AXI-003 · The blend ledger

Method

The engine draws two kinds of translucent structure: glass (an overlay-grid block at 80% opacity, in signal, sky or navy) and slivers (the near-miss columns, signal at 85%). Both use mix-blend-mode: multiply, and browsers multiply in gamma-encoded sRGB, per channel. With the pane's opacity folded in, the whole law is one line:

out = base × (1 − a + a × top) — per channel, 0…1; glass a = .8, sliver a = .85

Read it once more: a pane multiplies whatever is under it by a factor that depends only on the pane. Factors commute. So a stack of panes is order-free, a sliver crossing a glass is just two factors, and the entire closed palette is generated by four numbers. Everything below is derived from the live tokens in the stylesheet at load — change a color in snx.css and the ledger re-rules itself.

Two skins share the engine. On the client site the signal is cyan; here it is purple. The ledger computes both, because a ruling that only covers the studio's joke would be a poor gift upstream. Skin switches everything on this page.

The ruling, in numbers

What the ledger measures before it decides

Every threshold is published so it can be argued with. They were chosen by looking at the products on screen, then written down; the ledger applies them without exception, which is the point of a ledger.

Rule	Value	Why
law	out = base × (1 − a + a·top)	Multiply with opacity over an opaque backdrop, in sRGB. Verified against the canvas compositor on every product; verified by eye in Study A.
panes	glass .8 · sliver .85	The engine's actual opacities (snx.css v0.1.3). The ledger rules on what ships, not on full-strength multiply — though the hover shows it.
null	ΔE*ab(out, base) < 6	The pane leaves no mark on this field. Not harmful, just pointless: a composer who puts sky glass over navy has drawn nothing.
ghost	ΔE*ab(out, other token) < 6	The product reads as an existing color, slightly off — a misregistration, not a design. Rejected on sight because the eye will read it as a mistake.
identity	nearest token is the pane itself	Over white or a field close to it the pane simply becomes its own solid. Legal, unremarkable, and not a new color.
faint	ΔE*ab(out, base) < 8	Technically a mark, practically not. Named, so it can be used on purpose; flagged, so it is not used by accident.
type	ink ≥ 4.5 · chalk ≥ 4.5 · display ≥ 3	WCAG contrast for body and display sizes. Every product is cleared for ink type, chalk type, display only, or no type — and the composer reads that before placing a paragraph on a pane.
order	commutative	Factors multiply. The ::before sliver under the glass and the ::after sliver over it produce the same color, so the engine's asymmetry costs nothing.
source	live tokens	Read from the stylesheet at load, not copied into the script. The shipped `ledger.css` is checked against the live computation and the readout says whether they agree.

The extension

ledger.css — thirty-six tokens, named and ruled

The ruling serialised as custom properties, generated by build.js from the same module the page runs. A score that knows what lies under a pane may use the token as a solid fill instead of a blend: identical in every browser, no backdrop surprises on an open stage, one less compositing group. Panes are cut where the field beneath them changes — which is how a composer would draw them anyway.

loading ledger.css…

Findings

Dusk was already a product. Navy glass over butter or mint lands within ΔE 2–3 of dusk, the plane −1 recess color — the palette was closer to closed than anyone designed. But a pane that reproduces a plane −1 color impersonates a window, so the ledger calls it a ghost: the eye will read depth where there is none. Navy glass belongs over chalk, white, sky and paper, where it makes four genuine blues.
The spec's own example has an answer now. "Cyan over butter yields a specific green; we decide once whether we like that green." The client's cyan glass over butter is verdigris (#30BEBA), approved, ink type clear. Sky glass over butter, by contrast, is a ghost of mint (ΔE 5.1) — the one pale pane that must not cross the one pale field.
Purple is a worse blending color than cyan, and the joke has a cost. Every purple product over the paper tier lands at L* 50–59, the contrast trough: ink clears body type only over the four lightest fields, chalk never does, and over mint, sky and every sliver crossing the ruling is display-only. Cyan's light products all carry ink body type with room to spare. AXI keeps its house color, and keeps paragraphs off its panes.
Dark fields eat light panes. Sky glass over navy or ink, navy glass over ink, and on the client skin every cyan pane over navy or ink: null. Structure darks should be treated as opaque to translucent structure. Only purple survives over navy (midnight, #13128C), and barely over ink (pitch, faint).
Closure is an algebra, not a list. Because factors commute, the thirty-six doubles generate every triple and beyond, and drawing order is irrelevant. This also means a pane over a pane is safe by construction: glass on glass is just one more factor, already darker than either.
The arithmetic matches the browser. The seam test finds no seam at XL, M or S, and the canvas compositor agrees with the formula to within one level on every product. One caveat for the record: this holds for sRGB tokens blended in sRGB. If the client site ever adopts display-P3 tokens, the ledger must be recomputed in that space; the module's arithmetic is space-agnostic, its parser is not.
Blend where unknown, fill where known. Recommendation upstream: authored set pieces should render panes as solid ledger tokens cut at field boundaries; generated stages, where the tiling is only known at runtime, keep mix-blend-mode — or, better, the generator looks its own tiles up in the ledger and never draws a null or a ghost. That second half is a small change to the generator and a candidate for a utility run.
Graduation candidate, with the usual condition. The module and the stylesheet are AXI-x until the client site has a scored passage that uses the tokens in place of a blend. The ledger already covers the client skin, so that passage can be built without waiting on the studio.

The blend ledger

Study A · The assay

Study B · The ledger

Study C · Triples

What the ledger measures before it decides

ledger.css — thirty-six tokens, named and ruled

Findings