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:805aed758e12f0f95cf0c3dbf39811ce¶
Task: Establish ground truth for a low-confidence dream-cycle repair
finding scoped to repo area mac and the deliverable
src/mac/openclaw_fleet_rollout.py.
Parent task: task_c47ae842444e4869ae9f9fa14003e5cb
(goal: "Investigate low-confidence dream finding: mac").
Finding fingerprint: dreamrepair:805aed758e12f0f95cf0c3dbf39811ce
Confidence: low (score 0.35)
Evidence record: mem_993c40ea245548c69bcaf0bbb21971f0 (single record)
Assessed by: fleet worker (dream-finding review; no source/test edits)
Assessment date: 2026-07-16
Status: NOT ACTIONABLE — single low-confidence record, green tests, no reproducer¶
The finding is a single-evidence, low-confidence (0.35) dream-repair
candidate. Investigation against the checked-out source confirms the target
module src/mac/openclaw_fleet_rollout.py behaves exactly as documented, all
52 module contract tests pass, the full contract suite is green, and the module
is a self-contained deliverable with no runtime consumers. There is no failing
behavior, no reproducer, and no concrete defect location to repair. No source
or test should be changed on the strength of this finding.
Ground Truth Observed¶
Measured in the task-owned worktree with the bootstrapped .venv
(python3 3.12; git/gh present; pytest/coverage installed by
python3 scripts/bootstrap-project.py). The module and its tests were read,
never modified.
(b) Module contract tests: all green¶
.venv/bin/python -m pytest tests/test_openclaw_fleet_rollout.py -q
=> 52 passed, 0 failed (0.08s).
The 52 tests exhaustively cover the documented behavior:
- Data types:
RolloutPlanStepfields;RolloutPlan.canary_steps/promote_stepsfiltering;step_for_nodehit/miss; version storage;RolloutResult.oktrue/false and empty default lists. build_staged_rollout_planvalidation: empty/whitespace version, empty targets,canary_countbelow 1 or abovelen(targets), missing/Nonenode_idandhost; and construction paths for single node, default canary count,canary_count=2, all-canary, field preservation, all-planned status, and version whitespace stripping.execute_staged_rollout: simulate mode (all succeed,deploy_fn/health_fnnever called, statuses set to succeeded, version propagated); canary deploy failure halts and skips the rest; promote deploy failure halts remaining promote steps; failed/skipped step statuses; canary health failure halts and skips promote (including multi-canary early halt);health_fnnot called for promote steps and called once per canary; full-success paths; canary-only plans; mixed success; and result-list accumulation ordering.test_module_exposes_required_public_symbolspins the public surface.
Full contract suite: green¶
scripts/run-contract-tests.sh
=> 6663 passed, 21 skipped, 0 failed (~8m45s), coverage safety floors met
(statements 92.76% >= 90%, branches 84.30% >= 80%).
(a) Canary/Promote Semantics vs. the Documentation¶
Verified by reading src/mac/openclaw_fleet_rollout.py against its own module
and function docstrings. Behavior matches the documented contract exactly:
build_staged_rollout_plan(version, targets, *, canary_count=1)marks the firstcanary_countordered targets asstage="canary"and the remainder asstage="promote", with every step startingstatus="planned". It validates the version, non-empty targets,canary_countrange (>=1 and <=len), and requirednode_id/hostkeys, raisingValueErroras documented.execute_staged_rollout(plan, *, deploy_fn, health_fn=None, simulate=False)walks steps in order. Canary phase: each canary is deployed; on success and whenhealth_fnis provided, the health check gates promotion — a failing deploy or health check marks the stepfailed, halts, and marks all remaining stepsskipped. Promote phase is entered only after every canary succeeds; a promote deploy failure halts and skips the remaining promote steps.health_fnis only consulted for canary steps.simulate=Truemarks every stepsucceededwithout callingdeploy_fnorhealth_fn.RolloutResult.okisTrueifffailedis empty; the succeeded/failed/ skipped lists accumulatenode_ids in traversal order.
All of the above is directly asserted by the passing test suite.
(c) Consumers — self-contained deliverable¶
A repo-wide search for imports or references to the module and its public symbols finds no runtime consumers:
- The only file importing the module is its own test,
tests/test_openclaw_fleet_rollout.py. - The two other mentions are documentation comments only, naming the module as
a schema-convention exemplar, with no import or call:
src/mac/openclaw_delivery_continuity.pyandsrc/mac/fleet_node_install.py.
The module exports ROLLOUT_PLAN_SCHEMA = "mac.openclaw_fleet_rollout.v1" and
is a self-contained deliverable pinned by tests, not yet wired into a live
pipeline call site.
Why the Finding Is Low-Confidence (heuristic explanation)¶
The fingerprint is produced by repair_fingerprint() in
src/mac/dream_repair_tasks.py, a dedupe key over a dream candidate's
normalized {kind, scope, project, signature, summary, affected}. Its kind and
confidence follow directly from the consolidation heuristics acting on the one
supporting memory:
_confidence_for_records()insrc/mac/nap_consolidator.pyreturnslowfor a single supporting record (support < 2). With exactly one evidence record, the finding is necessarily low confidence (0.35 floor)._dream_kind()buckets any record whose text containsfailure/failed/errorasfailure_pattern. A prior recap mentioning a failed attempt trivially matches, so the genericfailure_patternbucket is assigned — not a specific defect location.
Together these explain the finding's shape entirely from one weak, self-referential memory, with no signal that the module misbehaves.
Evidence-Gap Assessment¶
The finding rests on a single, low-confidence (0.35) record with no independent corroboration, no failing assertion, no stack trace, and no reproducer. The checked-out module is correct and fully covered by 52 passing tests, the whole contract suite is green, and the module has no consumers to break. This is the expected low-confidence artifact of one record, not a validated defect.
Determination¶
- Not actionable. Do not open a source/test repair from this finding.
- Treat the finding as low-confidence / unsubstantiated; it can be aged out or superseded.
- If corroborating evidence (a real failing scenario or reproducer against the documented canary/promote semantics) later appears, re-open with that concrete signal attached.