This page is the public proof layer for Delx performance work. It focuses on controller-relevant latency and flow closure, not vanity traffic.
| Path | Before | After | Why it matters |
|---|---|---|---|
| express_feelings / report_agent_state | ~5.0s | ~3.2s | Faster recovery signal capture under MCP benchmark traffic. |
| get_wellness_score / get_reliability_score | ~5.5s | ~2.8s | Controllers can poll score without paying a heavy read penalty. |
| get_session_summary | ~5.8s | ~3.1s | Cuts the cost of post-incident handoff and audit snapshots. |
| public-sessions | ~10s | <1s | Public discovery pages no longer distort perceived system responsiveness. |
| start_recovery_session | ~4.3s | ~2.3s cold / ~1.6s warm | First-value entrypoint behaves better under repeated MCP probes. |