_{Comment @coderabbitai help to get the list of available commands.}

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.00%. Comparing base (18cce45) to head (a17d096).

@@           Coverage Diff           @@
##             main     #863   +/-   ##
=======================================
  Coverage   92.00%   92.00%           
=======================================
  Files          95       95           
  Lines       28809    28809           
  Branches      332      332           
=======================================
  Hits        26506    26506           
  Misses       2295     2295           
  Partials        8        8           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

Adversarial review complete (M-tier, two independent skeptics): CLEAN, zero findings.

Skeptic 1 (gate teeth + coverage conservation): the 176 = 165 + 11(+1 re-annotated) accounting proven at content level, not counts; all eight mutation classes fail loudly (stale, removed, malformed incl. the skip-prase typo path — the exact silent-rot class #606 represented — dangling, duplicate, unsupported-stage, non-vera target); regenerated site assets byte-identical; the original #606 bulk-shift repro (two edits at different positions, +10 line shift) is now a non-event with identical gate output and nothing to renumber.

Skeptic 2 (blast radius): each of the 12 formerly-suppressed blocks individually re-run through the real gate runners — the 11 removals genuinely pass their stages now (including the complete, correct fn add block that the #606 mis-anchoring class had wrongly suppressed in shipped main), and the one re-annotation is load-bearing (differential: removing it fails the gate with the exact documented error); all 165 placements mechanically verified outside fences; zero vera:skip leaks into docs/; zero surviving instructions referencing the deleted script in CI, CLAUDE.md, or CONTRIBUTING.md; full pre-commit green with the hook gone.

@coderabbitai All six round-1 findings addressed in 234ba09 — the two inline threads have direct replies; the four outside-diff findings (no threads) are resolved as follows:

spec/07-effects.md §7.5 (safe_parse taught parse_int as an Exn<String> thrower) — rewritten: parse_int is the pure Result<Int, String> built-in, so the example now wraps it in a parse_or_throw helper that converts Err into throw, with the effect Exn<E> declaration included so the block is self-contained. The rewritten block passes the full parse+check+verify pipeline (4 Tier-1 obligations), so its INCOMPLETE skip-check annotation is removed — spec check-stage annotations drop 7 → 6, exactly the shrink the stale-detection design is for.

spec/07-effects.md §7.6/§7.6.1 and spec/05-functions.md §5.9.1 (direct @Fn.0(...) application) — option_map and with_logging now apply the stored function value with apply_fn(...) per the §11.10.5 invocation rule (checker-typed since #854), with a cross-reference added in the prose. The blocks remain FRAGMENT-annotated for their inline fn-type parameters — the gate re-parses annotated blocks, and its stale detection stayed quiet, confirming the annotations are still honest.

spec/09-standard-library.md (mis-anchored suppression reason) — the annotation on the array_sort_by signature fence carried a string_chars/lines/words description, a #606-era allowlist mis-anchoring carried through the migration verbatim; the fence genuinely needs its suppression (bodyless signature), so the reason now names array_sort_by. The real string-utilities fence further down already had its own correct annotation.

Also in 234ba09: the unterminated-fence/<pre> hard failures and the run_parse_only_gate extraction from the inline threads, a CHANGELOG ### Fixed entry, and refreshed TESTING.md pipeline counts (type-check 86 pass / 6 annotated, verify 85 / 1). Gates: full pytest 5812 passed, six doc gates green, doc counts consistent, site assets byte-identical.

#!/bin/bash
git log --oneline -5
git show --stat 234ba09 2>/dev/null | head -50
echo "---parse_int check---"
git show 234ba09 -- spec/07-effects.md 2>/dev/null | head -150
echo "---stdlib annotation---"
git show 234ba09 -- spec/09-standard-library.md 2>/dev/null | head -80
echo "---changelog---"
git show 234ba09 -- CHANGELOG.md 2>/dev/null

@aallan Thanks for the detailed writeup. I attempted to pull the branch to independently re-verify commit 234ba09 (checking the parse_int/Exn<String> rewrite, the apply_fn conversions, the array_sort_by annotation fix, and the stale-count shrink), but the sandbox clone failed with a repository authentication error, so I can't re-run the gates myself in this session.

Based on your description, the fixes look sound from a correctness standpoint:

• Making parse_int a pure Result<Int, String> builtin and pushing the throw conversion into an explicit parse_or_throw wrapper is the right way to keep the effect signature honest — an Exn<String>-throwing parse_int would have been a real type-safety lie in the spec. Good that removing the now-unnecessary INCOMPLETE skip-check annotation was verified as a genuine 7→6 shrink rather than just deleted.
• Routing option_map/with_logging through apply_fn(...) instead of @Fn.0(...) direct field access matches the checker's typing rule from #854 — direct field application on a closure record would have been unsound outside the invocation path the checker actually type-checks.
• Renaming the array_sort_by annotation reason (rather than dropping the suppression) is correct, since it's genuinely a bodyless signature fragment and needed the skip; fixing the copy-pasted string_chars/lines/words reason text closes the mis-anchoring smell without weakening the gate.

If you'd like, I can re-run a fresh full review pass once I have working repo access, to independently confirm the stale-annotation counts and byte-identical site assets rather than taking the summary at face value.

@coderabbitai review

@coderabbitai review

@coderabbitai review

@coderabbitai review