The canon has a way of growing: a stratum is briefed, built with its falsifier stated first, measured, reviewed by someone who reruns the harness, and either shipped with its row in the ledger or kept as a brief. Seven strata went through that in the seed-identity round. These are the next ones, in the order they earn their place, with what each would have to prove.
Briefed and waiting
Text-mode rendering
Every surface re-rendered as a grid of glyphs from the two faces: the field as density characters, the plates as half-block mosaics, the type as itself. It is the oldest 1-bit technique there is and the canon does not have it yet. What it must prove: that a text-mode frame is byte-reproducible from the seed; that the glyph ramp is measured on the actual faces (Departure's block glyphs have a specific weight curve) and not copied from a 1990s ramp; that it reads at 390px without horizontal scroll. What it must not do: introduce a third face for the blocks.
The sigil engine
Austin Osman Spare's method, done as code: a statement of intent, its letters deduplicated, the survivors fused into one mark on the lattice, then the mark forgotten. Under this canon the statement is the day's figure and the seed; the sigil is drawn by the lattice with the sigil colour, lines only, and never repeats across two dates. What it must prove: that the construction is deterministic; that no two dates in a year collide (365 sigils, hash them); that the mark contains a period, because every mark here does. The brief exists (brief-zesign-craft.md, §7) and the risk is that it becomes decoration; the falsifier is a reader who cannot tell the sigil from a glyph of the mark.
The steganographic layer
The seed, the date and the citation hidden in the least significant choices of the dither: which of two equal-density cells is on, which of two equal ranks the Bayer tie breaks toward. A plate saved from the page then carries its own provenance and can be checked. What it must prove: that the payload survives a JPEG re-encode at the qualities people actually screenshot at (measure, do not hope); that the visible plate is pixel-identical in density to the unmarked one; that extraction needs only MN_derive and the plate. The brief exists (brief-zesign-craft.md, §5). The falsifier: any measurable density shift.
Not yet briefed
Paper, actually shipped
The paper substrate is specified and its ratios computed. Nothing has ever been rendered on it. The next surface built should be built on paper first and void second, so the substrate stops being a promise. What it must prove: the ladder floors on #f2f0e9; the essence dot still reads as the one living thing against ink; the dither gate against a warm white.
A plate that leaves the screen. The lattice pitch becomes a physical size; a 1-bit image is the native language of a laser printer and a risograph. What it must prove: that a pitch chosen for 8px at 96dpi survives 600dpi without moiré; that the 12pt floor holds; that the citation prints with the plate, because a print with no date is a plate with no seed.
Instances
A namespace is defined as something that can be instantiated twice: HORCRX #001, HORCRX #002. No instance has been rendered under the canon. What it must prove: two instances under one namespace differ in seed only, and the five-token contract is enough to tell one namespace from another at a glance.
The figure, extended
The figure is played as sound on gesture. The same figure could set the rhythm of the beat (--beat modulated by the day's hexagram lines, broken lines longer), which would make the pulse date-specific too. What it must prove: the pulse stays sin⁴, only its period moves; reduced motion is unchanged; the modulation is audible in the frame timing and measured, not felt.
An unfamiliar agent
The least evidenced claim in the ledger is the purpose of the whole document: that an agent given only DESIGN.md and the skill produces conforming work. The trial is cheap: one fresh session, one product brief from a namespace, no other context, then review.mjs on the result. Until it runs, the canon is a description of work rather than a way of making it.
The rule for adding to this list
An idea joins this page with its falsifier written. An idea leaves this page when it has a directory under artifact/strata*/ with an evidence.md and a row in §16, or when it is dropped with the reason written here in its place. Nothing on this page is a promise.