omp-fabric 1.4.1 → 1.4.3

package/docs/compaction.md

@@ -34,7 +34,7 @@ The selected-session reconciler stores a source generation/checkpoint and resume
 
 ## Summary maintenance
 
-The dedicated model is selected from `compaction.summaryModel` through OMP's public model registry and session-aware resolver. The active session model is the fallback when the dedicated route is unavailable. Leaf and condensed prompts XML-fence transcript evidence and treat it as untrusted data.
+The dedicated model is selected from `compaction.summaryModel` through OMP's public model registry and session-aware resolver. Pick it in `/fabric settings` under Compaction > Summary model, which lists the models OMP has available and offers Inherit to fall back to the active session model. The active session model is the fallback when the dedicated route is unavailable. Leaf and condensed prompts XML-fence transcript evidence and treat it as untrusted data.
 
 Maintenance defaults are bounded to 32 model calls, 2,000,000 input tokens, and 128,000 output tokens per project per UTC day, with a nested 16-call session limit. One project model job runs at a time. Jobs use 30-second leases renewed every 10 seconds, 60-second call limits, exponential retry backoff capped at 15 minutes, and three attempts. A failed call does not publish a ready node or increment usage counters.
 
@@ -96,3 +96,11 @@ bun run check
 ```
 
 This command includes typecheck, a fresh `dist/` build, the lazy-graph assertion, the full test suite, and dead-code lint.
+
+## Performance gate
+
+`bun run benchmark:lcm` drives the fixed fixture from the LCM operational contract: 10 sessions of 1,000 raw entries at 1 KiB each, 20 compaction cycles through the built `session_before_compact` hook, a repeat pass that must reuse the ready frontier, and a second process appending to the same ledger. It fails when a row is lost or duplicated, when a node references an uncommitted source or child, when the hook issues any model call, when hook p95 exceeds 250 ms, when recall precision or exact expansion regresses, or when a restart loses rows or nodes. The run prints and writes a JSON report covering ingest throughput, hook latency percentiles for both the emergency and ready-frontier paths, maintenance backlog, recall and expansion latency, migration counts, database and WAL bytes, checkpoint outcomes, and lock errors observed by the competing process:
+
+```sh
+bun run benchmark:lcm -- --out lcm-report.json
+```

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omp-fabric",
-  "version": "1.4.1",
+  "version": "1.4.3",
   "description": "A programmable tool and agent runtime for Oh My Pi (OMP)",
   "type": "module",
   "main": "./dist/index.js",
@@ -36,6 +36,7 @@
     "benchmark:real-resume": "bun run build && node scripts/benchmark-real-resume.mjs",
     "benchmark:memory-heads": "bun run build && node scripts/benchmark-memory-heads.mjs",
     "benchmark:codemap": "bun scripts/benchmark-codemap.mjs",
+    "benchmark:lcm": "bun run build && bun scripts/benchmark-lcm.mjs",
     "lint:dead": "knip --no-gitignore",
     "check": "bun run typecheck && bun run build && bun run assert:lazy-graph && bun run test && bun run lint:dead",
     "assert:lazy-graph": "node scripts/assert-lazy-graph-host-free.mjs",