OpenClaw RTT

Channel round-trip timing data for OpenClaw. This repo stores normalized QA results and publishes the dashboard below; the harness itself lives in openclaw/openclaw.

Each row measures how long a real channel takes to receive an OpenClaw agent reply after the test driver sends a controlled message. So yes: a Discord row is the agent-turn reply time observed back in Discord; Telegram, Slack, WhatsApp, and future channels measure the same loop in their own channel/scenario.

RTT covers the whole observed path, not just model time:

channel test driver -> OpenClaw channel transport -> gateway/agent turn -> outbound channel send -> reply observed by driver

That path can include channel API latency, polling/webhook timing, gateway routing, provider turn time, outbound send, and driver observation delay. p50 is the median successful sample; p95 is the tail sample. RSS appears when the importing workflow collected process resource metrics around the sampled command; older release rows stay blank until an RSS backfill run updates only the resource fields.

Treat cross-channel numbers as coverage and regression signal, not a pure transport ranking.…