Historical field note
This record preserves prior investigation or planning evidence. It is not a current operating contract; use the numbered book and current runbooks for instructions.
Dream-Finding Assessment: dreamrepair:3dc2cf317ea21e032952a355c3550f88¶
Task: Establish ground truth for a low-confidence dream-cycle repair
finding scoped to the OpenClaw fleet-rollout deliverable for project mac.
Parent task: task_5287b80a718c432ca0e78807ccf45911
(goal: "Investigate low-confidence dream finding: mac").
Finding fingerprint: dreamrepair:3dc2cf317ea21e032952a355c3550f88
Finding kind / scope: failure_pattern, scope project, repo_area mac.
Confidence: low (score 0.35), backed by exactly one evidence record.
Origin task of the one evidence record: task_663c141c19494c76b0a0d28caa8fa07a
Evidence record: mem_a718733f11c040e98ae608ad93307df6
(record_type deployment_learning:mac)
Assessed by: fleet worker (dream-finding review; no source or test edits)
Assessment date: 2026-07-16
Status: DOES NOT REPRODUCE — historical learning record, not a live defect¶
The finding is a failure_pattern with project scope and low
confidence (score 0.35), backed by exactly one deployment_learning:mac
evidence record. Investigation confirms the referenced deliverable and its test
suite are intact, committed in the baseline, and fully green. The finding is a
second-order artifact of a prior implementation/repair learning record, not
evidence of a current defect in the checked-in openclaw_fleet_rollout module.
No source or test should be changed on the strength of this finding.
Ground Truth: Deliverable Is Intact¶
src/mac/openclaw_fleet_rollout.py:1exists (209 lines) and exposes every required public symbol:RolloutPlanStep—src/mac/openclaw_fleet_rollout.py:31RolloutPlan—src/mac/openclaw_fleet_rollout.py:41RolloutResult—src/mac/openclaw_fleet_rollout.py:65build_staged_rollout_plan—src/mac/openclaw_fleet_rollout.py:83execute_staged_rollout—src/mac/openclaw_fleet_rollout.py:139- Plan-schema identifier
ROLLOUT_PLAN_SCHEMA = "mac.openclaw_fleet_rollout.v1"—src/mac/openclaw_fleet_rollout.py:23 - Both files are committed in the baseline commit
a77007875519a49e9a089af0db8bc0589d1002b7("MAC OpenShell sandbox baseline");git logon each path shows no post-baseline modification. The working tree is clean.
Ground Truth: Test Suite Covers the Contract¶
tests/test_openclaw_fleet_rollout.py:1 (545 lines, 52 tests) covers all
areas the finding would implicate:
- Builder validation — empty/whitespace version, empty targets,
canary_count < 1,canary_count > len(targets), missing/Nonenode_id, missing/Nonehost(tests/test_openclaw_fleet_rollout.py:121). - Canary/promote staging — single-node canary, default two-node split,
canary_count=2, all-canary when count equals length, stage assignment and version stripping (tests/test_openclaw_fleet_rollout.py:175). - Executor simulate path — all-succeed with no
deploy_fn,deploy_fnandhealth_fnnever called, statuses set to succeeded, version propagated (tests/test_openclaw_fleet_rollout.py:224). - Executor deploy path — canary/promote deploy failure halts and skips the
remainder, failed/skipped statuses recorded
(
tests/test_openclaw_fleet_rollout.py:272). - Executor health path — canary health failure halts and skips promote,
health_fnonly called for canary steps, called once per canary step (tests/test_openclaw_fleet_rollout.py:313). - Module-contract regression guard — asserts the five required public
symbols exist and have the right kinds
(
tests/test_openclaw_fleet_rollout.py:522).
Test Results (After Bootstrap)¶
Environment bootstrapped with python3 scripts/bootstrap-project.py
(editable install of mac==0.1.0 succeeded).
- Module suite, direct pytest:
.venv/bin/python -m pytest tests/test_openclaw_fleet_rollout.py -q→52 passed in 0.08s(exit 0). - Module suite via canonical hermetic runner:
scripts/run-contract-tests.sh tests/test_openclaw_fleet_rollout.py -q→52 passed in 0.04s(exit 0).
No failing assertion, error, or regression reproduces. The failure_pattern
does not manifest against the checked-out code.
What the Single Evidence Record Represents¶
The lone backing record mem_a718733f11c040e98ae608ad93307df6 is a
deployment_learning:mac recap from the original implementation/repair task
task_663c141c19494c76b0a0d28caa8fa07a. It captures a historical build/repair
learning about the rollout deliverable — not a live, reproducible defect. It
contains no failing assertion, stack trace, current reproduction, or named
offending code path. It is therefore an original-implementation "failure/repair"
learning artifact, not a live defect signal.
Why Confidence Is Low (0.35)¶
The 0.35 score is the classifier's structural floor, assigned deterministically when a finding is backed by a single evidence record:
docs-level contract of the classifier states "low — Signal is present in the artifact text, but only a single evidence record backs it. Score ~= 0.35" (src/mac/dream_cycle_classifier.py:14).- The threshold table sets
"low": ("low", 0.35)(src/mac/dream_cycle_classifier.py:87). _confidence_for(...)returns the low threshold wheneverevidence_count < 2and the candidate is not high/medium (src/mac/dream_cycle_classifier.py:233), matching this finding's singledeployment_learning:macrecord exactly.
So low confidence here is a property of the evidence volume (one record), not a measure of severity or of a confirmed defect.
Evidence Gap¶
- Only one evidence record exists, and it is a historical learning recap rather than a reproduction. There is no independent, current observation of a rollout failure.
- No failing test, trace, or offending code path is attached to the finding.
- The classifier floor (single record -> 0.35) is the sole driver of the low score; nothing in the checked-out repository corroborates a live failure.
Conclusion¶
The finding dreamrepair:3dc2cf317ea21e032952a355c3550f88 does not correspond
to any real current defect in the repository. The openclaw_fleet_rollout
deliverable and its 52-test suite are intact, committed in the baseline, and
pass cleanly under both direct pytest and the canonical hermetic contract
runner. The single low-confidence evidence record is an original-implementation
learning artifact, and the 0.35 score is the classifier's structural
single-record floor. No source or test changes are warranted.