[jack-arturo]

Prefix filters now normalize slash and colon separators consistently across graph, vector and postfilter paths. Legacy graph nodes use hierarchy-aware fallback rather than string-prefix matching. Exact matching retains separator distinctions. Documented backfill covers existing Qdrant payloads without tag_prefixes.

Closes #261.

Validation: initial regression suite had 8 failures on base; 42 focused tests pass. Full isolated Python suite: 683 passed, 1 existing skip; bridge: 43 passed, 5 optional live skips. Independent real FalkorDB/Qdrant replay passes 10 cases covering legacy/indexed nodes, any/all, separator equivalence, exact matching and hyphen negatives. Changed-file Black/isort/flake8 pass.

Babysit validation: 49 focused tag/recall tests pass; one local Codex documentation review was clean. The hosted documentation finding was clarified and resolved. Required CI passed on the documentation fix.

Required snapshot comparisons completed against main 3caa9d5: LoCoMo-mini 195/235 → 195/235; LoCoMo-full 1210/1542 → 1207/1542 (-0.1945pp). No judge/OpenAI similarity scorer; 69/444 category-5 questions excluded. Full has four lost/one gained QA rows. Existing unordered entity decomposition selected different supplementary queries between Python processes; with identical PYTHONHASHSEED=0 and shared stores, all five changed QA verdicts agree across three paired repeats (15/15), with identical query lists. Context ordering is not claimed identical. No ranking policy was changed. See benchmarks/EXPERIMENT_LOG.md and benchmarks/baselines/pr266-tag-prefix-20261021.json for refs, hashes, settings, counts, and the variability caveat.

Implementation harness: Codex desktop via AutoJack GitHub App. Existing original commit attribution preserved; babysit documentation/evidence commits use autohub-codex.

[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 2026-10-21T19:19:36.176444Z	6b51182	Draft marked ready

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: 6b5118226c

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".

[autohub-codex]

Codex desktop via AutoJack GitHub App: verified green at 0c3ae40ad8104e51dd59c7d7beb59b8e1280a8c6.

All six required CI checks pass on this head; PR is open, non-draft, and mergeable. The initial Codex review's documentation finding was fixed, replied to, and resolved. Smart auto-review did not fire for the final documentation/evidence head within its ten-minute grace; no current-head reviewing reaction remains. No manual review request was posted.

Required snapshot comparison is recorded in benchmarks/EXPERIMENT_LOG.md and benchmarks/baselines/pr266-tag-prefix-20261021.json: mini 195/235 → 195/235; full 1210/1542 → 1207/1542 (-0.1945pp), with 69/444 category-5 rows excluded and no judge/OpenAI similarity scorer. Existing hash-dependent entity decomposition selected different supplementary queries between server processes. All five changed QA verdicts agree across three paired shared-store repetitions with identical PYTHONHASHSEED=0 (15/15); original full scores and context-order variation are preserved honestly. Focused tag/recall checks: 49 passed.

Size 278 changed lines (start 147). Label babysit:ready; ready for human merge. No merge or auto-merge performed.