VERDICT: REVISE
F1: RESOLVED — bolero-harness/bolero/prompts/system.md:33 explicitly makes execution-reporter the terminal’s sole writer and leaves the main call ownership only of invoking it.
F2: RESOLVED — bolero-harness/bolero/prompts/d3_executor.md:402-415 requires reporter re-issue after gate failure and forbids the main context from editing the terminal.
F3: RESOLVED — bolero-harness/bolero/stages/d3_executor/.bolero/skills/execute-research-plan/SKILL.md:27-32 now runs only defect-free actions that do not depend on the defective action.
F4: PARTIAL — bolero-harness/bolero/prompts/d3_executor.md:471-475 now requires every failure’s action result.md, but line 242 still says preflight and undispatched failures “need no package.”
F5: PARTIAL — bolero-harness/bolero/stages/d3_executor/.bolero/agents/execution-reporter.md:41-49 permits result-cited small evidence, but execution_runner.py:228-269 exposes columns and bytes that the tabular guard does not bound and supplies no authoritative marker preventing a ≤200-row row-level table from qualifying.
F6: RESOLVED — bolero-harness/bolero/stages/d3_planner/.bolero/agents/feasibility-prober.md:31-38 uses Python -E -s -B and Rscript --vanilla, matching execution_runner.py:389-396.
F7: RESOLVED — bolero-harness/bolero/prompts/d3_executor.md:49-58 explicitly restores comparison-direction validation and condition-versus-rationale grain inventory without choosing or repairing a construction.
F8: RESOLVED — bolero-harness/bolero/prompts/system.md:134-136 now states that the Planner writes three advice files.
NEW:
- [MAJOR] bolero-harness/bolero/prompts/d3_executor.md:404,413-415; bolero-harness/bolero/stages/d3_executor/.bolero/skills/execute-research-plan/SKILL.md:179-181 — “exactly once”/“called once” contradicts the required reporter re-issue after a gate failure, leaving incompatible recovery instructions — specify one initial call plus at most one reporter recovery re-issue.