[takashi-matsuyama]

Summary

Fixes from two evaluations of basou board at board_version 3 (0.71.1) on boards outside basou. No shape changes: no board_version, record_version or method version moves, and no key is added to board.yaml or to a record.

• A refused key names what its kind takes. A key that no version of its kind takes, in a measure or an observation, is refused with the keys that kind takes in the version the file is read by: observe[0]: unknown keys 'count', 'capture' (kind 'http' takes: key, kind, url, words, count_pattern, capture_pattern, notes). Both evaluations had to find the keys of http, github_check, file_name and total by guessing from one refusal after another; one approved a declaration that read words as a count.
• basou board guide points to the reference for its own version (docs/spec/board.md at the version's tag), in step 2b (axis review) and in step 3 for a workspace with no board. Nothing basou printed named those keys.
• A string measure is printed and drawn without its unit (latest README.md, not latest README.md name). A number keeps its unit, and so does the count of a file_paths. unit stays required.
• Records are listed by their own recorded_at. The board page's list of records and the last record in basou board guide gave the time of the record's file name, which for a record written before 0.70.1 differs from its recorded_at (as the page's heading shows) by a second or so. The list reads recorded_at off the head of each regular file, only in the UTC form basou writes, and falls back to the name's time. basou board record now takes recorded_at after it measures the board again, not a few milliseconds before.

docs/spec/board.md and the Unreleased section of CHANGELOG.md say the same.

Verification

• On ae95c0c: pnpm install --frozen-lockfile, pnpm -r build, pnpm typecheck, pnpm lint, pnpm lint:lang, pnpm lint:paths, pnpm test and pnpm test:scripts all exit 0 (core 2921 passed and 1 skipped, sdk 67, cli 1566, scripts 21).
• Tests added or changed for each fix: the kind's keys at board_version 2 and 3 and with no version or a version that is not a number, none for a key outside a measure or an observation (including one named kind in a section or a lane); the reference line in both guide modes; string and number units in board measure and on the page; the list of records with pretty and compact JSON, a nested or misplaced recorded_at, non-UTC forms, a file that cannot be read and a link named as a record; recorded_at after measured_at.
• Mutation testing in a copy without .git: 16 mutations of the author's own (15 caught, 1 equivalent) and the survivors of the review's own 32 (all caught after the second commit, or equivalent).
• One round of convergent review (review-only): 5 low findings, all applied in the second commit (a guide sentence and a reference row that said too much, non-UTC times in the list, opening a name that is not a regular file, two tests that did not pin what they claimed).