@@ -0,0 +1,54 @@
1+# RTT regression audit checkpoint
2+3+Status: branch-local checkpoint, not release notes.
4+5+## Signals
6+7+- `openclaw-rtt` Discord main rows appeared to jump from about 5-7s p50 to about
8+ 24-27s p50 after the 2026-05-16 main window.
9+- Downloaded fast/slow Discord artifacts showed the old "fast" run included
10+ observed message `triggerTimestamp` and `timestamp`, while newer redacted runs
11+ kept only scenario metadata.
12+- `openclaw-rtt` `scripts/import-discord-rtt.mjs` falls back to whole summary
13+ duration when observed-message timestamps are missing. That made a redaction
14+ shape change look like transport RTT regression.
15+- Slack and WhatsApp RSS rows showed recurring first-sample max RSS outliers
16+ around 6-9GB while later warm samples sat far lower. That points at
17+ command-level cold-start RSS measurement before retained gateway heap.
18+19+## Fixes in this branch
20+21+- `extensions/qa-lab/src/live-transports/discord/discord-live.runtime.ts`
22+ preserves safe timing fields through metadata redaction so importers can keep
23+ measuring reply RTT without exposing Discord IDs or content.
24+- Discord QA scenario summaries now include `rttMs` for direct importer use.
25+- `extensions/qa-lab/src/suite.ts` records gateway-process RSS start/end/peak
26+ and checkpoint samples in `qa-suite-summary.json`, giving RTT importers a
27+ gateway-level metric separate from `/usr/bin/time` command max RSS.
28+29+## Proof so far
30+31+- Focused local wrapper:
32+`node scripts/run-vitest.mjs extensions/qa-lab/src/live-transports/discord/discord-live.runtime.test.ts`
33+ passed 30 tests.
34+- Focused local wrapper:
35+`node scripts/run-vitest.mjs extensions/qa-lab/src/suite.test.ts extensions/qa-lab/src/suite.summary-json.test.ts extensions/qa-lab/src/live-transports/discord/discord-live.runtime.test.ts`
36+ passed 50 tests.
37+- Testbox `tbx_01krvces8y0c99nzra2a90jg13` ran
38+`pnpm openclaw qa suite --scenario channel-chat-baseline` and emitted gateway
39+ RSS trace fields. Observed sample: wall `15784ms`, gateway RSS
40+`664403968 -> 689852416`, peak `689852416`.
41+- Testbox `tbx_01krwb9k7cbktytpjprxcydfbk` ran `pnpm check:changed` and the
42+ command exited 0. The wrapper Actions run `26008757251` still reported
43+`in_progress` after the box was stopped.
44+45+## Still weak
46+47+- No retained-heap regression has been proven. Heap snapshots should be taken
48+ only if the new gateway RSS samples grow across repeated warm samples.
49+- The branch fixes OpenClaw artifact quality. `openclaw-rtt` still needs an
50+ importer/report follow-up to prefer summary `rttMs` and optionally ingest
51+ gateway RSS fields.
52+- Gitcrawl data was stale for the newest RTT window, so live `gh` history was
53+ the source of truth for 2026-05-16 and 2026-05-17 PR attribution.
54+此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。