BUG-04-299 resolved, and the first live plan-to-bug engine exercise
This section carries two outcomes that cannot be separated: the bug’s resolution, and the FIRST live exercise of the engine’s plan-to-bug path. A standalone engine test cannot exercise the return/fold without a real child, and the child cannot own root closeout — splitting them would produce a section that structurally cannot prove its own deliverable.
Measured at authoring: BUG-04-299 is open, severity high, plan_dir: null, subsystem
codegen-llvm, 490 occurrences across ~48 functions, blocking dev-to-master CI.
Mechanism: EffectMismatch { field: "may_share", inferred: false, realized: true } —
inference uniformly MORE optimistic than realization, which reads as one root cause.
Engine facts: only the plan workflow declares replaces_skills: ["continue-roadmap"];
bug.json declares []. Per workflow-state-machines.md sec 1.7/9 the child returns
typed obligations and MUST NOT run /tpr-review or /impl-hygiene-review;
invocation_protocol.py:351-355 registers root-only closeout obligations.
Deliverables
- RE-MEASURE the bug against the post-section-03 tree BEFORE any fix work. Sections 01-03 may have cured the root cause.
- Branch on that measurement. STILL VALID: push a nested
/fix-bugchild. CURED BY 03: prove the original reproducer is gone and close through the sanctioned OBE route WITHOUT minting a fix plan, preserving attribution to section 03. Both branches require the section-01 frozen baseline comparison. - Structural engine preflight before the child push: confirm the root-owned closeout obligations and the child’s forbidden gates.
- If the engine misbehaves, SUSPEND the bug path and invoke
/improve-toolingagainstworkflow_engine; resume this same section only after the engine defect is verified fixed. An engine miss is a Critical workflow failure, never a workaround opportunity. - Fold the typed child evidence and obligations; the ROOT alone runs aggregate review and closeout.