01 — Dual-path Baseline + OOM Incident Recheck

DB / containers

PostgreSQL 16.13, postmaster up since 2026-04-17 (PID 1 survives crash-reinit). 11 containers up; incomex-nuxt healthy (image s174), directus healthy.

OOM incident — NOT closed (live evidence wins over prior checkpoint)

Postgres logs showed 5 signal-9 OOM crashes on 2026-06-05: 05:28:27, 05:54:43, 05:58:47, 05:59:34, 06:04:02 UTC. Each: backend killed → "terminating any other active server processes" → automatic crash recovery (WAL redo, ~1–2 s) → "ready to accept connections". NO data loss. Triggers were the prior session querying/EXPLAINing the deep-composite landmine VIEW v_rp_ui_deploy_final_readiness_guard (pre-fix) and v_rp_ui_current_production_acceptance_dashboard. EXPLAIN-alone crashes confirm planner-side cost. No signal-9 since 06:04:02 (DB stable >75 min through this session's work). My function call to fn_rp_ui_deploy_final_readiness_guard() was cancelled cleanly at 5 s (query_pg timeout) — function = safe, same-named composite view = landmine.

Baseline (dual-path: query_pg read-only == ssh→docker→psql, each deep view STANDALONE)

• Contract guard: PASS · 87=87 · invariant_real_fail 0 · expected_blocker 4 · static stale 0 · stale_v1 0 · alias 7/7
• Smoke 15/15 · anti-false-green 6/6 · generated_v2 backbone 87 · invariant_v2 87 pop / 0 real-fail
• birth_registry baseline (background drift continuing; all DDL later proven before==after)

No-fake

ownership 0 · AX-PROCESS official RP 0 · axis_active 0 · PROC-OWN 5 req / 0 votes · events 30 · real_run false · trigger_guard_alerts 129.

Verdict: DB stable now, all green, no-fake intact — but a live OOM landmine remained → Phase 2.