame splices, robos=born leet-glitches, spooky=one name/metadata glitch, solstice=inversions+negatives, ghosts=the double-mint, review=timing holds. Per-drop prefix elevens (1,297 sets, spooky pinned to each candidate) swept in essence at GAN and mint order - see result column of theory 126 when recorded.","status":"tested","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":124},{"id":126,"num":126,"title":"THE PER-DROP ELEVENS SWEEP (follows theory 125)","picked":"1,297 elevens built from number-ordered prefixes of each drop's own glitch class (skull splice pieces, robo leet/swap pieces, one of the three spooky candidates, solstice negatives, ghost listing anomalies, review timing pieces), at least four drops represented in every set.","tested":"Essence register, GAN order and mint order, both directions - 3,436 orderings, about 440,000 phrases across both queues, plants recovered in each.","happened":"No match. The per-drop SELECTION principle (theory 125's medium map) remains live; its essence reading at basic orderings is now closed. Next depths if revived: other registers on the strongest per-drop composition, or a different rule for how many pieces each drop contributes.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":125},{"id":127,"num":127,"title":"THE SIGNATURE DICTIONARY - EVERY NAME'S VISUAL RULE, AND ITS VIOLATORS","picked":"Michael's question: has anyone mapped the visual signature behind every Type/SubType/GANg name? Built completely: 56 label values measured across all 512 (color, motion, loop shape, direction), each stated as one plain sentence with statistics. Discovery: some names are TEMPORAL - nagaGAN's palindrome runs through name/code/loop; Twin pieces animate against their drop's current; animation direction is a drop convention (ghosts emerge, solstice descends, Review splits into two currents, widening the old reversal cluster into drop-wide structure). New violators found: #468 (Flaming label, multicolor art - already the speed11 'random' piece), #281 (the only glitchGAN that never glitches - mirror of the #47 story), #395 (a Twin running backward, essence 'position'). Honest count: 7 strong violators, not eleven. Five ambiguous cases escalated to Michael's eye.","tested":"The strong seven plus every choice of four from the eleven mediums/alternates (including the three new ones and #390) - 900 orderings at GAN and mint order both directions, essence register, 115,200 phrases, plant recovered.","happened":"No match. The dictionary itself (signature_dictionary.md + signature_violations.json) stands as reference infrastructure; Michael's verdicts on the five escalations will finalize its ground truth.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle write-up\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":126},{"id":128,"num":128,"title":"THE FODDER-FED ROUND-2 HYPOTHESES (leet eleven, Fibonacci-as-number, re-save sessions)","picked":"Two hypothesizers, fed the curated 40-item fodder list, converged over two critique rounds. Joint #1: the LEET ELEVEN - the eleven pieces whose TEXT spells a fault word in leet (ten G1itch1ng/G1itchb0t titles {136,140,142,146,148,158,170,172,182,188} + #191 whose essence err0r is the ONLY non-BIP-39 essence in all 513, repaired to 'error'). Argument: 'looking for glitches all wrong' = the glitches are pieces that SAY glitch, not ones that glitch visually. Self-confirming: #146 (hollow) is also one of four held-delay timing files. Also: Fibonacci pieces read as wordlist[number] not essence (abandon ability able above abstract account action aerobic almost arch ball); the 11 XMP re-save sessions' last-touched essences.","tested":"Leet eleven at full stateable depth; Fibonacci number-as-word + secondaries + Named variant; re-save-session phrases both orderings. 1,391 orderings, 178,048 phrases, plant recovered.","happened":"No match. Leet eleven's honest weakness: all eleven are in the robo drop (fails breadth, though consistent with the born-glitched-robos finding). Fibonacci-as-number and re-save-sessions closed at tested depth.","status":"ruled-out","links":"[]","media":"[]","sort":127},{"id":129,"num":129,"title":"VERIFIED: REPEATED-WORD PHRASES WERE NEVER SILENTLY DEDUPED","picked":"Hypothesizer B flagged a possible coverage gap: sets holding both #14 and #75 say 'announce' twice, #390 says 'century' twice - if the harness deduped repeated words, that space would be wrongly marked covered.","tested":"Audited the registry (81,662 keys) and the derivation engine directly.","happened":"RESOLVED - no gap. The registry contains 2,847 keys with repeated words, and seed derivation treats duplicated words as distinct positions (two phrases differing only by one duplicated word give different addresses, verified). Past coverage is honest; repeated-word space was genuinely swept where those sets were tested.","status":"tested","links":"[]","media":"[]","sort":128},{"id":130,"num":130,"title":"THE ELEVEN RE-SAVE SESSIONS (\"a glitch is a save\")","picked":"GAN-lore reading: the 77 post-export Photoshop re-saves fall into eleven editing sessions; if a glitch = a save event, the eleven glitches are the eleven SESSIONS. Word per session = first-saved or last-saved piece's essence. NOTE: Pindar does NOT literally say 'a glitch is a save' - the article 'A Glitch' describes a save happening; the link is inference. And the count is threshold-dependent (13 sessions at a 1-hour gap vs 11 in the worker's model).","tested":"First-saved and last-saved essences of the 11 sessions, in session/GAN/mint order + reverses, 128 endings, plant recovered.","happened":"No match. Closed at natural orderings. Both caveats (inferred premise, soft count) stand.","status":"ruled-out","links":"[{\"label\": \"bitGANs journal\", \"url\": \"https://bitgans.com/journal/page016\"}]","media":"[]","sort":129},{"id":131,"num":131,"title":"THE MISLABEL ELEVEN (SubType-vs-art contradictions)","picked":"The calibrated visual census of pieces whose SubType label contradicts their art = exactly eleven: strong {179,423,231,511,509} + medium {204,150,490,478,92,491}. #511's essence is literally 'negative'. The most literal 'glitches' reading solvers overlooked: nobody compared label AGAINST art.","tested":"Essence register at full stateable depth, 2,856 orderings, 365,568 phrases, plant recovered.","happened":"No match at essence. The STRONGER extraction (read each piece's TRUE class word, not its stated label - Pindar's own 'inverted BIP39' precedent) needs Michael's namings first and is untested. The medium/weak boundary is instrument-set and wants a human ratification pass.","status":"ruled-out","links":"[]","media":"[]","sort":130},{"id":132,"num":132,"title":"DIMENSIONS AS ORDERING / DIMENSION-DISTINCT ELEVEN (Michael + synthesis)","picked":"Michael's idea: Dimensions trait (values 2-8,11; NO 1,9,10) orders the eleven like a Sudoku - 8 pinned positions + gaps at 1/9/10 deduced. Forces a cross-drop, mixed-glitch-type set (no single type spans the dimension range). #512 = only Dim-11 piece = pinned final. Rare dims pin to flagged pieces (Dim8={129,130,141,161}, Dim7={487,507}, Dim2=ground-truth glitchers {67,71}).","tested":"(a) dimension-ORDERED readings of existing candidate sets (qdim, ~5.8M seeds). (b) dimension-DISTINCT construction: 8 slots from confirmed glitchers + 3 wild gap-pieces, dimension-pinned order (qddim, 86,400 orderings, 11M seeds). Both plant-validated.","happened":"No match. The STRUCTURE is the strongest convergence found (drops fall out automatically, rare dims hit flagged pieces) but the word-reading is dead at essence. Caveat: loose slots (Dim4=37 options, Dim5=17) were filled with best-guess confirmed glitchers; a different filling is untested.","status":"ruled-out","links":"[]","media":"[]","sort":131},{"id":133,"num":133,"title":"MICHAEL VISUAL NOTES + COLOR-GRADIENT ELEVENS (ghost drop)","picked":"Michael hand-annotated all 512 (michael_visual_notes.json). Ghost drop confirmed as a rainbow gradient in GAN order (his obs). Color-backward-steppers swept: full-picture 11 {329,347,357,368,371,376,394,403,421,427,440} and figure-median 12; #357 (ess seed) is the robust core, flagged 4 independent ways.","tested":"Essence, all 4 number registers, GAN/mint/column/reverse, AND self-hue order (color-ordered), plant-validated. Figure-12 leave-one-out too.","happened":"No match at any register. Color SELECTION is a strong drop-native insight (his eye); word-EXTRACTION unresolved (essence dead, color-names aren't BIP39). Membership is metric-dependent (11 vs 12); needs Michael's eye to lock. Morai-echo 11 also swept negative (count soft per Michael).","status":"ruled-out","links":"[]","media":"[]","sort":132},{"id":134,"num":134,"title":"THE GHOST COLOR-GRADIENT ELEVEN(S) + #357 ANCHOR","picked":"Michael: ghost drop cycles through the rainbow by GAN number; a piece stepping BACKWARD in color is a glitch. Full-picture metric = 11 backward-steppers {329,347,357,368,371,376,394,403,421,427,440}; figure-median metric = 12 (different set). Robust core (both metrics) = {347,357,368,394,403,421}.","tested":"Essence + all 4 number registers + GAN/mint/column/reverse + SELF-HUE order (color-ordered) + figure-12 leave-one-out. All plant-validated.","happened":"No match at any register. Color SELECTION is real+drop-native but too noisy for a clean eleven, and color-names aren't BIP-39 so no word-reading. OUTCOME: #357 promoted to CONFIRMED ANCHOR (over-determined 6 ways: 2 eyes, 2 color metrics, montage no-match, essence=seed). Ghost medium=color (theme), #390 double-mint = its planted defect.","status":"ruled-out","links":"[]","media":"[]","sort":133},{"id":135,"num":135,"title":"FEELING-NAMES ELEVEN + MORAI-ECHO ELEVEN","picked":"Michael idea: pieces named for human emotions (4 Shy + awkward/anonymous/friendly/angry/hungry/gazing/staring = 11, 5 drops); and the Morai-echo class (ghost extra-heads echoing #1 the triplet Moirai template).","tested":"Feeling-11 essences full stateable depth; Morai-echo 11 full stateable depth. Plant-validated.","happened":"Both no match. Morai count is SOFT (Michael: faint echoes may be unstated). Feeling-words register unsweepable (most non-BIP39). #357 Shy skullGAN essence=seed noted.","status":"ruled-out","links":"[]","media":"[]","sort":134},{"id":136,"num":136,"title":"MICHAEL FULL 512 VISUAL NOTES INTEGRATED (29 glitch types)","picked":"Michael hand-annotated all 512 by eye (michael_visual_notes.json). Turned into 29-type catalog + full cross-reference. Ground truth for color-flash (130,133,462,463,464,497), hidden painted numbers (#165=8941, #207=#4), eye taxonomy, wrong-drop pieces, no-glitch negative filter (19 pieces).","tested":"Integration pass, no sweep. Materials updated: glitch_type_catalog.md, michael_notes_analysis.md, glitch_types.json, signature_dictionary.md.","happened":"Data mappings verified correct (essence + dimension row-indexed); all past sweeps valid. Lens finding: Michael's VISUAL axis and our STRUCTURAL/byte axis barely overlap — the eleven likely UNIONS both.","status":"tested","links":"[]","media":"[]","sort":135},{"id":137,"num":137,"title":"TYPE-RULE-BREAKER CLASS + NON-BIP-ESSENCE CLASS + latent-class method","picked":"Michael's creative method: one-offs that seem individual form hidden classes. FOUND: (a) type-rule-breakers (20, all 6 drops) - piece fails to be what its type declares (#281 glitchGAN-not-glitching, #14 wrong-art, #206 wrong-drop, etc.); (b) NON-BIP-39 ESSENCES = exactly 3: #191 err0r, #363 TRUE, #473 FALSE (computational/boolean); (c) PAIR pattern (day/night, beige, scanline, sine-flash, painted-numbers).","tested":"Class compilation, no sweep yet.","happened":"Recorded as Tier-1 classes + pools. Standing directive: cluster one-offs before treating as unique. type_rule_breakers.json saved.","status":"tested","links":"[]","media":"[]","sort":136},{"id":138,"num":138,"title":"META-ESSENCE LAYER (self-referential essences)","picked":"Essences that name the puzzle: seed(#357)/random(#468)/final(#512) = the clue's own words. 32 puzzle/computational meta-essences total; robo cluster EXPLAINED (its own solved puzzle's clue was a sentence containing the seed) so DISCOUNTED. Act-1-relevant layer = ghost(14)+review(5): chaos one crack seed code true pair name swap crash key sign letter cycle solve random number false final.","tested":"Read as instruction-sentence in GAN/mint order (identical: essences alphabetical-by-GAN pre-Review).","happened":"Not a clean sentence - suggestive vocabulary cloud. Held as on-theme sanity-check + possible scattered-instruction. PRECEDENT logged: robo puzzle solved by 'seed words are in the clue sentence' - Act 1 answer may be words in a sentence.","status":"tested","links":"[]","media":"[]","sort":137},{"id":139,"num":139,"title":"DIMENSION-DISTINCT SUDOKU - SET-1 (best-constructed eleven yet)","picked":"Michael's Sudoku: Dimensions {2,3,4,5,6,7,8,11, no 1/9/10} give partial word-ORDER; 8 pinned slots + 3 wild gaps(1,9,10). Anchored #357@Dim4, #512@Dim11-last. SET-1 = {14,67,141,207,231,357,479,480,507,511,512} - dimension-distinct, all 6 drops via own medium, 3 byte-confirmed glitches, Tier-1 built from Michael's real glitch-IDs.","tested":"Full stateable depth + Sudoku dimension-order + mint order + extraction variants (#14->announce true-class; #207->wordlist[4] hidden-number). ~392k phrases, plant recovered.","happened":"No match. SET-2/3/4 + G1-hedge queued (theory_iteration_2026-08-26.md). Best-constructed candidate to date; structure sound, words dead at essence.","status":"ruled-out","links":"[]","media":"[]","sort":138},{"id":140,"num":140,"title":"WRONG-CREATURE-FOR-DROP CLASS (compiled, not yet swept)","picked":"'Breaks its drop convention' = a piece whose GANg/species is rare in its drop (intruder). 20 pieces: skull{5whale,8blob,13spirit,89shark}, robo{130qubit,131/137/145cyber}, sol{209/300 Solar/Lunar skulls}, ghost{327ice,407robo}, review{465/466spooky,479,480,497,507,511,512xxxxxGAN}. #512 GANg=xxxxxGAN (redacted species!). Dims cover 2,4,5,6,7,8,11 (all but 3) - nearly a ready Sudoku pool.","tested":"Compiled; NOT yet swept (Michael thinking / permutation options open). #357 is NOT in this set (color axis, not species).","happened":"OPEN. 20>11: narrow by truly-singular species / one-per-drop / reconcile with #357. wrong_creature.json saved.","status":"open","links":"[]","media":"[]","sort":139},{"id":151,"num":151,"title":"Cracking the hidden-type code","picked":"Every piece in the first three drops carries a hidden on-chain trait called HiddenType: a single letter, a to e or A to E. Pindar's journal hinted it tracks number types. We asked what the letter actually means for every piece.","tested":"We compared each letter with the piece's position inside its own drop, using each drop's own way of counting: skulls count in plain decimal (1 to 128), robots in six-digit binary (the code at the end of every robot name), spookies in octal.","happened":"Fully decoded. The letter is the piece's number class: a = Fibonacci number, b = a prime that is not Fibonacci (the class that glitches), c = square number, d = other odd number, e = even number. The same code runs in all three drops, each on its native counting. Robots add an F for the eight binary palindromes. The later drops carry no letter, so the code stops after the spooky drop. Only four pieces break it, and they break it on purpose (theory 153). The decoded meaning is now on every piece as the 'Hidden-type meaning' tag.","status":"tested","links":"[]","media":"[]","sort":151},{"id":152,"num":152,"title":"The eleven pieces outside the hidden-type alphabet","picked":"Eleven pieces carry a HiddenType that is not a letter a to e: nine marked F (the eight palindrome robots plus spooky #193) and two whose trait is literally the text '32' (#32 and #161). A clean, on-chain, exactly-eleven set that nobody reading letters would notice.","tested":"Sixteen ways of turning a piece into a word (essence, piece number, mint number, binary position, name words) in mint, piece-number and binary orders. Every run carried a planted control seed, which was recovered.","happened":"No match. Once the code was decoded (theory 151), F turned out to be a legitimate palindrome class rather than a corruption, so this eleven was two different things glued together.","status":"ruled-out","links":"[]","media":"[]","sort":152},{"id":153,"num":153,"title":"The five hidden-type misfiles","picked":"With the code known, exactly five pieces wear the wrong letter: #32 and #161 carry the literal text '32' instead of a letter; #119 is odd but filed as even; #187 is even but filed as square; #193, the first spooky piece, carries F, a class that exists only in the robo drop (the collector's ruling, 13 Sep 2026), where its position 0 should read A. #32 and #119 are the two pieces Pindar flagged in his own journal as 'typo or important?'.","tested":"Checked all 512 against the decoded rule: five misfiles, no others. Recorded as a glitch class and swept as a small group with the other small classes (theories 155 and 159).","happened":"A real, authored five-piece class, not an eleven on its own. The two '32's exist because 32 can only be written as a position in the skull drop (decimal 32) and the robot drop (binary 100000); the spooky drop's octal range stops at 15. The count of two is forced by the number systems, not chosen at random.","status":"tested","links":"[]","media":"[{\"caption\": \"#32\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/32.gif\"}, {\"caption\": \"#119\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/119.gif\"}, {\"caption\": \"#161\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/161.gif\"}, {\"caption\": \"#187\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/187.gif\"}, {\"caption\": \"#193\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/193.gif\"}]","sort":153},{"id":154,"num":154,"title":"Scoring the whole collection by number class","picked":"If the hidden-type code is Pindar's habit, maybe it runs across all 512 as one sequence rather than drop by drop. We scored every piece by its overall number (0 to 512) and by its mint number using the same five classes.","tested":"Compared every ratified glitch class against the global scores, and profiled the fourteen pieces sitting at global Fibonacci positions {0,1,2,3,5,8,13,21,34,55,89,144,233,377}.","happened":"The code is drop-native, not global. Only 13 of the 73 ratified glitch pieces are globally prime-not-Fibonacci, and no glitch class maps onto one global class. The global Fibonacci set is Pindar's Named eleven plus three later pieces (#144 burnt, #233 label contradiction, #377 space invader): an archetype axis, not a glitch axis. The scores stay on every piece as the 'Number class' tags.","status":"ruled-out","links":"[]","media":"[]","sort":154},{"id":155,"num":155,"title":"Bookends: adding Pindar's anchor pieces to nine- and ten-piece sets","picked":"Several strong glitch sets have 9 or 10 members, not 11. Pindar's anchor pieces, #0 'universe' (before the first), #1 'abandon' (the first seed word) and #512 'final' (the last), could fill the missing slots at either end.","tested":"Three tiers: 64 hand-picked configurations; every 9- and 10-piece subset of the twin, entangled and prime-robot classes with anchors at each end (27,742 sets, 3.5 million seeds); every one-per-pair combination and each drop's native set trimmed to 9 or 10 (1,270 sets). Essence words, several orders, planted control recovered each time.","happened":"No match anywhere. Anchoring does not rescue any ratified glitch set.","status":"ruled-out","links":"[]","media":"[]","sort":155},{"id":156,"num":156,"title":"A passphrase as a hidden thirteenth ingredient","picked":"A seed phrase can carry an optional extra password. Five years of sweeps had never tried one. 'Universe' (piece #0, 'before the first') and 'infinite' were the natural candidates.","tested":"144 of the strongest dead sets, two orders each, with the passphrase attached, plus 'universe' as a literal twelfth word. 74,720 wallet derivations, planted control recovered.","happened":"No match. A passphrase does not revive the strong dead sets.","status":"ruled-out","links":"[]","media":"[]","sort":156},{"id":157,"num":157,"title":"Ordering the backwards-Fibonacci eleven by where the glitch first appears","picked":"The eleven robots whose animation re-inserts a frame in a Fibonacci pattern. Michael asked whether the frame where the first re-insertion lands could set the word order.","tested":"Measured the onset frame for all eleven (seven share frame 37; the rest range from 19 to 29). Swept onset-then-mint, drop-then-onset and reversed variants with essence, piece-number and binary-position words.","happened":"No match. Onset location is partly distinctive but does not order the words.","status":"ruled-out","links":"[]","media":"[]","sort":157},{"id":158,"num":158,"title":"Act 1 Review pieces as remakes or mirror images of earlier pieces","picked":"Michael's idea: the 68 Review pieces are not just tributes but transformations of specific earlier pieces. If so, each Review piece points at an earlier one.","tested":"Every frame of all 68 Review pieces compared with all 22,753 earlier frames, as-is and flipped left to right, with controls for the generic symmetry of pixel art.","happened":"No authored map. One strong match exists, #458 'soup' echoing #150 'fire', and Michael judged it a coincidental family resemblance. Flipping never helped. Review pieces do not systematically remake earlier ones.","status":"ruled-out","links":"[]","media":"[{\"caption\": \"#150\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/150.gif\"}, {\"caption\": \"#458\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/458.gif\"}]","sort":158},{"id":159,"num":159,"title":"32 as the building block","picked":"Every bitGAN is 32 by 32 pixels, the drop sizes are powers of two, and two pieces carry a literal '32'. A seven-lens brainstorm produced 24 distinct ways 32 could pick pieces or turn them into words.","tested":"All 24 resolved: plus-or-minus-32 hops (killed by a control: shifting by 16 lands on glitches more often), the doubling eleven {0,1,2,4,...,512}, 32-by-32 pixel-grid reads, base-32 and octal readings, 32-row bit strings, distance from 32, the bitGANg 1001 roster, and the #32/#95 burn pair. Planted-control sweeps wherever a word set existed.","happened":"Every word reading was dead. Two facts survive: #32 and #95 were burnt in the same transaction (a real pair, now the 'Co-burned' tag), and 32 is a genuine boundary across the number systems, which explains why exactly two '32's exist. 32 is structure, not a selector.","status":"ruled-out","links":"[]","media":"[{\"caption\": \"#32\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/32.gif\"}, {\"caption\": \"#95\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/95.gif\"}]","sort":159},{"id":160,"num":160,"title":"The untested methods, swarmed","picked":"This site lists 47 word-registers (ways to turn a piece into a word) and 33 orderings, many never run. A four-family swarm matched every untested one against every glitch set and produced 28 recipes.","tested":"Each 'most promising' recipe was checked against the registry before running; the top pick turned out to be an already-dead register in disguise. The tribute eleven, the silver eleven and the twin-inverter cut (#157 and #298, the only two twins that put 'Twins' in Type instead of SubType) were swept with planted controls.","happened":"No match. Most untested registers collapse into dead ones, fall short of eleven, or need a human judgement call. That last group led to Michael's no-guess rule (theory 162). The Extraction and Ordering tabs now show what has and has not been tested for every method.","status":"ruled-out","links":"[]","media":"[]","sort":160},{"id":161,"num":161,"title":"Filename words that appear nowhere in the metadata","picked":"85 Arweave filenames carry a hand-written word that is not in the piece's traits: mostly seasons on the ghost drop, plus 'ancient', 'whale', 'spirit' and 'elder' on Named skulls. An objective register nobody had read.","tested":"Extracted every filename word; swept the skull-descriptor set and a creature-hybrid set (the creature word where it is a seed word, otherwise the essence).","happened":"No match. The distinctive words are the Named pieces' true identities (whale, spirit, heart, warrior, ancient), but only five or six of the eleven Named pieces have one, so the register cannot fill eleven. Kept as a partial register and as the 'Filename word' tags.","status":"ruled-out","links":"[]","media":"[]","sort":161},{"id":162,"num":162,"title":"Michael's no-guess rule: the word must be visibly there","picked":"A ruling rather than a theory. A seed word must be objectively present on the piece: a metadata field, a mark painted into the art, or a creature anyone can see. It must never be inferred from outside knowledge. Pindar's three solved puzzles were all mechanical in exactly this way.","tested":"Applied to every untested register on the site.","happened":"It closed the artist-tribute, homage-target and lore-vocabulary registers for good, because each needs a guess. It left a narrow objective frontier: four painted marks and the visible-creature words (theories 163 to 165).","status":"tested","links":"[]","media":"[]","sort":162},{"id":163,"num":163,"title":"The visible-creature eleven","picked":"Read what each piece visibly IS. A census of all 512 found exactly eleven rare creatures whose names are seed words: whale, spirit, heart, warrior, cat, dog, wolf, clown and three more. Objective, visible, not the essence, and exactly eleven.","tested":"17 orders including mint. Then Michael reviewed the animations and corrected three (#264 is a dog but a tribute piece, #268 is a bird, #412 is a monster); we swept his corrected reading, then every creature-versus-essence combination (128 of them) in seven orders. About 530 combinations, planted control recovered.","happened":"No match, including mint order, which solved both earlier Pindar puzzles. The clean eleven was partly a census artifact; the genuine rare creatures number about eight. The creature register does not yield an eleven. The verified creature identities stay on the site as the 'Visible creature' tags.","status":"ruled-out","links":"[]","media":"[{\"caption\": \"#264\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/264.gif\"}, {\"caption\": \"#268\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/268.gif\"}, {\"caption\": \"#412\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/412.gif\"}]","sort":163},{"id":164,"num":164,"title":"#300 as the one file that holds the whole phrase","picked":"The solved byteGAN puzzle hid its key inside one file. #300 'Lunar skullGAN' was thought to be the only Photoshop-saved file in the set, so it might carry all twelve words.","tested":"Full byte, XMP and Photoshop-block extraction of #300, and a check of the premise across all 512.","happened":"No payload: no layers, no history, no word list. The premise was false too: all 512 carry a Photoshop creator tag. #300's only uniqueness is a save-path fingerprint.","status":"ruled-out","links":"[]","media":"[]","sort":164},{"id":165,"num":165,"title":"Painted marks as pointers","picked":"Four pieces have something painted into the art: #165 '81', #207 '4', #179 '10.00', #379 'AI'. Read each as a wordlist position or a piece number.","tested":"81 gives 'any' or piece #81; 4 gives 'above' or #4; 10 gives 'access' or #10; 'AI' is letters, not a number.","happened":"Too small: at most four objective items, one of them non-numeric, and no objective seven-word register to pair them with. Closed as a component.","status":"ruled-out","links":"[]","media":"[{\"caption\": \"#165\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/165.gif\"}, {\"caption\": \"#207\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/207.gif\"}, {\"caption\": \"#179\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/179.gif\"}, {\"caption\": \"#379\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/379.gif\"}]","sort":165},{"id":166,"num":166,"title":"Pindar confirms it: the intentional errors are the glitches","picked":"Michael found a Discord message from 1 January 2022, before the clue existed. A collector noticed a Rainbow skull (#58) missing its SubType. Pindar replied: 'There are some glitch bitGANs with intentional errors - you will find others - they are glitched.' Exactly eleven pieces carry an on-chain trait that was deliberately corrupted: #16, 17, 44, 48, 58, 62, 66, 81, 173, 281 and 309.","tested":"Each corruption was classified by its move: transpose ('Primoridal'), reverse ('wobniaR'), delete (a missing SubType or a missing name suffix), swap (Type and SubType exchanged), substitute (Rainbow became Flaming, Ghost became Smoke). The three Rainbows, glitched three different ways, read as a Rosetta stone for the grammar.","happened":"The only exact-eleven an artist statement directly confirms. Its words are still hidden: the essences, the trait values themselves, the verbs as number operations (theory 167) and the undo-pointers (theory 168) are all dead. Finding this eleven's word channel is the open problem. UPDATE (13 Sep 2026): the 'exactly eleven' was counted on an incomplete chain pull. Recomputed field by field on the complete pull, the sheet and chain disagree on 26 pieces (theory 174). The class stands; the count of eleven does not. STATUS (14 Sep 2026): the class is settled and the artist-confirmed selection is now theory 179's elevens A and B; this card is the history of the discovery, not an open question.","status":"tested","links":"[]","media":"[]","sort":166},{"id":167,"num":167,"title":"The glitch verbs as instructions on numbers","picked":"If each corruption is a verb (reverse, delete, swap, substitute, transpose), maybe it says what to do to a number the piece already carries: its wordlist position, piece number or mint number.","tested":"Every reading was enumerated (about 90 candidate words across eleven slots). A real cipher uses one operand consistently, so only logics that fill all eleven slots were swept: piece number and mint number, counted from 0 and from 1. The five 'substitute' pieces have no numeric meaning on a wordlist position, so those logics cannot fill eleven. Planted controls recovered, including a logic missed on the first pass.","happened":"All four full logics dead. The verbs are not arithmetic on the piece numbers. The surviving reading is that they describe how each true word was hidden, acting on something not yet identified.","status":"ruled-out","links":"[]","media":"[]","sort":167},{"id":168,"num":168,"title":"Undo the corruption and follow it as a pointer","picked":"Repair each of the eleven errors to its true value (Rainbow, Ghost, Ancient, Primordial and so on) and treat that value as a pointer to other pieces.","tested":"The true values name sets, not pieces (Rainbow is 82 pieces, Ghost is 105). Nine collapse rules were swept: set counts as wordlist positions, the first, last or Nth member's essence, and hybrids.","happened":"No match. The words are not in this eleven's metadata, its numbers, or the sets its true values name.","status":"ruled-out","links":"[]","media":"[]","sort":168},{"id":169,"num":169,"title":"Glitches teach the method: right index, wrong list","picked":"A six-lens swarm on the idea that each glitch demonstrates a method to be applied somewhere else. Every lens converged on the same lesson: the glitches teach indirection, a correct index aimed at the wrong list or read in the wrong direction.","tested":"Two recipes were computable: transplant the confirmed eleven's mint ranks onto the title-alphabetical list and read those pieces' essences (goat fox bullet cross air grab goddess blind eight divide bean); and read the backwards-Fibonacci binary values from the END of the wordlist (work wonder wisdom winter win width whisper wheel welcome weapon swing). Both were independently recomputed, then swept with reverses.","happened":"Both dead. The swarm's by-products were real, though: a deck-wide typo census landed on exactly eleven deliberate pairs (theory 170), and the idea that structural glitches encode word ORDER rather than words (theory 171) is still open.","status":"ruled-out","links":"[]","media":"[]","sort":169},{"id":170,"num":170,"title":"Exactly eleven deliberate pairs","picked":"Two pieces that carry the same one-off defect that nothing else in the 512 carries: an identical file, a shared misspelling, one shared flash, day and night, TRUE and FALSE. Michael ratified nine pairs; a census found two more ({9,73} with 'gltich' in both filenames; {170,175} with 'Raibow' in both SubTypes). Exactly eleven pairs, 21 pieces, with #175 in two of them.","tested":"Every one-member-per-pair eleven (1,536 distinct sets), essence words, mint order, planted control recovered.","happened":"No match on words. The pairing itself is a locked fact and now a glitch class on this site: select the 'Deliberate pair' tag in the Collection to see each pair side by side with the reason.","status":"ruled-out","links":"[]","media":"[{\"caption\": \"#9\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/9.gif\"}, {\"caption\": \"#73\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/73.gif\"}, {\"caption\": \"#170\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/170.gif\"}, {\"caption\": \"#175\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/175.gif\"}]","sort":170},{"id":171,"num":171,"title":"Structural glitches encode the word ORDER","picked":"The backwards-Fibonacci gaps 5,3,2,1,1 sum to 12 and walk a 12-slot cycle that lands on slots {5,8,10,11,12}; mirrored, they give the Fibonacci-protected slots {1,2,3,5,8}. Two held-frame pieces land on two of the leftover slots. Perhaps the glitches place the words rather than name them.","tested":"Not yet run. It is an ordering, not a word source, so it needs a word set to order.","happened":"Open. If a word register for the confirmed eleven (theory 166) appears, this is the first ordering to try instead of plain mint order. UPDATE: with elevens A and B now fixed (theory 179), this ordering can be tested the moment a word register for them exists; mint order is the default until then.","status":"open","links":"[]","media":"[]","sort":171},{"id":172,"num":172,"title":"Rainbow is Pindar's demonstration word","picked":"Rainbow is the most-glitched value in the collection: reversed (#48), deleted (#58), substituted (#66), and missing a letter on two robots (#170 and #175 'Raibow'). In the robot drop it becomes 'Rainbot', the only robot name stem that is cut short; every other stem (Firebot, Ghostbot, Smokebot) keeps the whole trait word.","tested":"Compared every Rainbow, Rainbot and Raibow across the drops; checked the robot naming grammar and the robot essences, which Pindar chose by hand rather than alphabetically.","happened":"'Rainbow' is not a seed word; 'rain' is, and the first Rainbot (#134) carries the essence 'rain'. Rainbow reads as the word Pindar mutates to show the moves, and to show that an off-list word can point at an on-list one. An observation about method, not a word source.","status":"tested","links":"[]","media":"[{\"caption\": \"#134\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/134.gif\"}, {\"caption\": \"#175\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/175.gif\"}]","sort":172},{"id":173,"num":173,"title":"The ten seed-list colour words","picked":"The BIP-39 list holds exactly ten plain colour words (black, blue, brown, gold, green, olive, orange, pink, silver, yellow) and every one is an essence exactly once. Seven colour-ish essences (color, bright, pumpkin, avocado, coffee, rose, cinnamon) also appear once each. Ten colours plus 'color' is a natural eleven.","tested":"The ten colours with each colour-ish eleventh word first, last or sorted in; every eleven from the seventeen colour-ish words; piece, mint, alphabetical, spectrum and drop orders forwards and backwards. 94,106 orderings, 12 million seeds, planted control recovered.","happened":"No match. The completeness of the set says something about how Pindar assigned essences (five sit in the robot drop, where he chose them by hand), but it is not a word source. The ten pieces carry the 'BIP colour word' tag.","status":"ruled-out","links":"[]","media":"[]","sort":173},{"id":174,"num":174,"title":"The corrupt chain is bigger than eleven","picked":"We recomputed every field of Pindar's Arweave sheet against the raw on-chain traits for all 512 pieces (fourteen Review pieces with binary or hex counters had been mislabelled as missing in our pull; recovered 13 Sep 2026, they match the sheet exactly). Any piece where a name or trait value differs, in either direction, counts.","tested":"Each difference was classified by the kind of move: two letters swapped, written backwards, something missing on one side, Type and SubType exchanged, a different real word, a capital letter, a different number. The rarer side was named from the rest of the collection; where both values are real words it is marked as a judgement.","happened":"26 pieces, not 11. New chain-side oddities of the same kinds as the eleven: #4 'SkullGAN', #88 'skulGAN', #97's suffix cut to 'Slender Glitch', #100 'Primoridal' (the same move as #16), #124 'skullGANs', a stray backtick on #66, GANg cyclopsGAN on #39 and roboGAN on #509, three numeric differences, and a systematic Twin-to-Twins rename on all five sheet-'Twin' pieces (excluded as a convention). The selection is now 'Sheet-vs-chain corruptions (26)'. Every piece's page shows both values. No word reading has been run on the 26 yet. Resolved into the witness piles of theory 177 and the elevens of theory 179.","status":"tested","links":"[]","media":"[{\"caption\": \"#4\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/4.gif\"}, {\"caption\": \"#88\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/88.gif\"}, {\"caption\": \"#97\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/97.gif\"}, {\"caption\": \"#100\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/100.gif\"}]","sort":174},{"id":175,"num":175,"title":"Everything confirmed against Arweave itself","picked":"We went to Pindar's own Arweave drive rather than trusting our copies: enumerated every file entity, downloaded the properties sheet, every loose GIF and all six drop archives (about 3.4 GB), and hashed everything.","tested":"MD5 of every archived file against our 512 GIFs and our copy of the sheet.","happened":"All 512 GIFs and the sheet are byte-identical to ours. The only Arweave GIFs with no local match are two studies of piece #0, outside the 512. The pair #14/#75 is one byte sequence stored under two filenames. The archives also supplied Pindar's own filenames for all 512, replacing the folder names we had been using. The fourteen Review pieces we thought had no chain record were in our chain pull all along, mislabelled because their names carry binary or hex counters; they match the sheet exactly.","status":"tested","links":"[]","media":"[]","sort":175},{"id":176,"num":176,"title":"What changed on 31 January 2022","picked":"Michael's idea: if the sheet is the file Pindar modified to make the puzzle, its differences from the chain are where to look. We reconstructed the day from ArDrive and Arweave block timestamps, the zip archives' internal modification times, and every token's transfer history.","tested":"Compared each archived GIF's disk time with its Photoshop save time; checked ArDrive for earlier revisions of the sheet; placed every piece's first transfer out of Pindar's wallet relative to the sheet and the clue.","happened":"31 January was the publication of the complete Act 1 archive: sheet saved 19:42 UTC, uploaded 19:51 with the ghost and review archives, journal 015 updated with every link and archived on Arweave at 20:12. No GIF in those archives was edited after export, so the sheet is the artifact of the day, and no earlier revision exists to diff against. Every skull-drop disagreement's chain record predates the sheet by five months. The ghost drop first left Pindar's wallet on 1 to 3 February; the prize piece #445 moved into the mystery wallet on 5 February at 11:12 UTC, the day the clue was posted; most of the Review drop, including all fourteen binary- and hex-numbered pieces and #512, first left his wallet on 24 to 28 August 2022, around his 'looking for glitches all wrong' tweet.","status":"tested","links":"[]","media":"[{\"caption\": \"#445\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/445.gif\"}]","sort":176},{"id":177,"num":177,"title":"Which side is wrong: three witnesses","picked":"For every sheet-versus-chain disagreement we asked three witnesses which side holds the odd value: whether the value is used anywhere else in the collection, the Arweave filename Pindar wrote at export (which sits in time between the chain record and the sheet), and his journal (page 003 calls #39 'the cyclopsGAN').","tested":"Applied to all 26 disagreements; Michael then ruled on the substitutions by eye.","happened":"Four piles. The sheet corrects a chain error on thirteen pieces (letter-level moves: a capital, two transpositions, a reversal, a displacement, a swap, a stray backtick, three cut suffixes, a dropped and an added letter). The substituted word describes what the art shows on seven (Michael's ruling: Rainbow/Flaming, Ghost/Glitch, Ghost/Smoke, Ancient/Primordial, glitchGAN/bitGAN, skullGAN/cyclopsGAN, skullGAN/roboGAN). The sheet introduces the error on three (Cronenber, RainbowShimmer, Raibow). Four number or capital differences are undecidable. Two things reversed the old reading: #17's filename says 'glitch', agreeing with the chain, and #73's 'Gltich' is in Pindar's sheet and filename, so the chain's correct spelling is the odd record there.","status":"tested","links":"[]","media":"[{\"caption\": \"#17\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/17.gif\"}, {\"caption\": \"#73\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/73.gif\"}]","sort":177},{"id":178,"num":178,"title":"The substitutions echo what you can see","picked":"Michael's observation on the seven substitutions: each swapped word names a visual element of the piece that the data cannot see. Flaming on a piece that visibly flames, Smoke on a smoky one, cyclops on a one-eyed skull, roboGAN on a robot-shaped skull.","tested":"Recorded as the collector's ruling on each of the seven and as a new visual category, 'Name describes a visual element', for him to adjudicate across all 512.","happened":"Not errors on either side: the label bending toward the picture. It says the picture is the authority over the metadata, which fits Pindar's line about looking for glitches all wrong. Five of the seven visible truths are seed words or cut to one (robot, smoke, flame, ancient, and rain via Rainbow); glitch and cyclops are not. Open: the same reading applied to the eleven chain-error pieces, by eye.","status":"open","links":"[]","media":"[{\"caption\": \"#66\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/66.gif\"}, {\"caption\": \"#173\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/173.gif\"}, {\"caption\": \"#509\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/509.gif\"}]","sort":178},{"id":179,"num":179,"title":"Two candidate elevens, and the word registers closed on them","picked":"Read strictly, 'every malformed chain label the sheet corrects' gives thirteen pieces. Two of them are arguable under the rule itself: #73 (the chain is spelled correctly) and #66 (its only remaining chain oddity is a stray backtick). Removing them gives eleven B; keeping #66 and taking only the skull drop gives eleven A. Both are rules Pindar could state in a sentence; order is mint, which for skulls equals piece order.","tested":"Under the standing rule that nothing runs without a stated selection, word rule and order: the essence (every eleven inside the thirteen and inside the thirteen plus the sheet-error pieces, 151,164 orderings); then 'undo the error and cut the restored word to the list, repeats allowed'; then 'the filename word as the unpublished column'. Three further readings could not fill eleven slots under their own rules and were not run: the hidden-type class names (only 'square' is a seed word), applying each piece's own move to its word (lands on the list for none of the eleven), and reading the sheet-introduced spellings corrupted (only one cuts to a seed word). A stacked mix-and-match read (four substitution words plus seven letter-error words) was started and halted by Michael as brute force without a theory.","happened":"No hit on any register run. The selection and order stand as the best-motivated in the project; its word channel is still unknown. Pindar's moves are not rewarded by the wordlist on these words (reversal carries five wordlist words to others, an adjacent swap six, a dropped letter 198, none of them these). The reading left is Michael's: the word is what the corrected label describes visually.","status":"open","links":"[]","media":"[{\"caption\": \"#4\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/4.gif\"}, {\"caption\": \"#16\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/16.gif\"}, {\"caption\": \"#48\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/48.gif\"}, {\"caption\": \"#58\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/58.gif\"}]","sort":179},{"id":180,"num":180,"title":"Pindar's filenames are his private construction code","picked":"The archive filenames, now read from Arweave for all 512 pieces.","tested":"Compared every filename token against the piece's traits, number class and drop.","happened":"The 's_' prefix marks exactly the square-numbered skulls that take letter c (#1 is Fibonacci first; #121, eleven squared, is the one unmarked square). 'smooth' appears only on #5, #13, #89, the Fibonacci primes the glitch rule protects. Primordial is written 'ancient' eight times and 'elder' once, on #16, one of the two Primoridal pieces; both are seed words and Primordial is not. Solstice filenames carry '1221' and a season code that tracks the Date trait the solved puzzle used. Review filenames group the drop into reprise00 to reprise13 with no reprise07, plus repriseaa, reprisebs, highdef, line, test, skullall and the tributes; the fourteen binary/hex-numbered pieces are exactly the reprises of the robo, crypto and spooky pieces. The layer encodes the logic he later published, and it is the only layer that has turned an off-list trait word into on-list ones.","status":"tested","links":"[]","media":"[{\"caption\": \"#16\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/16.gif\"}, {\"caption\": \"#100\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/100.gif\"}, {\"caption\": \"#121\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/121.gif\"}, {\"caption\": \"#5\", \"url\": \"https://api.bitganspuzzlebook.com/media/gifs/5.gif\"}]","sort":180},{"id":181,"num":181,"title":"The moves as instructions, and what cipher history says","picked":"Each intentional error is a move: transpose, reverse, delete, insert, swap fields, displace, change a capital, substitute. Read in both directions (undo it, or apply it) each is an instruction.","tested":"Checked which moves carry a wordlist word to another wordlist word; read the history of deliberate errors in ciphers.","happened":"Reversal works five times (edit/tide, evil/live, liar/rail, loop/pool, net/ten), an adjacent swap six times, a dropped letter 198 times. Pindar has already demonstrated the reversal move in plain sight: the naga names are reversal pairs and Evil Olive gives two seed words. History rhymes: the Kryptos sculptor planted three deliberate misspellings that are now argued to encode the key to its unsolved passage; null ciphers use deliberate errors to mark which words count; Bacon's cipher hides a binary code in capitals and typefaces; book ciphers index words in a public text, and the BIP-39 list is the obvious public text here. Four statable hypotheses came out of this (theory 179 has their results).","status":"tested","links":"[]","media":"[]","sort":181},{"id":141,"num":141,"title":"The backwards-Fibonacci frame glitch","picked":"Eleven pieces where one animation frame comes back again and again, with the gaps between its returns shrinking 5, 3, 2, 1, 1 — the Fibonacci sequence run backwards. We found these by rebuilding the glitch detector from scratch and matching it to Michael's eye; exactly eleven pieces carry the complete run: #148, 152, 158, 160, 166, 170, 172, 176, 182, 190, 287.","tested":"We turned the eleven into full seed phrases using each piece's essence word, its piece-number and mint-number as word-list positions, its name's BIP-39 word, and its binary position — across seven orderings (mint, reverse mint, piece number, binary position, glitch timing, and alphabetical). Every phrase was checked against the wallet with a planted control to prove the checker works.","happened":"Ruled out — all straight readings of this eleven are dead. It stays the strongest structural candidate, because it is an exact eleven with no trimming, so its non-obvious readings (the glitch's own numbers, its gap sequence) are still worth trying.","status":"ruled-out","links":"[]","media":"[]","sort":281},{"id":142,"num":142,"title":"Chain-is-corrupt-side eleven","picked":"The exactly eleven pieces where the blockchain copy of a trait is the corrupted one while the official spreadsheet has it right: #16, 17, 44, 48, 58, 62, 66, 81, 173, 281, 309.","tested":"Built into seed phrases using essence, piece-number, mint-number, and name words across six orderings; checked against the wallet, plant-validated.","happened":"Ruled out on every straight reading. SUPERSEDED (Sep 2026): the 'exactly eleven' was counted on an incomplete chain pull; the complete diff gives 26 disagreements, split by three witnesses into four piles (theory 177), with two candidate elevens A and B (theory 179).","status":"ruled-out","links":"[]","media":"[]","sort":282},{"id":143,"num":143,"title":"Every group of eleven inside the fifteen 'Entangled' pieces","picked":"The fifteen pieces Pindar named 'Entangled'. Since the answer needs eleven, we tried every possible group of eleven out of the fifteen — all 1,365 of them.","tested":"Each group's essences in mint order, checked against the wallet.","happened":"Ruled out. The loaded essences in this class — key, solution, book, believe, false, final — made it tempting, but no group of eleven matched.","status":"ruled-out","links":"[]","media":"[]","sort":283},{"id":144,"num":144,"title":"Every group of eleven inside the fifteen glitching robots","picked":"The fifteen robots that glitch (all at prime binary positions — see the discovery below). We tried every group of eleven out of the fifteen.","tested":"Essences in mint order, and binary positions read as word-list positions, checked against the wallet.","happened":"Ruled out as a solve — but this exact work produced a real discovery about how the robots glitch.","status":"ruled-out","links":"[]","media":"[]","sort":284},{"id":145,"num":145,"title":"DISCOVERY: the Prime Rule extends to the robot drop","picked":"Pindar wrote a rule in his own journal 002: in the first (skull) drop, a piece glitches exactly when its number is a prime that is not also a Fibonacci number. Everyone thought the rule stopped after the skulls. We checked the robots against the binary number spelled inside each robot's name, instead of its piece number.","tested":"For all 64 robots we compared 'does it glitch' against 'is its binary position a prime that isn't Fibonacci'.","happened":"CONFIRMED, with zero exceptions: every glitching robot sits at a prime-not-Fibonacci binary position — 15 of 15 — and the only prime positions that stay clean are 2, 3 and 5, the Fibonacci primes. It is the same rule as the skulls, keyed to each drop's own numbering. The old master spreadsheet missed it by testing the wrong number. This doesn't open the wallet by itself, but it turns the glitch selection into a provable rule instead of a guess, and ties the whole puzzle back to Fibonacci. UPDATE: theory 151 shows why the rule holds without exception. The number class is written into each piece's hidden trait letter, and 'b' is exactly prime-not-Fibonacci.","status":"near-miss","links":"[]","media":"[]","sort":285},{"id":146,"num":146,"title":"Numbers painted into the art (the byteGAN move)","picked":"In Pindar's earlier byteGAN puzzle the key word was painted right into the picture. So we scanned all 512 animations, every single frame at full size, hunting for numbers or letters the artist painted in by hand.","tested":"Six helpers scanned the whole collection; Michael then graded every candidate by eye on all-frames grids. Blind controls — pieces we knew carried marks but hid from the hunters — were all recovered, which proves the scan works.","happened":"Confirmed four carriers: #165 shows '81', #179 shows '10.00', #207 shows '4', and #379 shows 'AI'. This is a finished, tested census. But four painted marks is not eleven, so it is one piece of the answer, not the whole thing. The numbers are parked as possible word-list positions or pointers to other pieces.","status":"tested","links":"[]","media":"[]","sort":286},{"id":147,"num":147,"title":"Eyes that never move","picked":"Michael noticed while browsing that on some pieces the eyes stay perfectly still while everything else animates.","tested":"He checked candidates by eye; we censused the whole collection for the effect.","happened":"Confirmed: eight pieces have static eyes. Because two of the eight palindrome robots do NOT have it, this is a deliberately chosen class of eight, not automatic behavior — which makes its two odd members stand out: the qubitGAN #130 (the piece lore calls the wallet-cracker) and #161. Not yet turned into a solve, but a clean authored group worth using.","status":"near-miss","links":"[]","media":"[]","sort":287},{"id":148,"num":148,"title":"Maybe 'eleven glitches' means eleven work sessions, not eleven pieces","picked":"The hidden file metadata shows Pindar re-saved 77 of his files, and those re-saves cluster into exactly eleven separate Photoshop sessions.","tested":"Not yet run. Each session has a date and a first-saved piece — either could give a word, the way the solved Solstice puzzle turned dates into words.","happened":"Ruled out. We ran both halves: the piece-based reads (each session's first-saved and last-saved piece, through essence / piece-number / mint-number / name across four orderings) and the date-based reads (each session's date turned into a number — day-of-year, month-day, month, days-since-first — read as a word-list position). All dead, plant-validated. The elegant idea that 'eleven glitches' meant the eleven work sessions does not hold on any straightforward reading.","status":"ruled-out","links":"[{\"label\": \"Solstice puzzle used dates as words\", \"url\": \"https://bitgans.com/news/pattern-revealed\"}]","media":"[]","sort":288},{"id":149,"num":149,"title":"Let the glitch's own number be the word","picked":"The glitch pieces are defined by numbers — their binary position, their frame count, the exact frames where a repeat lands, how many times it repeats. Instead of a piece's essence, use one of those numbers as a word-list position.","tested":"We tested the simplest version — binary position read as a word-list position on the Fibonacci eleven — and it was dead. The richer versions (the gap sequence, the repeat count) are not yet built.","happened":"Mostly ruled out. The numbers the glitch itself encodes — its binary position, its frame count, the sum and the first/last frame where a repeat lands — were each read as word-list positions and swept dead, plant-validated. The gap sequence (5,3,2,1,1) is identical on every piece by definition, so it can't tell them apart. What remains under this idea is the eyes-as-binary reading (see the next entry), not the frame geometry.","status":"ruled-out","links":"[]","media":"[]","sort":289},{"id":150,"num":150,"title":"Read the eyes as ones and zeros","picked":"On four pieces the white eyes are literally the '0' digits of a binary strip sitting in front of the creature — the eyes ARE binary digits. The robot drop teaches binary reading right in its names, which all end in a 6-bit code.","tested":"Not yet run — it needs the eye positions marked first, by eye or a targeted detector, before a computer can turn them into a number.","happened":"Still open. The most Pindar-elegant idea we have not executed: if the eyes pick out digits of a binary number, that number can be read as a word.","status":"open","links":"[]","media":"[]","sort":290}]