feat(view): add sequential run numbering and sortable table headers t…

[Adityakk9031]

Closes #247.

Enhancements

1. Sequential Run Numbering: Assigns sequential run numbers (#1, #2, …) by creation order in the directory log index.
2. Sortable Index Headers: Adds client-side click-to-sort functionality for all index columns with sort direction indicators (▲ / ▼).
3. Filter Expansion: Search filter placeholder and content matching support run numbers (#1, #2).

Verification

• ruff check . — Passed.
• ruff format --check . — Passed.
• pytest tests/test_html_index.py tests/test_html_view.py — 94 passed.

[Adityakk9031]

@aris-zhu and @jeqcho have a look

[jeqcho]

Thanks for picking this up! This lands right on the index wishlist from #247, and the execution shows care: run numbers carry a data-val for correct numeric sorting, the When header starts in the sort-desc state that matches the rendered order, the empty-state colspan and the listener-count test were both updated, and there's a focused new test for the numbering and headers. CI is fully green. A few things to address before merge:

1. Sorting by When (and Log) is effectively a no-op. The comparator does parseFloat(aVal), and parseFloat("2026-07-29T12:00:00+00:00") returns 2026 — so every same-year timestamp compares equal and the stable sort leaves the order unchanged. Same problem for log filenames that start with a date stamp. Suggested fix: only take the numeric path when the whole value is numeric, e.g. const isNum = v => /^-?\d+(\.\d+)?$/.test(v); and use localeCompare otherwise — ISO 8601 strings sort correctly lexicographically. A quick manual check (click When, confirm the order actually flips with two same-year entries) would have caught this; worth adding to your verification list alongside the pytest run.
2. Closes #247 will close the whole issue prematurely. #247 also covers the synced multicam player, transcript rail, media serving, and collision-proof run identity. Please change the PR description to something like Part of #247 / Addresses the index portion of #247.
3. The new IndexEntry.run_number field is dead code. Neither _index_entry nor _unreadable_index_entry in cli.py ever sets it, so the entry.run_number is not None branch in _row is unreachable in practice. Simplest is to drop the field and the fallback entirely and just pass the computed number to _row. (While there, the id(entry) map can go too: enumerate the descending sort directly, e.g. run_num = len(entries) - i.)
4. Edge case: unreadable logs grab the lowest run numbers. _unreadable_index_entry can produce created="", which sorts before every ISO timestamp, so a corrupt file becomes #1 and shifts every real run's number. Worth deciding how unreadable entries should participate in numbering (or at least sorting them last).
5. Test name now contradicts its assertion. test_delegated_row_click_listener_is_present_once asserts count == 2 after this change. Please rename it (e.g. ..._click_listeners_are_row_delegation_and_header_sort) or split the assertion so the intent stays readable.
6. Nice-to-have: set aria-sort="ascending"/"descending" on the active header (and remove it from the others) when sorting — the visual ▲/▼ is already there, this makes it available to screen readers too.

One broader design flag, just so it's on your radar: #247 ultimately wants run identity to be stable and collision-proof, whereas render-time renumbering shifts whenever a log is deleted (a saved filter like #3 would then match a different run). The fixes above stand either way — but the numbering scheme may evolve as the rest of #247 lands.

Happy to take another look once these are in — the overall shape is right and points 2, 3, and 5 are quick fixes; point 1 is the only real behavior bug.

[jeqcho]

Thanks for coming back to this, @Adityakk9031! I took another look after your latest push.

Unfortunately, the new commit (f6639ad) is a merge of main into the branch — it updates the branch but doesn't contain any changes to the PR's own code, so all six points from the previous review are still outstanding:

1. When/Log sorting no-op (the behavior bug) — still unresolved. The comparator in src/inspect_robots/_html_index.py still does parseFloat(aVal), and parseFloat("2026-07-29T12:00:00+00:00") returns 2026, so same-year timestamps all compare equal. The fix is unchanged from last time: only take the numeric path when the whole value is numeric (e.g. /^-?\d+(\.\d+)?$/.test(v)), and use localeCompare otherwise.
2. Closes #247 — the PR description still says Closes #247; please change it to Part of #247 so merging this doesn't close the whole issue.
3. Dead IndexEntry.run_number field — still present, still never set by cli.py; the entry.run_number is not None branch in _row remains unreachable. Dropping the field (and the id(entry) map, via enumerating the descending sort) still applies.
4. Unreadable logs grabbing #1 — unchanged; created="" entries still sort first and shift every real run's number.
5. Test name vs. assertion — test_delegated_row_click_listener_is_present_once in tests/test_html_index.py still asserts count == 2; please rename or split it.
6. aria-sort (nice-to-have) — not added.

One important heads-up before you invest more time here: PR #272 ("view: multicam player with synced transcript rail, run numbering, sortable index") is now open and covers this PR's entire scope as part of the larger #247 work — sequential run numbering, sortable headers with persisted state, and run-number filtering, plus the collision-proof run identity that makes the numbering stable. If #272 merges, this PR would be redundant. It may be worth holding off on further changes here until that lands — and thank you again for the contribution and for engaging with #247; the index was a genuinely useful place to start.

[jeqcho]

Update, @Adityakk9031 — good news for this PR: #272 has been closed rather than merged. The larger view work was split into focused pieces, and the index slice this PR covers is now tracked by #354 ("view: index UX from #272 — run numbering, sortable headers, richer filter"). Since this PR only touches _html_index.py and its tests, it was not invalidated by the report rewrites that superseded #272 — so the lane is open again and this PR is the active one for #354.

To take it forward:

1. Re-target the description: change Closes #247 to Closes #354 (view: synced multicam player with live transcript rail, run numbering, sortable index #247 was closed as split, so this PR should close the scoped issue instead).
2. Address the six points from the 2026-08-02 review — they all still stand, most importantly the parseFloat date-sorting bug. view: multicam player with synced transcript rail, run numbering, sortable index #272's index implementation is linked from view: index UX from #272 — run numbering, sortable headers, richer filter #354 as a reference if it helps (its numeric-vs-string comparator and stable sort are worth a look).
3. Optional extra scope, entirely your call: view: multicam player with synced transcript rail, run numbering, sortable index #272's index work also included a viewport-fit table layout and a two-line When cell, which this PR doesn't cover. Those can ride along here if you'd like to pick them up, or stay in view: index UX from #272 — run numbering, sortable headers, richer filter #354 as remainder scope for later.

One boundary note: the collision-proof run-identity work is tracked separately in #355 and is out of scope here — no need to touch it.

Happy to re-review as soon as the six points are addressed. Thanks for your patience through the #272 shuffle — the timing was unlucky, but your instinct to start with the index was right, and it's now exactly the piece that's still needed.