PLAN_INDEX.json

{
  "version": 1,
  "plans": [
    {
      "id": "PLAN-20260418-LOCAL-LLM-FILTER",
      "title": "Local LLM Filtering Evaluation and Integration",
      "kind": "feature",
      "status": "in_progress",
      "priority": "P0",
      "owner": "Codex",
      "file_path": "docs/plans/active/PLAN-20260418-local-llm-filter.md",
      "confirmed_by_user": false,
      "created_at": "2026-04-18T14:15:40Z",
      "updated_at": "2026-04-19T11:28:00Z",
      "archived_at": null,
      "notes": [
        "The benchmark corpus is now at a 60-sample pause point. The shipped deep-filter path has moved from the earlier `priors-v2` baseline to the benchmark-backed `priors-v4 + post-processing` contract, the rerun under an aligned resolver contract keeps `qwen3.5:2b` as a valid but clearly weaker fallback than `qwen3.5:4b`, a desktop verification pass fixed a Rust-byte-offset versus JavaScript-string-index mismatch so the in-app reviewed-result preview now matches the benchmark-backed redaction behavior, and targeted frontend/Rust regression tests now cover review/apply logic plus provider/JSON failure handling."
      ]
    }
  ]
}