cmyk
provenance and rarity
cmyk is a print shop rebuilt as a generative system, and a shop writes everything down. This is the record: what each mark on the sheet means, how the press behaves, what the failing press actually does, and the odds for every trait in the edition of 144.
The rates here are the weights declared in the engine itself, and they are verified against ten thousand test renders before each release build rather than estimated after the fact.
the press
The piece does not draw a picture of printing. It runs a press, and the sheet is whatever comes off it. Six mechanical facts govern every token.
Plates lay down in order. A process job puts cyan down first, then magenta, then yellow, then black, and the color you see is what those four passes make together. One sheet in ten runs the black first instead, an underprint, and the whole piece changes character because the later inks are sitting on a dark bed rather than on paper.
Ink is subtractive on paper and additive on black. On a light stock the inks multiply, the way ink actually behaves on white paper: overlap two and you get a darker third color. On a black stock they screen, because on a dark ground an ink can only add light. This is why Rhodamine on Black C glows and a black tint ladder on black paper would be invisible, which is why the engine never prints one.
Dot gain is real. Ink spreads slightly where it meets paper, so midtones come out heavier than the plate says. Every halftone in the set is drawn with that swell in it.
Registration is a bet. Each plate has to land where the last one did. Proofs can drift or sit plainly off. Finals are dead-on, because a final that was not would have been pulled.
The sheet measures itself. The calibration bar is computed from the token it sits on, not decorated onto it, and the ink-limit tick fires when the piece runs past what paper could physically hold.
Bad days happen. Ghosting, streaks, and hickeys appear only on proofs, at the rates a working shop would recognize.
Every token carries this as structured provenance: its plate order, its measured coverage, its registration offsets, its dot gain per plate, and the position of every press error on the sheet. Nothing about the sheet is decorative in origin. It all came from somewhere.
the failing press
One token in ten runs on a press that is failing, and it fails while you are looking at it.
Leave the piece open. The first slip comes within ten seconds of the print settling, and after that the plates walk out of register at intervals from half a minute to five minutes, each event shifting the color plates a little further from each other. What you are watching is a press losing its grip in real time. The drift is capped: the piece degrades to a point and holds there, the way a pressman would catch it before the whole run was scrap.
One token in the edition, roughly, runs a catastrophic press. That one has no cap. It keeps walking, and the plates will eventually leave the sheet entirely. Nobody catches it. Someone owns the broken one.
The recipe
The decay is not random and it is not animation. It is computed:
- The token's own hash seeds a second, separate number stream, used for nothing else in the piece.
- That stream sets the schedule of decay events and the direction and distance the plates walk at each one.
- Because the stream comes from the hash, the sequence is identical every time that token is displayed, on any machine, forever. Your failing press fails in its own particular way, and it is the same particular way next year.
- The decay stream never touches the art stream. Two viewers watching the same token see the same decay at the same moments.
- The captured image is the minted state. Decay cannot begin before the piece has finished printing and been captured, so the token's canonical image is always the clean sheet.
- Reload and the piece is perfect again. The damage is real while you watch and it is never permanent. That is the whole point.
A clean run stays clean. It is not that nothing is happening; it is that this press is having a good day, and 89 sheets in 100 do.
the marks
A press sheet is covered in writing that is not the artwork. Crosses, bars, ticks, and labels: the printer's instrumentation, put there to be checked and then trimmed away. cmyk keeps them, because in a shop they are the most honest thing on the sheet. Here is what each one is.
Registration cross
The circled crosshair. On press it is printed by every plate in turn, and if the plates agree, the four crosses become one. If they disagree, you see it here first, before you see it in the image. Proof sheets can carry crosses that drift or sit plainly out of register. Final sheets carry exactly one cross, dead-on, printed in the sheet's own ink and placed at the edge of the field: the piece signing off on itself.
Crop marks
The right angles that sit outside the trim, at the corners, pointing away from the image. They tell the guillotine where the page ends. They are drawn outside the live area on purpose, exactly as they fall on a real sheet, so they never touch the artwork they define.
Fold marks
Short rules at the head and tail, again outside the trim, marking where the sheet would be folded if this were a job on its way to becoming something.
Calibration bar
The row of stepped patches, and it is not decoration: it measures the sheet it sits on. Each patch reports real ink coverage for that token, and the small red tick appears when the piece runs past the total area coverage a press could actually hold on paper. A sheet with the tick showing is a sheet that would have been pulled.
Control strip
The bar of solid patches along the head or tail edge. On a process job it carries the four inks, their overprint secondaries, and gray tints, the classic four-color control bar. On a single-plate job it carries that one ink alternating solid and screened, because one plate cannot print four colors, and a shop that pretends otherwise is not a shop.
Tint ladder
One ink stepped from solid down to nearly nothing, hugging the trim edge. The densitometer's subject, made the subject.
The working words
CYAN, MAGENTA, YELLOW, KEY, BLACK. The plate labels, set in the piece's own letterforms. On some sheets they are large enough to be the composition; on proofs they sometimes end up buried under the ink that came after them, which is what happens to labels.
The hash, written in numerals
Every token writes its own hash in the seven hand-drawn Fibonacci numerals, one Zeckendorf decomposition at a time. Zeckendorf's theorem says every positive integer is a unique sum of non-consecutive Fibonacci numbers, so each character of the hash resolves to exactly one stack of numerals and no other. A miniature registration cross stands in for zero. It is the token's name in the only alphabet this piece has.
The mark
A small m, printed in the sheet's own ink, low in the corner. The shop's imprint, on one sheet in twenty-one.
Press errors
Three things a press does when it is having a bad day, and only proofs carry them. Ghosting is the faint impression of the sheet before this one. Streaks are ink pulled along the direction of travel. A hickey is a speck of dust or dried ink that lands on the plate and prints as a spot with a halo, sitting in the ink rather than on top of it.
The job ticket
The card the piece opens with before the plates come down: the registration cross, the four ink swatches, the shop's m, and the token's hash written in numerals. On press, the ticket is the paperwork that travels with the job.
the furniture glossary
In letterpress, furniture is the wood and metal spacing that locks the type into the chase. It carries no ink and never prints. The press furniture Form is that idea taken seriously: a sheet where the artwork never arrived and the infrastructure is the subject.
One furniture sheet in four is a glossary plate, laying the mark vocabulary out as numbered figures on a light grid, the way a printer's manual would. The figure numbers are written in the piece's own numerals, so the plate needs this page to be read. That is deliberate: the token holds the specimen, the archive holds the key.
| Figure | Mark |
|---|---|
| 1 | Registration cross |
| 2 | Crop marks, outside the trim, pointing away |
| 3 | Fold mark |
| 5 | Control strip, honest to the sheet's plates |
| 8 | Tint ladder |
| 13 | Calibration bar with its ink-limit tick |
| 21 | The shop imprint, the m |
Two cells on the plate are always empty. Furniture keeps its margin.
traits and rarity
Every rate below is the weight declared in the engine, and every one is verified against ten thousand test renders before each release build. The edition is a draw, not a quota: each mint takes its own hash, so the actual counts in the 144 will land near these numbers, not exactly on them. That is the honest arithmetic of a long-form set.
Plates
Whether the sheet runs four-color process or a single Pantone spot.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| process color | 70% | — | 101 |
| spot color | 30% | — | 43 |
Stock
The paper. Marshmallow is the warm off-white the shop ran most; Black C is the rich black that makes the inks glow rather than sit.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| 11-4300 TPX (Marshmallow) | 45% | — | 65 |
| black | 30% | — | 43 |
| Black C | 15% | 1 in 7 | 22 |
| white | 10% | 1 in 10 | 14 |
Form
What is on the sheet. Twenty-two of them: the classic fields the piece was built on, the press-native forms (the ramp, the separations, the makeready), and the quiet end (the inkless emboss, the single mark).
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| outline | 8.8% | 1 in 11 | 13 |
| halftone ramp | 8.0% | 1 in 13 | 12 |
| thirteen | 7.0% | 1 in 14 | 10 |
| packed polygons | 6.0% | 1 in 17 | 8.6 |
| squares | 5.0% | 1 in 20 | 7.2 |
| random polygons | 5.0% | 1 in 20 | 7.2 |
| progressive proof | 4.8% | 1 in 21 | 6.9 |
| plate separations | 4.8% | 1 in 21 | 6.9 |
| ink draw-down | 4.8% | 1 in 21 | 6.9 |
| makeready | 4.8% | 1 in 21 | 6.9 |
| Zeckendorf field | 4.8% | 1 in 21 | 6.9 |
| imposition | 4.8% | 1 in 21 | 6.9 |
| Fibonacci | 4.0% | 1 in 25 | 5.8 |
| curvilinear polygons | 4.0% | 1 in 25 | 5.8 |
| regular polygons | 4.0% | 1 in 25 | 5.8 |
| cubes | 4.0% | 1 in 25 | 5.8 |
| blind emboss | 4.0% | 1 in 25 | 5.8 |
| 02 | 3.7% | 1 in 27 | 5.3 |
| press furniture | 3.0% | 1 in 33 | 4.3 |
| moire | 2.0% | 1 in 50 | 2.9 |
| type specimen | 1.5% | 1 in 67 | 2.2 |
| single element | 1.2% | 1 in 83 | 1.7 |
Sheet
A proof carries the marks of checking. A final is dead-registered and clean, and carries one perfect registration cross in the field.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| proof | 50% | — | 72 |
| final | 50% | — | 72 |
Fibonacci
Which hand-drawn numeral set writes the sheet. The 1 set is declared twice in the engine and the 8 set only once at a low weight: a hand-era artifact from 2023, kept rather than corrected, so the published odds are the real ones.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| 1 | 25% | — | 36 |
| 13 | 20% | 1 in 5 | 29 |
| 2 | 13% | 1 in 8 | 18 |
| 3 | 13% | 1 in 8 | 18 |
| 5 | 13% | 1 in 8 | 18 |
| 21 | 13% | 1 in 8 | 18 |
| 8 | 5.0% | 1 in 20 | 7.2 |
Grid
How many cells the field divides into. Fewer cells, larger marks.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| 1 | 25% | — | 36 |
| 2 | 25% | — | 36 |
| 3 | 25% | — | 36 |
| 4 | 25% | — | 36 |
Run Condition
The press this token runs on. A failing press walks out of register while you watch it. A catastrophic one never stops.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| clean run | 89% | — | 129 |
| failing press | 10% | 1 in 10 | 14 |
| catastrophic | 0.7% | 1 in 143 | 1.0 |
Conditional traits
Three traits only vary inside a gate. Outside it they are fixed by construction, not by chance: a final sheet is dead-on and clean because that is what a final is, and a spot job is one plate because it has one ink. Both rates are given: the rate within the gate, and the blended rate across the whole edition.
Lay-down
Varies on process color sheets (70% of the edition); otherwise single plate.
| Value | Rate within gate | Blended rate | Odds | Expected of 144 |
|---|---|---|---|---|
| cmyk | 90% | 63% | — | 91 |
| single plate | — | 30% | — | 43 |
| kcmy | 10% | 7.0% | 1 in 14 | 10 |
Registration
Varies on proofs (50% of the edition); otherwise dead-on.
| Value | Rate within gate | Blended rate | Odds | Expected of 144 |
|---|---|---|---|---|
| dead-on | 30% | 65% | — | 94 |
| drift | 60% | 30% | — | 43 |
| off | 10% | 5.0% | 1 in 20 | 7.2 |
Press Errors
Varies on proofs (50% of the edition); otherwise none.
| Value | Rate within gate | Blended rate | Odds | Expected of 144 |
|---|---|---|---|---|
| none | 85% | 93% | — | 133 |
| ghosting | 6.0% | 3.0% | 1 in 33 | 4.3 |
| streaks | 6.0% | 3.0% | 1 in 33 | 4.3 |
| hickey | 3.0% | 1.5% | 1 in 67 | 2.2 |
Events
Not metadata traits, but things to hunt. The mark is the shop imprint: Michael’s m, printed small in the sheet’s own ink, one sheet in twenty-one.
| Value | Rate | Odds | Expected of 144 |
|---|---|---|---|
| bleed | 25% | — | 36 |
| buried marks (proofs only) | 15% | 1 in 7 | 22 |
| hero element | 13% | 1 in 8 | 19 |
| tilt | 10% | 1 in 10 | 14 |
| type as image | 8.0% | 1 in 13 | 12 |
| the mark (m) | 4.8% | 1 in 21 | 6.9 |
Ink
Ink is chosen against the stock rather than drawn from a flat table, so it has no single rate. Process sheets run neutral CMYK or one of four plate biases. Spot sheets run one Pantone: Red 032, Orange 021, Yellow 012, PMS 802, Blue 072, Violet, Rhodamine Red, Black C, Warm Gray 10, Cool Gray 9, or Cool Gray 11. The engine will not pick an ink that matches the paper, and a contrast floor keeps every spot ink legible against its ground: yellow on a warm stock deepens toward mustard rather than disappearing. Process inks are exempt from that floor, because process yellow is process yellow.
the token in your hands
A finished piece answers to the keyboard. Click it first so it hears you, then:
Nothing is upscaled: asked for a size, the piece reprints itself at that size from its hash. Pulling plates and replaying the ticket are display-only, and the token restores itself on reload.
determinism
Every token is a pure function of its hash. The same hash renders the same sheet on any machine, at any size, forever: the same plates in the same order, the same ink coverage, the same specks of dust in the ink. Nothing samples the clock, the screen, or the network. The piece can be reprinted from its hash alone long after any particular website is gone.
The drawn glyphs and polygon fields at the center of the work have not been machine-edited since December 2023. Everything around them, the press and its physics, was built in collaboration with an AI system in 2026, and the whole history of that collaboration is versioned and auditable.