#443's body cites have drifted since it was filed, and "all five call sites" enumerates a different five

[pbean]

#443 was filed on 2026-08-03 and its file:line cites were taken against the tree of that day. Almost none of them still land on what they name, and one count contradicts the issue's own removal list a few lines below it. #443 is the design-of-record for shape (1) of the repo_root != project work, so whoever implements it will resolve these cites first — and a stale cite that lands on real code is the failure mode conventions.md calls out, because it survives a spot-check.

Resolved by symbol against main @ 1a3d358b.

"all five of its call sites" enumerates a different five

The body says landing #443 deletes bmadconfig.worktree_isolation_conflict "and all five of its call sites", then names them as _reject_isolation_conflict at cmd_run/cmd_sweep/_resume_paused_run, the RuntimeError in the sweep factory, and the cmd_validate finding.

The actual five callers are cli.py:197 (inside _reject_isolation_conflict itself), cli.py:369 (cmd_validate), cli.py:1081 (_warn_preflight_would_abort), cli.py:2185 (the sweep factory), and tui/app.py:222 (_guarded). The parenthetical counts _reject_isolation_conflict's three callers (cli.py:1843, :2244, :2414) as three sites and omits _warn_preflight_would_abort and the TUI — both of which get their own removal bullets immediately below. This is the one genuinely miscounted number in the body.

Stale cites

Body says	Actually at
cli.cmd_validate's verify.worktree_clean(project) — cli.py:268	cli.py:423 (def :316); cli.py:268 is an unrelated resolve_profiles/ProfileError docstring
merge-back and teardown — worktree_flow.py 389, 613, 730	merge_local :1852, gc_run_worktrees :2251, discard_worktree :2266
sweep.py:666	discard_worktree :847, worktree_clean :882 and :1426
isolation predicate — recovery_flow.py:172	:828
cli.py clean-tree gates — 1023, 1261	:1860, :2247
_reconcile_stale — 179	def :301, calls :1867 and :2254
--reverify — 1813, 1855	_reverify def :2927, calls :3026 and :3068
the confirm commit — 1984	verify.commit_paths(paths.repo_root, …) at cli.py:3211, in _land_confirmation (def :3161)
cmd_clean — 2385	def :3720
load_paths' config requirement — bmadconfig.py:72-74	:188-190; :72-74 is inside worktree_isolation_conflict's docstring
ProjectPaths.rebased — bmadconfig.py:43-63	:45-65
hook-relay bake — worktree_flow.py:142-143	relay = repo_root / HOOK_SCRIPT_REL at :818, in provision_worktree (def :721)
skill trees / _bmad/custom/ copies — worktree_flow.py:238-245, 262-266	_seed_bmad_tree :373, base_skills_seed_incomplete :489, _worktree_skill_copy_candidates :96
seed_files / seed_globs — worktree_flow.py:196-200	consumed at :848 and :920

Two content claims that also drifted

• "probed under project" is imprecise. The body says skill trees and _bmad/custom/ overrides are "probed under project" by missing_base_skills / resolve_review_layers. Those functions take whichever root the caller passes: worktree_flow.py:98 and :111 call resolve_review_layers(repo_root, tree), and base_skills_seed_incomplete(worktree, repo_root, trees) probes repo_root. Only the preflight is project-rooted. The current wording reads as though the functions are inherently project-rooted.
• "Coverage gap to close" was narrowed by PR fix(verify): run review-gate commands in repo_root; spare an empty-proof park #742. It says the refusal's tests "construct the divergent pair but stop at the refusal". tests/test_verify.py:2692 (test_verify_review_gates_run_commands_in_repo_root) and the rows after it now go past it, built with dataclasses.replace(project, repo_root=…) — the house idiom an implementer of shape (1) would reuse. The provisioning gap the section describes is still real; the sentence is just no longer the whole picture.

Suggested fix

Re-resolve every cite by symbol and correct the five-call-sites parenthetical. Related to #551, which is blocked behind #443 and quotes some of the same cites in its answer.