perf: throttle trusted mute pruning

Co-authored-by: PR0M3TH3AN <[email protected]>

docs/agents/task-logs/weekly/2026-02-21_13-11-29_perf-deepdive-agent_completed.md

@@ -0,0 +1,28 @@
+# Weekly Perf Deep Dive Agent - Completed
+
+**Agent:** perf-deepdive-agent
+**Date:** 2026-02-21
+**Cadence:** Weekly
+
+## Summary
+Executed a performance deep dive focusing on the Moderation Service.
+
+### Tasks
+1.  **Exploration:** Identified `moderationStage` and `moderationService` as potential hotspots due to repeated lookups in the feed loop.
+2.  **Benchmarking:** Created `benchmarks/moderation_stage_bench.js` and `benchmarks/moderation_service_bench.js`.
+    - Baseline for service lookups: ~39ms for 10k reads.
+3.  **Optimization:** Identified that `getActiveTrustedMutersForAuthor` was performing an O(N) prune operation on every read.
+    - Implemented "Lazy Pruning" in `js/services/moderationService.js`, throttling the prune to run at most once per minute per author.
+4.  **Verification:**
+    - Re-ran benchmarks: Latency reduced to ~22ms (~43% improvement).
+    - Verified logic with `tests/moderation/*.test.mjs`.
+    - Confirmed file size compliance (after minor adjustment).
+
+## Artifacts
+- `weekly-perf-report-2026-02-21.md`
+- `benchmarks/moderation_stage_bench.js`
+- `benchmarks/moderation_service_bench.js`
+- `perf/hits-2026-02-21.json`
+
+## Next Steps
+- Consider further optimizing `moderationStage` by batching author lookups.