# Proof of humanity: fingerprints in the artblocks code

Do you think digital art is sterile? That a work made with a computer must somehow contain less feeling, less judgement, less of the person who made it?

Spend some time with the code.

There are spelling mistakes in it. There are abandoned colours, switches nobody connected, messages to friends, and numbers that do something slightly different from what their names suggest. There are things an artist tried and left behind. Sometimes a mistake changes what a collector owns. Sometimes the artist notices, likes the result, and keeps it.

AB5D’s essays on the first five hundred Art Blocks projects were commissioned to look closely: at the work, its construction, its mint, and particular outputs. Read together, they also describe the working habits of a young artistic culture. The finished images are accompanied by an unusually intimate record of how people made them. The [complete research notes](#research-notes) below collect every finding in Claude’s sweep, including the less conspicuous errors, mint decisions and artist anecdotes.

“Proof of humanity” is a name for those traces. It is not a test that distinguishes a human from an AI. Machines can make mistakes too. What matters here is that the traces belong to particular works, particular artists and particular histories. They give us something to read beyond the claim that an image came out of an algorithm.

{{construction-token-by-jeff-davis}}

## The first typo

The AB5D essay on Jeff Davis’s *Construction Token* records a tiny slip in the loops that read the token’s hash: `Sting`, where the expected type is `String`. The essay reports that Processing.js tolerates it. The works render. The missing letter remains. [^h01]

There is something pleasing about encountering this in the project whose first token precedes the first mints of *Genesis* and *Chromie Squiggle*. One of the founding documents of Art Blocks contains the sort of error anyone who writes at a keyboard recognises. Its ordered geometry does not announce the slip. You have to look underneath.

A painter can leave a thumbprint in wet paint. Here the impression is linguistic. It sits in the instructions that produce the image, rather than on the image’s surface. The machine’s ability to carry on makes the mistake almost invisible, and preservation makes it available to a later reader.

Other slips are more consequential. In John Gerrard’s *Petro National*, the series essay identifies a shader constant that defines pi as `3.1459`, rather than approximately `3.14159`. It traces the value into the cloud layer and the viewing-angle calculation for the oil’s iridescence. This is not evidence that every visible feature can be attributed to that number, or that a corrected render would look dramatically different. It is a specific discrepancy inside a highly elaborate illusion. [^h02]

The work turns national oil consumption into beautiful, uncomfortable images of petroleum on water. Its engineering serves an argument about the world. A short numerical constant survives inside that engineering. The two facts belong together: technical ambition and ordinary fallibility, preserved in the same object.

{{petro-national-by-john-gerrard}}

## The switch nobody flipped

William Mapan’s *Anticyclone* makes the question unusually visible. These images already seem to carry the pressure and hesitation of a hand. Fine strokes accumulate into currents. Colour catches in a grid. The paper looks worked, and the margins allow the drawing to breathe.

In a pre-release interview, Mapan said that he wanted viewers to slow down and imagine the movement. He explained the border in precisely those terms: he had left space “to let the air in”. The appearance of paper is part of an artistic intention, made through code. [^mapan]

The supplied technical follow-up identifies a pair of settings called `issqusse`, apparently an echo of the French *esquisse*, a sketch. The spelling is suggestive, though it cannot tell us what Mapan meant when he named it. The second member of the pair enables the Graphite treatment. The first is fixed at zero.[^h03]

That distinction matters. An earlier explanation described an unused switch that would add a sketched frame. The supplied experiment shows the opposite. Changing `issqusse[0]` from `0` to `1` removes the confinement of the colour to the frame, reduces the paper-grain treatment, and changes the colour balance around contours. The drawing runs to the edge; the paper looks cleaner and cooler.

{{comparison-0}}

For the comparison above, Claude’s supplied experiment used the same token hash and changed that one setting in the indexed script. The two panels let us see a decision that the ordinary token display conceals. Much of the structure persists, while the relationship between drawing, margin and paper changes.

The second comparison makes the point again with a different composition. A single dormant setting carries a coherent alternative treatment across more than one work.

{{comparison-100}}

These are experimental renders, not additional Anticyclone tokens or an artist-authorised new edition. They were supplied at 707 × 1,000 pixels per artwork, rendered in headless Chromium rather than the official Art Blocks renderer. Fine grain and colour can vary with the rendering environment. The exact intervention is documented in the experiment note below[^experiment].

We cannot infer from the unused switch that Mapan forgot something. He may have rejected the alternative, retained it for testing, or found the framed treatment more convincing. Nor can we claim that nobody had ever seen it before. What survives is narrower and more interesting: a route the program contains but the released setting does not take.

There is a related finding in the Anticyclone series essay[^h22]. It describes a table keyed to individual token numbers, affecting the empty cells of 634 outputs. Such a table makes the program more particular than a general recipe applied identically to every seed. The original essay goes further and infers a chronology of manual intervention. Establishing exactly when and how the table was assembled would require version history or testimony from the artist. The table alone does not settle that history.

{{anticyclone-by-william-mapan}}

## When punctuation becomes rarity

An error can stay quiet in a name. It can also decide the population of an edition.

In *The Source Code*, Ofir Liberman’s system assigns categories through numerical bands. The series essay describes tests that exclude the exact boundaries: a number must be greater than one limit and less than the next. Four values, 60, 70, 80 and 95, fall through the gaps into the final category. Its replay counts seven such tokens among the nineteen labelled Light. [^h07]

The distinction between “less than” and “less than or equal to” can therefore become a collecting fact. Seven objects occupy a category through the behaviour of the comparisons. Once those objects exist, their history includes that behaviour. A tidier imagined program would produce a different account of the edition.

The Fidenza essay describes two other kinds of discrepancy: an assignment of `"med"` inside a turbulence condition where a comparison might be expected, and a complete Retro palette absent from the palette-selection path. The first changes execution; the second preserves an unused possibility. The code can contain a colour world that no minted output reaches. [^h06]

Even absence can have a mechanism. The Unigrids study reports that its letter selection spans A to Y, leaving Z outside the range. Looking at one work would hardly reveal the omission. Reading the program alongside the edition makes it legible. [^h10]

None of these findings makes a work valuable by itself. They change the questions we can ask of it. Which variations were possible? Which were selected? Which existed in the artist’s working material but could never appear? A rarity list gives counts. A close reading can explain why those counts have the shape they do.

## The label and the thing

Collectors often meet a generative artwork through its traits. A palette has a name. A composition belongs to a category. A checkbox says a feature is present. Those labels feel authoritative because they arrive beside a token identifier and a contract address.

But a description is also made by someone.

The AB5D essays document cases in which the logic used to write traits differs from the logic used to draw. Hideo’s *Mecha Suits* offers a clear example. The artwork skips an extra-arms random choice when a suit has claws. The essay’s reconstruction finds that the published traits behave as though that choice were always consumed. From that point, later choices can read different positions in the sequence. The reported mismatches affect Jetpack or Inverted palette on eight of fourteen claw suits. [^h18]

Think of two people reading the same list aloud. If one skips an item and the other does not, the fact that they began with the same list will not keep them in step. Deterministic randomness still depends on the order in which it is consumed.

{{mecha-suits-by-hideo}}

Emily Xie’s *Memories of Qilin* poses a different question. Its series study identifies a Vertical composition routine whose outputs are labelled Sweep, and reports forty-eight examples. That may reflect a naming decision or an oversight; the code and labels alone do not establish intention. It does mean that the published category can conceal a formal distinction worth seeing. [^h16]

This gives the collector a useful responsibility. Look at the work. Use the traits, but allow the image to disagree with them. A label is a compact account of a much richer procedure, and that compression can lose information even when nobody has made a mistake.[^p04]

The same applies to an archive’s essays. They are arguments and research records, capable of correction. The Anticyclone switch is a small example: an attractive explanation became more precise when someone tried the alternative setting and looked at the result.

## The mistake an artist wants

Sometimes there is no need to guess whether a surprise was welcome. The artist tells us.

Writing about *Subscapes*, Matt DesLauriers recalled discovering a bug the day before release. In some Lattice outputs it made too many holes. He decided to retain the behaviour with a fifty-percent chance within that treatment. His account makes the act of judgement explicit: he saw what happened and chose to keep it. [^deslauriers]

That choice is familiar from other materials. Ink spreads farther than expected. A glaze breaks. An accidental interval suggests the next phrase. The event becomes part of the work through the artist’s attention to it.

Generative art gives this decision another reach. The artist can preserve a surprise as a rule that future outputs encounter. An accident becomes repeatable without becoming identical. The result still depends on its surroundings, on the other choices made for each particular work.

{{subscapes-by-matt-deslauriers}}

The artist’s presence is visible in this act of selection. It is there in what is allowed to recur, how often it can recur, and what other conditions accompany it. A released algorithm carries judgements about a field of possibilities. Some of those judgements begin with an outcome the artist did not plan.

## Notes to friends

Not every trace needs to alter the image.

Loren Bednar’s *Parade* contains a short acknowledgement in its script: `tx piter, you rock :)`. The AB5D study records the line and leaves its recipient unidentified. That restraint is useful. We do not need to invent a biography to recognise a thank-you. [^h25]

It gives us a glimpse of the social world around the work. Someone helped. Someone was grateful. The sentence has no obligation to appear in the collector’s image, but it belongs to the material that produces it.

{{parade-by-loren-bednar}}

The research sweep also points to recurring fragments of random-number and weighted-choice code across different artists’ projects. Similarity can reveal common tools and inherited assumptions. It does not, by itself, establish who copied from whom or the conversation in which a fragment travelled. That history needs its own sources. Still, the possibility is valuable: programs can carry evidence of a community’s working methods as well as an individual’s style.[^p01][^p02][^p03]

The mint has its own human record. A release involves timing, allocation and decisions about who receives a work. The final *Chromie Squiggle*, #9999, entered LACMA as a gift of the artist in August 2024. A series founded on generative variation ended with a particular act of giving.[^h20] [^lacma]

## Read the work, then read underneath

Arguments about AI and art often move quickly towards an abstract contest between humans and machines. The early Art Blocks record rewards slower attention. It gives us named people building systems, revising them, living with their limits, and deciding when the work is ready to meet an audience.

Code makes some of that labour unusually inspectable. We can read the typo, follow a branch, compare a description with a drawing, or change a setting in an explicitly labelled experiment. We can also reach the limits of what inspection proves. A dormant branch is not an explanation of why it was abandoned. A table is not a diary. An unusual variable name is not a confession.

Those limits make the record more human, too. The archive contains evidence, while motives remain partly with the people who made the work.

The AB[500] deserves this kind of attention because its images have substance, and because the means of making them can remain available beside them. The code extends the encounter. It lets us see the choices that support an apparently effortless surface, and the small things that escaped the artist’s notice or earned a place through it.

Look at the work first. Follow the line, the colour, the space left at the edge of the page. Then read underneath. There are people there.

---

### Sources and method

This essay is an AB5D editorial synthesis of the supplied Claude research sweep, the AB5D series studies and the supplied Anticyclone comparison experiment. Code findings and edition counts attributed to the series studies have been checked against those texts; this article is not an independent replay of every underlying script. The Mapan interview, DesLauriers’s account and LACMA’s acquisition record were consulted directly. The Anticyclone comparisons were supplied by Claude and were not independently regenerated for this article.

The six collection illustrations were checked against Art Blocks’ indexed ownership records on 21 September 2026. Each links to its token. They illustrate the discussed series; their inclusion does not imply that the pictured token exhibits every quirk discussed. Anticyclone #0 and #100 are comparison subjects, not identified here as AB5D holdings. Artwork rights remain with the respective artists under each project’s terms. Original essay prose: CC0.


<h2 id="research-notes">Complete research notes</h2>

All 65 entries from Claude’s supplied sweep appear below: 25 headline findings, eight cross-project patterns, 31 additional project entries and one group of artist contexts. Compound entries retain all the individual findings within them. The inventory covers errors, quirks, unused possibilities and deliberate choices; not every item is a mistake.

Claude reported searching all 500 series essays, approximately 1.11 million words, using about 60 signal terms. The search returned 839 paragraphs from 297 essays. Twelve headline claims were checked against the essay text. The sweep did not independently replay the claims against the scripts. The notes retain that reported status; corrections and qualifications are attached to the affected entries. Browser behaviour, holdings, supplies and registry values describe the research snapshots, not a continuously updated register.

Notes 1–25 contain the headline findings, 26–33 the cross-project patterns, 34–64 the additional project findings, and 65 the artist contexts. Notes 66–69 document the Anticyclone experiment and the primary sources used in the essay.

///Footnotes Go Here///

[^h01]: **Construction Token (Jeff Davis).** The hash-reading loops declare their variable as `Sting`, not `String`. Processing.js tolerates the typo, so all 500 tokens render. This is the first artwork minted on Art Blocks, and it has a typo in it.

    Source: AB5D’s [construction token series study](/series/construction-token-by-jeff-davis/).

[^h02]: **Petro National (John Gerrard).** The shared shader header defines pi as **3.1459** instead of 3.14159. The wrong value feeds the cloud layer and the thin-film viewing-angle term.

    Source: AB5D’s [petro national series study](/series/petro-national-by-john-gerrard/).

[^h03]: **Anticyclone (William Mapan).** The Graphite switch is named `issqusse` in the code, an apparent misspelling of the French *esquisse* ("sketch"). A second flag is always off, so the sketch-frame shader branch never runs.

    **Correction.** The supplied follow-up identifies the disabled member as `issqusse[0]`, the first setting, fixed at zero. The second controls Graphite. Enabling the first removes the frame confinement, reduces the paper-grain overlay and shifts contour colour balance; it does not add a sketch frame. The comparisons and experiment method are on this page.[^experiment]

[^h04]: **Misspellings in the published traits.** "verticle" appears in JPEG (Leegte), matching the script's own console messages. "Balaced" is on 38 Neophyte MMXXII tokens. "Acqua" and "vulcanic" are in Dino Pals. "Flourescent" is in The Eternal Pump. The Source Code has "Dmension" and "Settelment" in its code, although the published metadata spells both correctly.

    **Reading note.** These are spellings flagged by the research sweep. Their presence does not establish that every word is an error: “acqua”, for example, is an Italian word.

[^h05]: **Misspellings in the on-chain descriptions.** "Etherum" (Mind Maze). "We are so accustom", "it's path" and "effected me" (Solar Transits, Hodgin). "no-one will ever experience again" (Because Unless Until). "dependancies" (JPEG, Window). "comunicate" (Your Story). "entering and existing the stage" (Woman, Life, Freedom).

    **Reading note.** These are textual irregularities flagged by Claude, not all independently established grammatical errors. They are retained here as part of the complete inventory.

[^h06]: **Fidenza (Tyler Hobbs).** The turbulence condition *assigns* the string "med" instead of comparing against it, so high turbulence is gentler than intended. A complete **"Retro" palette** is defined but missing from the draw, so no Fidenza can have it. Some colour-weight tables sum to 1.02–1.04, which makes Luxe's darkest brown unreachable. 28 of the 35 spirals carry turbulence labels even though spirals set turbulence to zero.

    Source: AB5D’s [fidenza series study](/series/fidenza-by-tyler-hobbs/).

[^h07]: **The Source Code (Ofir Liberman).** Strict inequalities at the band edges send four exact values to "Light". That covers 7 of 180 tokens, about a third of all Light cards: "an artefact of four missing equals signs". A one-character indexing slip produces "White undefined".

    Source: AB5D’s [the source code series study](/series/the-source-code-by-ofir-liberman/).

[^h08]: **The Eternal Pump (Dmitri Cherniak).** The "Hexagons" are circles drawn with too few segments: the code asks for 3 instead of 50, and Three.js r124 doubles that to six sides. The last ring always has scale zero, so the published Rings value is one less than the loop actually draws.

[^h09]: **Gazers (Matt Kane).** A misplaced comma makes the "Aluminum" background always return the fourth grey, so two warm off-whites can never appear. A one-argument `constrain()` returns NaN in 375 of the 1,000 tokens. The lunar cycle is mapped to 29 days instead of 29.53.

[^h10]: **Unigrids (Zeblocks).** The random letter is scaled to 25 steps above "A", so **Z never appears**. Each of A to Y appears 38 to 60 times.

    Source: AB5D’s [unigrids series study](/series/unigrids-by-zeblocks/).

[^h11]: **Incantation (Eliya Stein).** The SZR-12 size class reads the bit derived from the *x* in the hash's "0x" prefix, so it can never occur.

[^h12]: **Neophyte MMXXII (Sterling Crispin).** Each palette table has 19 entries but the index only reaches 18, so the dark-olive palette is unreachable. A random test called without its scale always fails, so only one of three grammar fragments is ever used.

[^h13]: **Dino Pals (Hideo).** The albino test uses `<=` against 0.010, and #41 scored exactly 0.010. A strict `<` would have given the series one albino instead of two.

[^h14]: **Space Birds (Mark Cotton).** Seven tokens are labelled "Twisted Triangles: True" but draw a pentagon. The corrected counts are 39 triangles and 41 pentagons, not 46 and 34. Malformed colour codes render as opaque white: one is missing its "#", one has seven hex digits, one uses the letter "o" for zero, and one carries a stray quote. One branch works only "by accident", because an arrow function stands where a comparison was meant.

[^h15]: **Chroma Theory (Pawel Dudko).** 3 of 5 published traits do not match the live work. The first token computes a background hue of 83, but its trait says 183.

[^h16]: **Memories of Qilin (Emily Xie).** There is a hidden fifth composition, "Vertical", and it labels itself "Sweep". All 48 Verticals are published as Sweep.

    Source: AB5D’s [memories of qilin series study](/series/memories-of-qilin-by-emily-xie/).

[^h17]: **The Spring Begins… (Cole Sternberg).** On token 317000295, 10 of the 13 published traits disagree with the program. The other 486 tokens all match.

[^h18]: **Mecha Suits (Hideo).** The feature script always draws the extra-arms chance, but the artwork skips it for claw suits, which shifts every later draw. As a result, the Jetpack and Inverted traits are wrong on 8 of the 14 claw suits.

    Source: AB5D’s [mecha suits series study](/series/mecha-suits-by-hideo/).

[^h19]: **Calendart (Steen & n-e-o).** The palette labels "Energy" and "Coffee Shop" appear to be swapped.

[^h20]: **Chromie Squiggle.** Public minting was "permanently paused" at 9,069. The remaining 931 were given out by hand over three and a half years, ending with #9999, a gift to LACMA on 21 August 2024.

[^h21]: **Construction Token and the Genesis launch.** The launch was improvised by email "late on Thanksgiving", with backup scripts ready "just in case no one was interested". Davis "unpaused the project along with you and DCA". Construction Token #0 (15:58:01 UTC) predates both Genesis #0 and Squiggle #0.

[^h22]: **Anticyclone (Mapan).** The largest data structure in the script is a *hand-written table*, keyed by token number, that empties specific grid cells in 634 of the 800 tokens. It matches each token's hash exactly, so it had to be written after the mint, with every output in view.

    **Qualification.** The table and the reported 634-token coverage are the findings. “Hand-written” and “had to be written after the mint, with every output in view” are Claude’s interpretation, not an independently established production history. The matching table does not alone prove how it was assembled, the timing of every edit or whether a person inspected every output.

    Source: AB5D’s [anticyclone series study](/series/anticyclone-by-william-mapan/).

[^h23]: **Hard-coded artist proofs.** The Source Code #0 overwrites all eight parameters with literals. Recursion (Hevey) hard-codes the artist's own mint so it shows the system at full depth. BUSIEST renders five hand-picked seeds in place of its on-chain hashes. while(true) fabric (project 501) swaps in an "artist-selected" seed. In Cole Sternberg's work, every eleventh invocation is hand-set.

    **Scope note.** Claude included while(true) fabric, project 501, as a comparison. It is retained here without treating its project number as proof of membership in the AB[500].

[^h24]: **Supply that changed its mind.** Mono no Aware was advertised at 1,024 and cut to 414. The Source Code promised "1024 karma cards" and minted 180. FOCUS was "a 1000-image long meditation" and stopped at 567. Escherly Seeds was "set to 327" and completed at 30. Coalition sold a 37th token on purpose: "the algorithm self-destructs after mint #35".

    **Reading note.** Supply changes and release decisions are not necessarily mistakes. This entry records differences between announced and realised editions.

[^h25]: **Personal messages in the code.** Parade (Loren Bednar) opens with the bare string `'tx piter, you rock :)'`. BUSY (James Merrill) prints "Hey you! Look around coords", dates itself "April 2024 - March 2025" and thanks "Bantam Tools for making EPIC drawing machines". 70s Pop Series Two (Daniel Catt) begins with an unused string holding a Howard Moss poem. Blockbob Rorschach (eBoy) ships unminified working source with a live `console.log`.

    Source: AB5D’s [parade series study](/series/parade-by-loren-bednar/).

[^p01]: **Folk code: one snippet, many artists.** The same xorshift seeding snippet reads `hash.slice(0,16)`, "0x" prefix included, then truncates to 32 bits and gives one of only 1,000 values per draw. It appears in Amoeba, Blockbob Rorschach (released the next day, by a different artist), Mental Pathways, Return, Ringers, Sentience, The Eternal Pump, Archetype, Dino Pals and Unigrids. In practice, most of each 64-character hash does nothing. In Archetype, "the other 50 digits do nothing", and in Unigrids, randomising 56 of the characters changed no output. This is shared tooling passed around 2021 Discords: copied code, not independent invention.

    **Qualification.** Recurring code is the reported observation. The asserted route through “2021 Discords” and a specific direction of copying are not established by code similarity alone.

[^p02]: **The 0.99 picker.** A weighted-choice helper scales every draw by 0.99, so the last item in every list is starved. It appears in Return (Aaron Penne), where the inward direction drops from 10/300 to 7/300, and in Dino Pals, Mecha Suits, Sentience and Space Debris. It is a lineage you can trace, like a scribal error copied down a manuscript tradition.

[^p03]: **Each artist's personal fingerprint.** Jake Rockland's Patchwork Saguaros and Speckled Summits share an identical seed slice that starts one character late, reading only 32 of 64 digits, and both define the same colour-mixing function without ever calling it. Pseudo-named arguments (`invert=`, `a=`) that land in the wrong positional slots recur across all three Space Debris scripts, and the unused `GTYPE` variable counts up 0 → 2 across them. A stray global `color` shadows p5's `color()` in both Chromie Squiggle and Sentience.

[^p04]: **The two-script drift.** Art Blocks separates the *art* script from the *feature* script that writes traits. Wherever those two drifted apart, the metadata is wrong: The Source Code, Mecha Suits, Space Birds, 70s Pop Ghost, Chroma Theory and The Spring Begins. In The Source Code, the feature script *corrects* the artist's errors, so the traits are more correct than the artwork. This is the richest source of "misclassification" evidence, and it could carry an essay of its own.

[^p05]: **Bugs kept on purpose.** Artists admit to keeping some bugs. Meridian's **Prime** style is "a bug in my code reduced the palette to a single colour". Subscapes' **Lattice** holes came from a bug found the day before release and kept "as a feature that occurs 50% of the time". LIA's "cheeky blue/red" (Little Boxes) is hue taken mod 255 in a 360-degree space; there are actually four cheeky colours, not the two she names. The outer-polygon mistake in Autology "helped establish the eventual aesthetic". Sudfah is Arabic for "happy accident". The Ecumenopolis wind always leans left, "an unintended behavior whose cause he had not resolved".

    **Reading note.** This group combines documented artistic decisions, technical observations, a title’s meaning and an unresolved behaviour. They are different kinds of evidence, not all admissions of error.

[^p06]: **Machine-dependent output.** Some works give different results on different machines. Scheme 1 in 319 (Santiago) sorts with a random comparator, so colours can differ by browser engine. In BUSY, Safari, Firefox, Edge and Opera switch to multiply blending while Chrome does not. Return renders flat on Apple M5 (ANGLE/Metal) because of an undefined `smoothstep`. The Eternal Pump, Neophyte and while(true) animate at double speed on 120 Hz displays. 319 and Neophyte detect `HeadlessChrome` and behave differently for the platform renderer than for collectors. Fake It Till You Make It keeps "a fleck of live randomness" through unseeded `Math.random`.

    **Scope note.** Browser, device and refresh-rate results are those reported in the underlying essays, not a new compatibility test for this article.

[^p07]: **The ubiquitous #0.** Dozens of projects show invocation 0 minted days, weeks or months before activation. Examples: Dendro, three months before #1; Enigma, three months; Latent Spirits, with an unexplained #1 eleven days early; Quarantine, where 27 tokens were minted before opening. This is the artist-proof ritual, visible in the timestamps. World Flag #0 is Yemen, out of the emissions order (it would rank 119th), minted five days before activation and held by the artist.

[^p08]: **Registry misclassification (Art Blocks' own data).**
    - `curation_status` reads "factory" while the display fields say "curated" for Alan Ki Aankhen, Running Moon and Sudfah.
    - `display_static: false` is recorded on works that call `noLoop()` and draw one frame: CryptoNewYorker, Blockbob Rorschach and Mental Pathways.
    - Price fields are denominated in "BRTRF" and "RTHM" tokens instead of ETH, which are traces of Bright Moments' Golden Token claims.
    - Rotor loads a CDN path labelled p5 1.0.0, but the file there is **v0.10.2**.
    - The World Flag licence label says BY-NC-ND while the link points to BY-NC. Squares carries a licence string with a leading space.

    **Reading note.** These are index and dependency discrepancies reported in the sweep. A non-ETH denomination associated with a claim mechanism is context, not necessarily a registry error.

[^o01]: **24 Heures (André).** The wall-clock palette shift is a single interval, not a modulo, so it fires once, 24 hours after load, and never again.

[^o02]: **319 (Santiago).** 8 of 19 rocks carry a Colour label they do not show. The script fills 1,048,576 random values where 65,536 are needed. An undocumented "S" key turns on a glitch mode.

[^o03]: **70s Pop Ghost (Catt).** Tokens #1–3 have empty feature sets, so "Haunted: True" appears 61 times instead of 64. A missing bracket leaves the final edge of every straight tile unstroked.

[^o04]: **8 (Bard Ionson).** An auction designed to close in minutes took 655 days. The identity symbol is selected for every token but drawn on only 42 of 128.

[^o05]: **Aitherios (Ledezma).** Spanish variable names (`ruido`, `rumbos`) mark it as "a personal working tool".

[^o06]: **Being Yourself While Fitting In (LIA).** Greyscale brightness is chosen "at random" between 255 and 255. A six-way switch assigns the same 60,000 in every branch.

[^o07]: **BUSY (Merrill).** A missing `break` makes "down" fall through to "left". Highways, trains and sleepers can never appear.

[^o08]: **CryptoNewYorker (Qian Qian).** 34 garments are published under 31 names. The Accessory key is missing on 119 tokens. Every token was minted by a person standing in a gallery.

[^o09]: **Elevated Deconstructions (Luxpris).** The float seed loses its low bits. Hairline count depends on resolution: 7,500 at 2,400 px, but 3,750 at the size Sotheby's catalogued.

[^o10]: **Fluiroso (Brocher).** A diffusion bug amplifies density to about 178,000 by frame 300. The Transparency labels are inverted. The code name `noiseDensity` leaked into the metadata.

[^o11]: **Ilumz (Wolffia & Motus).** Every shape gets ID 1, so the collision skip never works. Neon is drawn in 13 tokens but invisible in 7. "Chance, mistakes, serendipity… sing of human touch."

[^o12]: **JPEG (Leegte).** At exactly 2,400 px the script logs "test" and draws at 1/24 scale, so every official render is a 100×100 JPEG enlarged 24 times.

[^o13]: **Karma (Kala).** The hash is parsed without a radix, so 96 of 264 chunks become zero. The Mohaniya Age label is off by one on 4 of 5 tokens.

[^o14]: **LeWitt Generator Generator (Chan).** `placeFiftyPointsAtRandom` places 12 to 51 points.

[^o15]: **Meridian (DesLauriers).** The "60 colours" option draws from 55. No token has 60.

[^o16]: **Patchwork Saguaros (Rockland).** The 1-in-54 "Forgotten Horizon" never minted ("we didn't get to see it at all"). #12's speckle comes from patch and dot-grid misalignment.

[^o17]: **Pre-Process (Reas).** The pause flag starts "paused" while the work is running, so the first P press does nothing. The speed maths mixes areas and lengths.

[^o18]: **Quine (Larva Labs).** Full "Field" and "Motif" engines exist but can never be picked. Two palettes are both called "BX". The symmetry trait lines are commented out.

[^o19]: **Reflection (Jeff Davis).** A p5 WebGL triangle fan leaves a diagonal seam (#91). One hue's direction is read from the wrong byte.

[^o20]: **Ringers (Cherniak).** Bullseyes include five hand-picked byte values (33, 43, 69, 109, 110). That is a human's favourite numbers written into the canon.

    **Qualification.** The five byte values are the reported code finding. Calling them the artist’s “favourite numbers” goes beyond what those constants establish.

[^o21]: **Sentience (pxlq).** Left and right neighbours are computed and never used, which is why every work reads as vertical bars. The duplicated palette colours were copied over from Cyber Cities.

[^o22]: **SLCE (Stranger in the Q).** It discards exactly 419 random values, the project's own number.

[^o23]: **Speckled Summits (Rockland).** #67 drew Sun Sync, but its label omits it, and it is held by the artist.

[^o24]: **Space Debris Ravaged.** Pseudo-named arguments put the grid on the twin strip. An `#000003` "colour" is the background, so those blocks act as erasers.

[^o25]: **Star Flower (Alexander).** A misspelt property routes every path through the alternative branch.

[^o26]: **Subscapes (DesLauriers).** A ternary precedence slip turns every coast into a plateau ending in a cliff.

[^o27]: **The Wrapture (Cherniak).** No published traits at all. The index allows 51 invocations; 50 were minted.

[^o28]: **Transparency (Usnea).** Paper colour names do not match the stored colours ("white" is tan in code).

[^o29]: **while(true) (Wander).** The gamut check tests the colour it has just written, so the fallback never fires.

[^o30]: **Woman, Life, Freedom (Fatemi).** The scroll listener leaks, so page scrolling stays blocked. There are typos in the in-work texts.

[^o31]: **World Flag (Gerrard).** Terrain names are mismatched ("Ash" is sand). The day-of-year calculation mixes UTC and local time.

[^context]: **Human context from the artists.** IEVA is named after the investigator who brought a criminal case against shvembldr "for making art". Recollection grew out of a false memory of climbing Mt Fuji. Speak To Me's title came from a note scribbled half-asleep. Gels simulates "novice researcher" lab errors. FOCUS can be recreated "with rulers, a calculator, and art supplies".

    **Scope note.** These are biographical and artistic contexts reported by the sweep, not code errors. They are included to preserve its complete set of findings.

[^experiment]: **Anticyclone: experiment method.** Supplied by Claude for tokens #0 and #100, using Art Blocks project 304 on contract `0xa7d8d9ef8d8ce8992df33d8b8cf4aebabd5bd270`. Within each comparison, the token hash stays the same. The only reported script intervention is changing the first member of `issqusse:[0,…]` to `issqusse:[1,…]`; the second member controls Graphite. The stated effects are removal of the frame mask confining the colour layer, reduction of the paper-grain overlay from 0.02 to 0.003, and blending towards a lower-saturation setting along contour edges. The supplied renders used headless Chromium, three.js r124 and 707 × 1,000 pixels per artwork. They were not independently regenerated for this publication. They are modified research renders, not new tokens or an artist-authorised edition; fine grain and colour may differ from the official renderer. This corrects the earlier account that the switch adds a sketch frame. The experiment does not establish why the setting was left off, when it was last used or who previously saw it.

[^mapan]: **William Mapan on Anticyclone.** Jeff Davis, “In Conversation with William Mapan”, Art Blocks, 27 April 2022. Mapan explains that the border leaves space “to let the air in”, and describes the grid as a structure through which the viewer can imagine missing pieces. [Original interview](https://www.artblocks.io/articles/in-conversation-with-william-mapan).

[^deslauriers]: **Matt DesLauriers on the retained Lattice bug.** “Subscapes (Part 2: Traits)”, 5 May 2021. The artist describes discovering excessive holes the day before release and retaining that behaviour with a fifty-percent chance within the Lattice treatment. [Artist’s account](https://mattdesl.substack.com/p/subscapes-part-2-traits).

[^lacma]: **The final Chromie Squiggle.** Dhyandra Lawson, “Erick Calderon’s Chromie Squiggle 9,999”, LACMA Unframed, 21 August 2024. The museum records #9999 as a gift of the artist. [Acquisition record](https://unframed.lacma.org/2024/08/21/erick-calderon%E2%80%99s-chromie-squiggle-9999).