TLDR
soak-100 exists in the QA catalog and the temporary confidence workflow can run it manually, but the latest beta.5 confidence proof did not run it because run_soak=false.
Latest classification from #80936:
{
"lane": "soak-100",
"status": "blocked",
"verdict": "environment-blocked",
"details": "Scheduled/Testbox soak runner did not upload artifacts for this proof bundle."
}Priority if OpenClaw moved fully to Codex today: P3 proof/completeness risk for optional long-run coverage; P4 product impact. This is not a default maintainer gate blocker, but it remains unfinished relative to the scheduled/Testbox proof request.
Current Coverage
The suite exists:
extensions/qa-lab/src/runtime-suite.ts maps soak-100 -> runtime-soak-100-turn
qa/scenarios/runtime/soak-100-turn.md defines the 100-turn same-session flow
The temporary fork proof workflow also exposes a manual dispatch input:
workflow: qa-runtime-confidence-proof.yml
input: run_soak=true
Latest full proof without soak:
OpenClaw baseline: v2026.5.10-beta.5
PR head: 3336dec6419c9cc9a87dc7cfa6f48118ca2d838e
Run: https://github.com/electricsheephq/openclaw-local-test/actions/runs/25719383976
Confidence tracker: #80936
Manual Command
OPENCLAW_BUILD_PRIVATE_QA=1 OPENCLAW_ENABLE_PRIVATE_QA_CLI=1 OPENCLAW_QA_TRANSPORT_READY_TIMEOUT_MS=180000 \
pnpm openclaw qa suite \
--provider-mode mock-openai \
--runtime-suite soak-100 \
--concurrency 1 \
--model openai/gpt-5.5 \
--alt-model openai/gpt-5.5 \
--runtime-pair pi,codex \
--codex-tool-loading direct \
--output-dir .artifacts/qa-e2e/soak-100
Acceptance Criteria For Closing
- Add or designate a scheduled/manual/Testbox job that runs
qa suite --runtime-pair pi,codex --runtime-suite soak-100 in the intended provider mode.
- Capture and link a real
qa-suite-summary.json and report artifact.
- Keep it out of the default maintainer gate unless maintainers explicitly choose to pay the 100-turn cost.
Links
TLDR
soak-100exists in the QA catalog and the temporary confidence workflow can run it manually, but the latest beta.5 confidence proof did not run it becauserun_soak=false.Latest classification from #80936:
{ "lane": "soak-100", "status": "blocked", "verdict": "environment-blocked", "details": "Scheduled/Testbox soak runner did not upload artifacts for this proof bundle." }Priority if OpenClaw moved fully to Codex today: P3 proof/completeness risk for optional long-run coverage; P4 product impact. This is not a default maintainer gate blocker, but it remains unfinished relative to the scheduled/Testbox proof request.
Current Coverage
The suite exists:
The temporary fork proof workflow also exposes a manual dispatch input:
Latest full proof without soak:
Manual Command
Acceptance Criteria For Closing
qa suite --runtime-pair pi,codex --runtime-suite soak-100in the intended provider mode.qa-suite-summary.jsonand report artifact.Links