v0.1.2 — Deep Review & Performance Tuning

v0.1.2 — Deep Review & Performance Tuning

Closes v0.1.1's 4 deferred-at-shipping items with measured, on-disk evidence.

What's New

CI Substrate Hardening (Phase 14)

• pg-lunaris container (pgvector + AGE + pgmq) and Moon service in GitHub Actions
• 4 EVAL-05 regression tests that fail on vanilla postgres:16, pass on pg-lunaris
• moon-it and pg-it integration features enabled by default on PR CI

Moon Performance Parity (Phase 15)

• source TAG field added to Moon FT schema with Filter::Eq push-down
• Moon p50 = 2.40 ms vs Postgres p50 = 5.97 ms (Moon 2.5× faster under clean conditions)
• Moon p99 = 4.04 ms (well under 20 ms target)
• Analytical hotspot audit: bottleneck is I/O-bound, no hash-map swap warranted

ActRConsolidator Default (Phase 16)

• ConsolidatorPipelineHandle default flipped from NoopConsolidator → ActRConsolidator
• promotion_rate SLO enforced at [0.00, 0.01] in EVAL-05 CI gate
• Worker tick p50 ≈ 72 µs on both backends (696× under 50 ms SLO)

HELIOS Live UAT (Phase 17)

• Criterion helios_p50 baseline script with pre-flight checks + VERDICT.json
• SIGKILL chaos 100×2 script with per-run evidence aggregation
• Both HELIOS-05 and HELIOS-06 lifted from deferred → validated

Audit Debt Closure (Phase 18)

• Zero-promotion root cause identified: unique episode_id per write → ACT-R activation = 0.0
• Operator scripts hardened against LUNARIS_CONSOLIDATOR_BACKEND=noop
• lunaris-recipes wired into CI integration workflow

Performance (EVAL-05 10K-turn replay)

Metric	Moon	Postgres	Target
p50	2.40 ms	5.97 ms	Moon ≤ PG × 1.10 ✓
p95	3.20 ms	92.69 ms	—
p99	4.04 ms	151.45 ms	Moon ≤ 20 ms ✓

Stats

• 5 phases, 23 plans, 17/17 requirements satisfied
• 0 critical blockers, 8 deferrable tech debt items
• Shipped in ~48 hours from Phase 14 kickoff

Full Details

See milestones/v0.1.2-MILESTONE-AUDIT.md for the complete audit report.

Full Changelog: v0.1.1...v0.1.2