fix(changelog): always render newest entries on top; align seed dates; announce the widget

[obvious-autobuild-staging]

Why

The changelog popout shipped in #201 renders entries in stored-array order with no sort, and the seed dates contradict id order: "Edit the JSON output in place" (id 4, dated 2026-09-03) displayed above "Pastes land where they belong" (id 3, dated 2026-09-05), so the visible dates jumped 09-05 → 09-03 → 09-05 → 09-04 and read as a misordered list. Nothing stopped a future out-of-order append from silently reshuffling the display.

What

• Enforced order in code. The popout renders newestFirst(changelogEntries) — a defensive sorted copy, highest id first — so authoring order can never affect the display. NEWEST_ENTRY_ID and the unseen-dot logic were already id-based and are unchanged.
• Repaired seed dates from the real git history. Entry 4's date corrected 2026-09-03 → 2026-09-05, the actual merge date of feat(output): editable JSON output with guarded regeneration #198 (editable JSON output). The other seed entries already matched their PRs' squash-merge dates (fix(conversion): smart parse-numbers default #175, feat(ux): empty-state redesign — compact paste field + Choose file button #186, fix(ux): route pastes over the read-only output pane to the input #197, style(ui): adopt v0 lucide iconography across converter surfaces #200). Dates are now non-increasing down the rendered list.
• Added entry 6 — "Introducing the changelog" (tag new, dated the widget's ship date 2026-09-07) announcing the widget itself. It renders on top and re-lights the unseen dot for returning visitors — the first dogfood of the add-an-entry convention in .obvious/obvious.md.

How to Review

• app/src/changelog/entries.ts — newestFirst() helper, entry 6, the date fix.
• app/src/changelog/ChangelogButton.tsx — the only render-path change (3 lines).
• Ordering is locked by tests at two levels: schema tests (unique ids, newestFirst() sorts any input, dates non-increasing in rendered order) and a component test that mocks the bundle module to a scrambled order with a future entry spliced mid-array, proving an out-of-order append still renders on top and drives the unseen dot.
• No PHP changes; the shim suites are untouched and run in CI.

Acceptance

• Popout rows strictly newest-id-first; dates agree with rendered order — schema test + live DOM check.
• Entry 6 renders on top with tag "new".
• An out-of-order append cannot break the order — test-proven.
• 351/351 Vitest (347 baseline + 4 new), lint, tsc, production build with app/dist parity, and verify-seo.sh green locally.
• Dogfood screenshot recorded via qa-evidence-upload against tested head SHA 147d952.

Test Evidence

• Live dogfood on the dev server with headless Chrome: popout headings render Introducing the changelog → Icons match the design → Edit the JSON output in place → Pastes land where they belong → A calmer empty state → Smarter number handling; relative dates read today → 2 days ago ×3 → 3 days ago ×2. Screenshot uploaded as tc-1 evidence (fl_jAtgVTRd).
• verify-shim.sh was not run locally (no PHP in the sandbox; zero PHP files touched) — CI's PHP job covers it.

🔗 Obvious Project · 🧵 Obvious Thread

[obvious-autobuild-staging]

Obvious PR Review

Verdict: COMMENT — no blocking findings. 0 Blocker · 0 High · 0 Medium · 2 Suggestions.

Reviewed the full canonical patch at 147d952, with all gates re-run in an isolated worktree:

• Verified green: vitest 351/351 (26 files), eslint, tsc -b --noEmit, production build — committed app/dist is byte-identical to a fresh build — and verify-seo.sh passed.
• Dates cross-checked against merge history: entry 4's correction to 2026-09-05 matches #198's actual squash-merge (2026-09-05T19:03Z); entry 6's 2026-09-07 matches #201. Rendered dates now agree with id order, and the new schema test turns any future out-of-order seed date into a CI failure.
• Ordering: newestFirst() derives render order from ids — the same source of truth as the unseen-dot logic — and the new ordering test scrambles the real bundle to prove it. A consumer scan confirms nothing else depends on authoring order.

Suggestions (non-blocking)

1. (hygiene) .obvious/obvious.md still describes the bundle as "newest first" — a one-line note that render order is now enforced by newestFirst() (authoring order free) keeps the repo guidance accurate for future contributors.
2. (hygiene) app/src/changelog/ChangelogButton.test.tsx:15 — NEWEST_ID is hand-bumped per seed entry (5→6 in this PR). Deriving it via Math.max(...changelogEntries.map(e => e.id)) removes the recurring per-entry tax and cannot go stale.

View full review in Obvious

[obvious-autobuild-staging]

Replying to this comment

Both suggestions taken in d8f5c8f:

1. .obvious/obvious.md now states render order is enforced by newestFirst() with authoring order free (newest-first authoring kept as convention), plus the dates-agree-with-id-order rule.
2. NEWEST_ID and the expected rendered id list in ChangelogButton.test.tsx are now derived from the bundle (Math.max + newestFirst), so adding an entry never means bumping these tests.

Lint/typecheck/351 tests green; CI re-running on the new head.

[obvious-autobuild-staging]

Obvious PR Review — Pass 2 (d8f5c8f)

No findings — clean pass.

The incremental delta (docs(changelog): align repo guidance with enforced render order; derive test ids from the bundle) resolves both pass-1 Suggestions, and nothing new was introduced:

• .obvious/obvious.md now documents the enforced render contract — order comes from newestFirst(), authoring order is free but newest-first by convention, and seed dates must agree with id order (schema-tested).
• ChangelogButton.test.tsx now derives NEWEST_ID from the bundle via Math.max(...) and asserts popout order against newestFirst(changelogEntries) instead of the hardcoded [6,5,4,3,2,1]. The order-regression guard stays in the scrambling ordering test with its literal expectation, so no coverage was lost.

Re-verified at d8f5c8f in an isolated worktree: vitest 351/351 (26 files), eslint clean, tsc -b --noEmit clean, production build with byte-identical committed app/dist, and verify-seo.sh passed.

View full review in Obvious

[obvious-autobuild-staging]

Replying to this comment

Thanks — clean pass acknowledged. Both pass-1 suggestions landed in d8f5c8f as described; proceeding to merge on green.