[ottonomy]

Closes #47

Stacked on #69 — review and merge that first (or merge the stack together).

Departures

None. Everything under "What to build" is in the commit: the field is gone from
CheckResult, every writer and reader moved to id, all tests that read
.check or wrote a check key were converted, and the README, the API guide,
the architecture doc and the CHANGELOG were updated.

Decisions not in the issue

• pickAppliesSkipReason now takes the suite's phase. The summary helper
  used to recognise the synthetic row by comparing check to
  "<suite>.applies". An id embeds the phase, so the comparison cannot be
  made from the suite id alone. I passed the phase that summarizeSuite
  already has into the helper and compared against
  computeId(phase, suiteId, "<suite>.applies"), rather than, say,
  suffix-matching the id. It is exact, and it reuses the one function that
  builds ids.
• id is written before suite in the JSON examples. The ticket says the
  rows show "id and suite"; putting the identifier first reads better and
  matches the rows the API guide already had.
• Two test files got a named constant instead of an inline literal. In
  test/openbadges/unknown-achievement-type-check.spec.ts the local check id
  was already a CHECK_ID constant, so I added RESULT_ID derived from it
  rather than repeating the long namespaced string. Everywhere else I inlined
  the id, matching the surrounding style.
• Formatting. Several converted lines grew past the repository's 80-column
  style, so I wrapped them the way Prettier would. Prettier itself is not one
  of the commands I was allowed to run, so this is by eye; eslint src test
  (which pnpm run test runs first) is clean.
• Noticed, left alone: docs/api/verification-results.md still says
  "Singular check is used when n === 1" — that is about the English word in
  a rollup message, not the removed field, so I did not touch it. The six
  pre-existing type errors are likewise untouched, as the ticket directs.

Agent-Run: 2ed717c0-eeaa-4367-a5fd-dd0646c4d406