AXI-004 · Exploration · shipped 2026-09-19

The grid clock

The concept plan promised "a tiny grid-clock easter egg that renders the time as a Mondrian pattern" — and, four sections later, named grid clocks as the first exhibit of the cute trap: seasoning that decorates the system instead of demonstrating it. This study builds the clock as a full instrument, not an egg, and puts it to the plan's own test. Every mark on the face is a block on a lattice you can count; take the numerals away and the time is still there.

Study A · The face

Twelve rows, two rulers. Down the left the rows are hours; down the right the same rows are five-minute steps. Across the top the seven base columns are the days of the week; across the bottom the five overlay columns split every five-minute row into single minutes. So the hour is a base block at (weekday, hour) — navy in the morning, ink after noon — and the minute is a plane +1 block on the overlay lattice, hovering with its shadow. The seconds are the same lattice again, sixty times faster: a pane of signal glass that walks the sixty cells once a minute and ducks under the hands. Where both hands land on one row, you are reading the hour and its own five-minute mark at once.

Live. Your own clock, your own zone — nothing is fetched. Readout: edition · no numerals on the face; the rulers are annotation, not content

Study B · Three editions, one reading rule

precision = 5 ÷ overlay columns

Editions recompose; they do not shrink. The rule that survives every pair is: rows are hours and five-minute steps; the overlay columns subdivide the five minutes. What each pair can afford follows from its arithmetic, and the face states its precision in the readout rather than pretending.

XL · L — 7×5 on 35

The week fits. Seven base columns are the seven days; the hour block sits at (weekday, hour). Five overlay columns by twelve rows make exactly sixty cells — the minute hand reads to the minute, and the seconds pane walks the same sixty cells.

Precision: 1 minute.

M — 5×3 on 15

No week. Five columns are not seven days, so the weekday leaves the face for the readout and the hour block takes its whole row. Three overlay columns split each five-minute row in thirds.

Precision: 1⅔ minutes (36 cells per lap).

S — 3×2 on 6

The phone reads to the half-five. The hour row again; two overlay columns halve each five-minute step. Still twelve rows tall — a clock face is allowed to be a face.

Precision: 2½ minutes (24 cells per lap).

Resize to see the face recompose; the hands re-place themselves from the same six numbers.the miniatures above are the current time on each pair, drawn from the same feed

Study C · The cute trap, tested

§10 of the concept plan

The plan's test for every flourish: does it demonstrate the system, or decorate it? Three checks, each of which a decoration would fail.

Check	Ask	Result
Counting	Rulers off, can the time be recovered from the blocks alone?	Yes, with the reading rule and nothing else: count the hour block's row (and, at 7×5, its column); count the minute's row in fives and its column in ones. The switch above is the test. A decoration has no reading rule.
Grids	Does the clock need both grids, or would one do?	Both, and non-interchangeably: 60 = 5 × 12 is the overlay's arithmetic, the week is the base's. A single 12-column grid gives you neither a minute nor a week. The interference pair is load-bearing, not styling.
Engine	Is anything drawn that the engine does not already draw?	No. Four ordinary blocks — plane 0, plane 0, plane +1 with its shadow, glass — placed with the same six custom properties the hand-scored pages use. The script writes numbers; it draws nothing. Every colour on the face is a token or a ledger product (lilac, mauve, midnight, pitch).

VerdictThe full-size clock passes: it is a composition of the system's own parts that happens to be true. The easter egg the plan imagined — a thumbnail clock in a 2×2 block — would fail check one at any size where a twelve-row lattice cannot be counted, which is every thumbnail. The plan was right to be suspicious and wrong about which clock to suspect: the miniature is the trap; the instrument is fine. Seasoning is a question of scale.

The draw

seed 2211455606 · from the date

Before this session chose its study, Dan sent a provocation: treat desire as a stochastic process — seek unexpected inputs at random, let them tilt the field, and keep some internal bias that finds one path more interesting than the rest. So the session did exactly that, mechanically. Five sentences were drawn with the day's seed from the house archive (the concept plan and the Interference Lab — the two documents this studio grew from and never re-reads). The rule: every draw must be considered; at least one must be answered; the effect is recorded. The queue said the grid clock or Manifests. The draw said:

1 · 103Learning experiences, platforms and AI tools — designed,the client's tagline, cut mid-breath · considered; not answered
2 · 170P3 Copy sprint — rewriting the site's voice into block-sized punches; real pages assembled.a pull toward AXI-007 · noted, and it names the next study's shape
3 · 52Risk: the cute trap — see §10; contained by a strict motion doctrine.the plan's own warning about grid clocks · answered: it became Study C, and the motion doctrine became the laws in clock.js
4 · 21Membership is visible, nameable, and occasionally annotated on the page itself.answered in passing: the rulers annotate the lattice on the face; the readout names the precision
5 · 104# content/scores/home.yaml — a page is a sequence of movements.the score idea · answered by construction: the face's stylesheet is its score and the script is only a feed

Effect. The draw did not change what was built; it changed what the study was about. Without it, this would have been a clock. With it, it is a test the plan set for itself three months ago and had not run.proposed as standing practice · charter §5, v1.2

The rules, published

argue with the numbers, not the pictures

Constant	Value	Why
rows	12	the dial; a row is one hour on the left ruler and five minutes on the right
hour block	(weekday, hour%12) · 1×1 base	navy before noon, ink after; at 5×3 and 3×2 it spans the row, since those pairs have no week
minute hand	(⌊(m mod 5)·over/5⌋, ⌊m/5⌋) · 1×1 overlay · plane +1	the one signal on the page; its shadow is one microcolumn, like every plane +1 block
seconds pane	same lattice, s for m · signal glass at .8	ledger-approved over every fill on the face: lilac, mauve, midnight; pitch (faint) over the afternoon hour
precision	5 ÷ over minutes	1 · 1⅔ · 2½ at the three pairs; stated in the readout, never rounded away
minute, hour	snick · 400 ms · snick curve	FLIP from the old cell to the new; a hand that moves once a minute is allowed a journey — except on first paint, tab return or edition change, when it snaps
seconds	tick · 0 ms	a hand that moves once a second jumps — quartz, not sweep; sixty journeys a minute would be the cute trap
hidden tab	stop; snap on return	a clock is never wrong on purpose — unlike pulses, it does not resume where it left off
reduced motion	no snick; seconds pane moves by row only	twelve steps a minute instead of sixty; the time still reads, only the motion is withheld
demo speeds	×60 · ×720	verification aids; the readout says "demo" so a fast clock is never mistaken for a true one

Findings

The feed/score split is the engine lesson. Six custom properties on the stage and every placement in CSS meant the editions cost three lines of media query, and the same script would drive any other instrument. Recommended upstream: live-data blocks should write numbers to -- properties and let the score place them; a widget that positions its own elements is smuggling layout into a script.
Two rulers on one set of rows is the interference pair made legible. The left ruler is the base grid's reading of a row (an hour), the right ruler the overlay's (five minutes). The Interference Lab needed a paragraph to say what one look at the face should say; whether it does is Dan's read, not mine.
Precision is an edition property. 60 = 5 × 12 is a coincidence the 7×5 pair enjoys and the others don't. Declaring "1⅔ minutes" honestly turned out to read better than a fake minute would; the small editions feel like smaller instruments, not broken ones.
Ticks and snicks want different curves. Decided before drawing, not after: the seconds pane jumps and only the minute and hour travel on the snick curve. Sixty journeys a minute would be exactly the seasoning the plan warned about, and the minute's 400 ms move reads as an event because the seconds do not move that way.
The ledger earned its keep on day one. A seconds pane in sky glass would have been a ghost of mint over today's butter column; a navy pane would have been a ghost of dusk. Signal glass is the only pane approved over every fill the face uses, so the study's colour was decided by a lookup, not a taste.
The cute trap is a question of scale, not of subject. A clock that can be counted demonstrates the system; a clock that can only be recognised decorates it. The thumbnail version belongs in the queue's easter-egg drawer only if AXI-011 Manifests finds a way to nest a countable lattice inside a block.
The draw works. One seeded pull from the archive gave the study its question. Proposed for the charter's cadence: five sentences per session, at least one answered, effect recorded on the study page. It is a bias-with-a-random-input, which is roughly what was asked for.