The token file's own contrast claims are decided by nothing, and both clear by two hundredths

[iderex]

The design token file makes two contrast claims about its own values, in the role
sentence of ink-3, and nothing in this tree decides either of them:

python -c "import json;print(json.load(open('data/design-tokens.json'))['surface']['tokens']['ink-3']['role'])"
Tertiary text: labels, units, the year beside a title. It is a third step of the same scale and not a fainter one: it clears the same 4.5:1 floor as ink and ink-2 against every surface, and the distance from ink-2 is under one and a quarter of contrast ratio in the widest case. Reach for it where the words are ones a reader passes over, never to make text recede, because it will not.

git grep -l -i 'contrast' -- . ':!LICENSE'
data/design-tokens.json
decisions/0012-the-browser-in-the-gate.md

Both run 2026-08-27 against 2a1e214. The second command returns the file that
makes the claim and a record that mentions the word, and no source file at all.

Both claims are true today, and that is the reason to act rather than not to

Computed from the declared sRGB values, with the script and the full output on
#36:

dark  ink-3  min 4.53
light ink-3  min 4.52
widest ink-2 to ink-3 distance  1.23

Run 2026-08-27 against 2a1e214. The floor the file states is 4.5 and the bound
it states is one and a quarter, so the tightest text ratio clears by 0.02 and the
widest distance clears by 0.02.

Six hex values decide those margins. A one-step edit to any of them makes the
file assert something that is false, every leg of the gate stays green, and the
page that renders the file goes on printing the sentence. The claim is in the one
file this repository declares as the authority for those values, which is what
makes an unchecked claim there worse than the same sentence in a document.

What has to be decided before it can be built

The floor is a number inside an English sentence. A route cannot read one out of
prose without guessing, so the number a check refuses against has to become a
declared field beside the values it governs, and the sentence then cites the
field rather than restating it. That is part of this issue rather than a
prerequisite for it, and it is the thing to get right first: a floor typed into
Go and a floor written in the file are two copies of one value, which is the
defect this repository counts as restated-not-referenced.

The focus accent is deliberately outside this. Its ratios are on #36, the file
declares no floor for a ring, and picking one is a call rather than a reading.
This issue is the text scale, where the number is already stated and only the
refusal is missing.

The means

Go, as a row in internal/invariant beside the rows that already read this file.
A contrast ratio is arithmetic over integers the standard library parses, the
package already carries the row shape, the fixture convention and a suite that
proves a row bites, and adding one needs no program the toolchain does not have.
That is the same reason the rest of the gate is where it is, re-asked for this
artefact rather than carried over.

Done when

data/design-tokens.json declares the text contrast floor and the step distance
as fields rather than as sentences, the role sentence cites them instead of
restating them, a row of the invariants verb reads both fields and refuses a text
token whose ratio against any surface in either scheme falls under the floor or a
step distance over the bound, each refusal names the token, the surface and the
scheme, and the row is shown to bite by a one-step edit to a hex value that makes
the file's own claim false and reds the run.

[iderex]

The first two conditions ask this board to author a file it vendors

The Done-when opens with data/design-tokens.json declaring the floor and the
step distance as fields and the role sentence citing them. That file is a pinned
copy of one published in another repository, and this repository's own record
says so in the clause that matters here:

sed -n '27,30p' decisions/0007-where-the-design-tokens-live.md
The machine-readable token file lives with the machine-readable data, in the
repository that holds the roster. This repository consumes it the way it
consumes the roster: a pinned copy in the tree, rendered by the build, compared
on a schedule against the published one. Nothing here defines a token value.

The copy agrees with the published file exactly today, so two fields added here
would not read as a declaration. They would read as a copy that has drifted:

go run . tokens
design tokens: 206 value(s) pinned in data/design-tokens.json
206 value(s) pinned, 206 published, 0 differing.

and that is how every landing against this file has been treated so far. All
three are re-pins or notes about the copy, and none of them adds a value:

git log --oneline -- data/design-tokens.json
cf589b2 Re-pin the design token copy to the published values [#198]
96fbd9c Bring the vendored token note level with the published file [#175]
7b4df72 Vendor the design tokens and compare the copy against what is published [#66] (#145)

All run 2026-08-29 at 951bcb1.

What that leaves open, and it is a call rather than a repair

Two answers are available and this issue assumes neither. The fields are
declared in the published token file, where the values they govern are authored,
and arrive here with the next re-pin - which makes this issue wait on a change
in Flowfin/hub, and no issue there holds that work:

gh issue list --repo Flowfin/hub --state open --limit 300 --json number,title --jq '.[]|"#\(.number) \(.title)"'
#154 Publish builds the catalogue and drops it on the runner floor

Run 2026-08-29. Or the floor is a rule this site holds about a file it renders
rather than a value the file states, in which case it is declared somewhere this
repository authors and the sentence in the copy goes on restating it, which is
the defect this issue names in its own words.

Neither is decided here, and the reason the row cannot be built in the meantime
is the one this issue already gives: a floor typed into Go beside a floor written
in the file is two copies of one value.

What did land, and what it does not do

The gate said nothing about any of this. It printed one owed row, and a reader
takes that line for the whole of what this gate is meant to carry. It now prints
the gap with what would end it, which is #215, merged as 951bcb1:

go run . invariants | tail -1
39 rule(s) decided, 2 owed and not decided.

Run 2026-08-29 at 951bcb1. That is a disclosure and not the row. Nothing about
the Done-when is met by it: no route decides either claim, and both still clear
by two hundredths, re-taken today with the script from #36 cut to the text scale:

python contrast.py
ink scale against every surface, floor 4.5 declared in the token file
  dark  ink    min 12.73  15.85 14.38 12.73
  dark  ink-2  min 5.52  6.87 6.23 5.52
  dark  ink-3  min 4.53  5.64 5.11 4.53
  light ink    min 16.93  18.47 19.27 16.93
  light ink-2  min 5.23  5.70 5.95 5.23
  light ink-3  min 4.52  4.92 5.14 4.52
widest ink-2 to ink-3 distance, under 1.25 declared in the token file
  1.23

Run 2026-08-29 at 951bcb1. Six hex values still decide both margins, and a
one-step edit to any of them still leaves every leg green.

Adding blocked-on-a-decision, whose description on this board is that no change
to this tree moves the issue. That is the state above: which of the two answers
is taken decides which tree the fields are written in, and until it is taken the
row has nothing to read on either.