Skip to content

v0.11.5 — patch (Pre-Path-B: QL_FIELD_OWNERSHIP_STRICT; Path B UNBLOCKED)

Latest
Latest

Choose a tag to compare

View all tags
@andyzengmath andyzengmath released this 03 May 18:23
· 8 commits to master since this release
v0.11.5
9c1fc3a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

4-story patch closing the Pre-Path-B field-ownership escalation policy (operator-decision item from v0.11.4 backlog). Path B (real-feature dispatch via --coordinator) is now UNBLOCKED.

QL_FIELD_OWNERSHIP_STRICT=true (env var; default OFF):

  • When set, the v0.10.8 N48 field-ownership snapshot-diff WARN escalates to WAVE_FAILED.
  • Wire site: lib/iteration-loop.sh:317-321 (inside existing N48 WARN block).
  • Pattern-consistent with v0.11.1 N43 QL_PARALLEL_POLL opt-in design (same ${VAR:-false} gate, same == "true" compare).
  • Default OFF preserves backwards compat: Tests 9 + 10 unchanged (verified 32/32 pre-Test-11).
  • Recommended for Path B real-feature dispatch to provide hardened wave-level data-integrity guarantee.

Semantic note: strict mode operates at the wave-signal level — it forces WAVE_FAILED so downstream parent-agent signal parsing gets a clear failure indicator. Per-story aggregation under WAVE_FAILED branch still derives status from review.* fields (cf. Test 2 partial-pass semantics), so a violating coordinator that wrote review.* fields will have stories aggregating to passed despite the wave-level FAIL. This is intentional: the strict mode flags the wave as untrusted while preserving potentially-valid review writes for triage. Documented in CLAUDE.md "Coordinator-related" section.

Test 11 (NEW; 3 sub-asserts): validates strict-mode escalation. Reuses existing field_ownership_violation stub mode + sets QL_FIELD_OWNERSHIP_STRICT=true. Asserts: WARN appears + FAIL log appears + Wave FAILED at wave-signal level. (PRD planned 5 sub-asserts; per-story status assertions removed after semantic discovery; retro-noted.)

30th p014 review trio: architect SHIP 92, code-reviewer SHIP 92, security SHIP 95. 17th career p014 catch in 30 applications (~57% career hit-rate; 5th consecutive cycle climbing past 50%).

49 consecutive LOW G30 self-validations (v0.6.5..v0.11.5).
8 test suites green: 220 total (was 217; +3 from Test 11).

The 19-cycle hardening arc (v0.10.6..v0.11.5) is complete. All infrastructure for Path B is in place. Operator can queue a real multi-story feature for first --coordinator dispatch on live repo.

Lessons learned (canonized PIPELINE_REPORT_v55):

  1. PRD off-by-one is structural across 5 cycles; future PRDs should distinguish "test groups" from "sub-asserts" in AC.
  2. Semantic discovery during implementation: Test 11's 5→3 reduction surfaced a real wave-signal-vs-per-story architectural choice not visible from PRD.

Next operator decision: queue a real feature for Path B, OR continue to architectural cycle (run_iteration_loop decomposition).

Assets 2
Loading