correct: Poll the deadline inside structural traversals

[BenWestgate]

Before: correct(CorrectionContext(Profile.MS), "ms10tests" + "q" * 122) returned after 14.2 s, despite the documented ten-second deadline. Some alignment traversals on repeated-symbol input did minutes of work without yielding, so _search_target never reached its deadline check: one traversal on a 126-character all-q string ran 237 s.

After: the same call returns at 10.00 s. A search that stops at the deadline in its last layer now reports itself incomplete.

How: indel._views takes the deadline. It checks it before each edit or swap, including swaps of two equal symbols that are then skipped, and every 1024 deletion choices, the way _competitors._known_swaps already does. Once the deadline passes, every level returns without visiting its remaining siblings, so a traversal stops within a millisecond of the deadline. _search_target returns False when the deadline has passed by the end of its layers. One consequence is deliberate: a search that finishes just after the deadline is now reported as incomplete instead of complete, which errs on the safe side.

Swap and insertion traversals on ordinary input show no measurable slowdown. On all-repeated input, where most edits are skipped, the deletion loop is about 20% slower (3.6 s to 4.4 s for one pathological layer) and so is a swap layer (0.52 s to 0.62 s at 1,023 symbols).

Security review (focused): ACK.

• Completeness: an early stop can't be reported as a complete search. The clock only moves forward, so the next layer check, the best cutoff path, _search_many and _search_competitors all see the deadline as passed.
• No change below the deadline: with no deadline or a distant one, the traversal yields exactly the same 928,728 alignments as the parent commit.
• Polling gaps: the longest stretch without a poll is about 0.07 s, measured across every shape on 119–135 character inputs. Codex then found that equal-symbol swaps skipped the poll, so a 1,023-symbol all-q swap layer ran 0.5 s past an expired deadline; the poll now comes before that skip and it stops at once. That follow-up only moves the check earlier, so the completeness argument above covers it unchanged.

Tests: four deterministic clock tests cover both traversal branches, the skipped equal swaps and the incomplete result. All four fail on the current code. The full suite passes (911 tests).

Stacked on #130: on reviewability-v1 this change would push the package past the 5,200-line budget, and #130 brings it down to 4,959. With this change it is 4,971.

Fixes #136. Found by Codex on #94.

🤖 Generated with Claude Code

https://claude.ai/code/session_01T233rKgZqE5wzDm3EVTHL1

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-08T07:48:54.370624Z	56bba33	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 86a789d0f4

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".