The manifest entry this table said did not exist

[ChelseaKR]

Part of #15. Stacked on #29; base is feat/a11y-in-a-real-browser.

What was wrong

The Standards Conformance table opened with:

Read one caveat first. The standards program's applicability manifest has no entry for this repository, so nothing has ever decided which of the fifteen standards bind it.

The entry was added on 2026-08-15, in STANDARDS commit 36747be, "fix(applicability): correct the manifest, and register the repo that owns it". Verified at that repo's HEAD, applicability.yml:1171:

  perimeter:
   archetype: civic-data-tool
   publication: cleared   # repo is PUBLIC on GitHub since 2026-08-08
   tier: B+C              # GitHub Pages site + local Python/Node build
   flags: { html: true, dockerfile: false, llm: false, bilingual: false, hosted: true }
   standards:
     ...
     AIEV: { na: "deterministic coverage counting over published CAL FIRE/FRAP datasets; no LLM/model component" }

The sentence was true when it was written and false the same week.

Why nothing noticed for twelve days

def test_the_table_says_the_applicability_manifest_has_no_entry_for_this_repo() -> None:
    """FIX-02. Remove this test on the day the manifest entry exists, not before."""
    assert "applicability manifest has no entry" in section

That test cannot fail while the sentence is present, and cannot notice when the sentence stops being true, because the only thing it reads is the sentence. It was a permanently green check over a false claim about this repository's own governance, in the section whose whole job is to state what is true today. It is the same shape as the four checks ADR-0004 was written about, in a place nobody thought to look.

What the table says now

It transcribes the entry, with the date it was read, and says plainly that the registry lives in another repository so nothing in CI can check the transcription against its source. That is the discipline docs/MARKERS.md already applies to CAL FIRE's documents: a source that cannot be re-read at build time is transcribed with a date, and the date is part of the claim.

One row moved to match the registry. Observability was Applies (Tier C); the registry says B+C, which is also what OBS section 0 asks a repository with two surfaces to state.

The B half is a genuine gap and is recorded as one rather than talked around. OBS section 8 wants a Core Web Vitals RUM beacon from a Tier B frontend, which means shipping a script that reports readers of a civic-data page back to somebody. These pages ship no script and this project takes no telemetry. That refusal is a position, not an oversight, and the standard has no N/A for it, so the row says so. The Lighthouse-CI lab gate on LCP, INP and CLS needs no beacon, is also not met, and is named as the half that could be built.

Proof each new check can fail

Against the pre-change README, with the new tests in place:

FAILED tests/test_standards_conformance.py::test_the_table_does_not_claim_the_manifest_has_no_entry
FAILED tests/test_standards_conformance.py::test_the_table_transcribes_the_manifest_entry[civic-data-tool]
FAILED tests/test_standards_conformance.py::test_the_table_transcribes_the_manifest_entry[`B+C`]
FAILED tests/test_standards_conformance.py::test_the_table_transcribes_the_manifest_entry[cleared]
FAILED tests/test_standards_conformance.py::test_the_table_transcribes_the_manifest_entry[added 2026-08-15]
FAILED tests/test_standards_conformance.py::test_the_table_transcribes_the_manifest_entry[Read 2026-08-27]
FAILED tests/test_standards_conformance.py::test_the_transcription_says_it_cannot_be_checked_from_here
7 failed, 50 passed in 0.06s

Positive controls, 31 of them in the same file, passing identically before and after:

$ uv run pytest tests/test_standards_conformance.py -q -k "no_row_is_blank or every_standard_has_a_row or pin_names_a_released_tag"
31 passed, 26 deselected in 0.02s

test_the_table_marks_exactly_the_one_standard_the_manifest_marks_not_applicable is also a control: it passed before this change too, because the table already marked exactly one. It is added because it is the one part of the transcription a test inside this repository can check, and it fails the day a second row is marked N/A without the registry saying so.

Gate output

$ make verify
...
Required test coverage of 90% reached. Total coverage: 100.00%
============================= 581 passed in 27.51s =============================
...
determinism: build/run-one and build/run-two are byte-identical (5 files)
MAKE_VERIFY_EXIT=0

Exit code read from $?.

What is still open on #15, and why this PR does not close it

• main has no ruleset and no branch protection. Verified today: gh api repos/ChelseaKR/perimeter/rulesets returns [], branches/main returns "protected": false, branches/main/protection 404s. The profile is committed at .github/rulesets/main.json and applying it is a live repository setting. A pull request cannot do it and this one does not pretend to.
• Dependabot alerts are disabled. gh api repos/ChelseaKR/perimeter/vulnerability-alerts returns "Vulnerability alerts are disabled". Also a repository setting, so SEC-15 still has nothing to read.
• zizmor and CodeQL now exist and run. Both were listed as missing in ci.yml calls its gates merge-blocking; main has no protection at all, and the repo was never registered with the standards program #15 and both are present. CodeQL passed on A field note is published output, and two of them were false #26 today, which is the first execution ADR-0004 recorded as its outstanding verification. Neither can be required until the ruleset is applied.

🤖 Generated with Claude Code