Week of 2026-10-19 → 2026-10-25
Three new mark-families, and a second substrate
The engine stopped being a single vocabulary on a single screen.
The week in one line: the engine opened three new mark-families (fractal, raster, text) and then crossed into a second substrate (the pen), where its marks had to be re-learned, not just re-rendered.
Conceptual learnings
- Three veins opened in six days — the archive's biggest blind spots, closed. The engine had zero fractal/recursive forms, zero raster/video register, and zero text. Dragon (recursive path) + subdivide (recursive partition) opened fractals; scanline opened raster; lineprinter opened text. The lesson: when a pioneer's essence is a whole class the engine can't express, that's the signal to change the program, and the payoff is a vein, not a piece. Next absent classes to watch — sound/waveform, network/graph, moving-image loop.
- "Change the program, not the drawing" scales to a new SUBSTRATE. The deepest move wasn't a character — it was re-voicing the same genome in ink. The plotter work is all export-only: the raster never changed (
7ddf2b3dheld through fifteen engine versions). The program isn't "how the engine draws pixels" — it's "how a canonical seed becomes a mark," and a pen asks different questions of the same seed. - The FORK RULE — the cleanest criteria-line we've drawn. A translation is a faithful export while it reproduces the parent's checksum (page, pitch, pen, caption are output-only). It becomes a study the moment a decision changes what is drawn. This gave the Plotter Studies track its spine and kept the daily archive's provenance clean.
- A medium teaches by failing. The first physical plot failed instructively: the digital ink-density ramp did not survive as a tonal order on paper — the pen read it as texture, not tone. That finding generated Study 02 (true solid black) directly. Principle: plot early, let the substrate contradict the screen, and treat the contradiction as the brief for the next study.
- Restraint lived in the RENDER, not in refusing the capability. Four new characters in five renders could have been creep. It wasn't — each debut was polarized to one signature (the new mark alone, other weights zeroed). The capability was justified by an absent class; the restraint was in the composition.
- The plotter is a DISCOVERY INSTRUMENT that feeds the main engine. Exploring Overstruck in ink surfaced a property the screen never exposed: the lineprinter's tone field has a latent two-tone structure. That discovery came home as a native screen capability (
lpColorMode, below). The pen asks a question the pixel didn't → the answer becomes a dial on the main instrument.
Visual learnings
- Each new mark has a distinct, nameable signature — and reads cleanest ALONE.
- dragon — deterministic order that looks grown (a space-filling fold).
- subdivide — a quadtree whose cells halve and darken; directional bias makes it a trajectory (coarse→fine = a visual argument).
- scanline — saturated horizontal bands interleaved with scan-text; the raster/CRT register.
- lineprinter — the charset AS a tonal palette; hard-edge nested-rectangle tone from overstruck glyphs.
- On paper, the glyph ramp is a TEXTURE vocabulary, not a tonal one. Darkness order is not preserved by the pen — "tone" on screen becomes "pattern" in ink. To restore tonal depth you must change the fill (full-cell merge → true black). The single most important visual fact the plotter taught.
- PEN WEIGHT is a second compositional axis — the week's best new lever. Pairing a bold dark pen with a fine light one makes the two colours differ in weight as well as hue: the thin layer recedes, the bold layer advances. A two-colour plot gains real depth — not two flat layers, but a near-ground and a far-ground.
- Direction + colour = a WEAVE grammar. Splitting marks by orientation turns the
#field into woven cloth; recolouring per cell gives textile structures — plain, swap (vertical emphasis), houndstooth (checker), twill (diagonal). The line-printer's crosshatch was always a latent weave; the pen made it literal. - A solid on paper wants to MERGE, not tile. Per-cell inset hatch reads as stripes; full-cell fill at the pen pitch lets adjacent dark cells merge into true black. Small cores stripe more than large swatches. Rule: remove the inset for the darkest tier so neighbours join.
The build — two-colour ribbon (lpColorMode, v3.20.0)
The plotter's two-tone discovery wanted to be native to the screen, because red/black is the classic typewriter / teletype ribbon — a real piece of computer-art material history, not just a plot trick. So lineprinter gained lpColorMode ('figure' = dense glyphs in a 2nd ink; 'nested' = alternate by nested ring) + lpInk2. A future daily text source that is two-colour — a red/black ledger, a two-ribbon typewriter poem, teletype with red overprint — now renders as a canonical piece with a real checksum, not only as a plot. Default 'mono' is byte-identical (Day 147 still 7ddf2b3d). The first capability the plotter studies gave back to the daily engine.
What this generates next
- Remaining absent classes (treat as priority, like fractals were): sound/waveform, network/graph, moving-image loop.
- A Fractals room is overdue (dragon + subdivide live) — and a Text / Line-Printer room becomes defensible on the 2nd text day.
- Plotter Studies open fronts: jet-black solid, a pen-first glyph vocabulary, >2 colours, and the richest — weight-contrast as a design primitive (every two-colour piece now asks "which layer recedes?").
- Next text sources to watch: Knowlton EXPLOR, Houédard typestracts, Alison Knowles House of Dust (1967, first computer poem).
The through-line: a mark is not fixed — it changes meaning with the class it belongs to (fractal/raster/text), the substrate it lands on (pixel vs pen), and the weight that draws it. Three new dials on the same instrument.