fix(sync): make the index delete check case-exact

[phernandez]

Part of #1627.

What was wrong

After a case-only rename with the watcher on, a ghost entity for the old lowercase path can appear (case/config.md next to the real case/Config.md). The ghost never went away. bm project index planned its delete, then LocalProjectIndexDeletePathVerifier.confirm_deleted_paths checked existence with (base_path / path).stat(). On a case-insensitive filesystem (APFS, NTFS) that stat succeeds for any case, so the verifier reported the path as present and the maintenance runner logged "Skipping planned index deletes for paths present in storage again" on every pass.

Fix

path_spelled_exactly_on_disk in src/basic_memory/index/local_project.py runs after a successful stat. It checks every path component against its parent directory's listing, which returns the stored spelling. Listings are cached per directory for the batch, so each parent is listed once. Names are compared in NFC because APFS is also normalization-insensitive and only case is in question here. Listing errors go through the verifier's existing OSError handling, so a failed probe still skips the delete instead of confirming it.

The old spelling now counts as absent, so the next index pass deletes the ghost. This is derived state converging through the existing index pass. No locks or extra machinery.

Not fixed here (follow-up)

The watcher race that creates the ghost in the first place (the stale lowercase path indexed as a new file after the move, and the brief -1 permalink rewrite of the file bytes) is not addressed. That needs separate work in the watcher's new-file handling. In the index-pass test below, the renamed file is created before the old row is deleted in the same pass, so it can still be assigned a -1 permalink; that is also part of the follow-up.

Tests

tests/index/test_local_project_index.py:

• path_spelled_exactly_on_disk with Case/Config.md on disk: the exact spelling is present; Case/config.md and case/Config.md are absent. On a case-sensitive filesystem those paths simply do not exist, which also reports absent. One listing per directory.
• NFC and NFD spellings of the same name both count as present.
• The verifier confirms case/config.md absent and keeps case/Config.md.
• An index pass after replacing case/config.md with case/Config.md (different content, so no move detection) leaves exactly one entity, case/Config.md.

On macOS (APFS) the two verifier/index tests fail without the fix and pass with it. On a case-sensitive filesystem they pass either way.

Verification

• just fast-check: passes.
• BASIC_MEMORY_TESTMON_SELECT_FLAGS="--import-mode=importlib" just fast-test tests/index tests/indexing test-int/test_project_index_delete_materialization_race.py: 793 passed, 3 skipped.

🤖 Generated with Claude Code

https://claude.ai/code/session_01APFUk2bjEwMptMQqpRhjea

[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 2026-10-09T02:37:34.137281Z	0971ee0	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: 72a06a4a89

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".