Reproducible Evidence
Observed Benchmark & Demo Metrics
All measurements below come directly from running cortex-mcp benchmark and cortex-mcp demo locally on the current release.
cortex-mcp benchmark (500 Facts · 100 Episodes · 60 Entities)
| Operation | Sample Size | P50 Latency | P95 Latency |
|---|---|---|---|
record_episode | 100 episodes | 0.004 ms | 0.008 ms |
upsert_fact (with supersession) | 500 facts | 0.004 ms | 0.006 ms |
traverse_subgraph (2-hop) | 60 entities | 0.158 ms | 0.227 ms |
compile_context (800 tok budget) | 50 queries | 5.750 ms | 7.410 ms |
verify_integrity (600 SHA-256 + Merkle) | 1 full pass | 1.756 ms | 1.756 ms |
Active Facts: 350 · Superseded Facts: 150 · Stale-Fact Suppression: 100.0% · Token-Budget Compliance: 100.0%
cortex-mcp demo Bi-Temporal Supersession Output
Across 3 entities, 6 active facts, 2 superseded stale facts (JWT_HS256 superseded by PASETO_v4_public; HMAC_SHA256_SHARED_SECRET superseded by Ed25519_Asymmetric_Keypair), and 4 episodes:
- Compiled token usage: 307 / 320 tokens (95.9%)
- Suppressed stale facts: 2 excluded (0% leak)
cortex-mcp benchmark --json
cortex-mcp demo --json
cortex-mcp audit-contradictions --auto-resolve