01 Dual-Path + OOM Baseline
01 — Dual-Path Baseline + OOM Stability
OOM stability (the safety gate)
Live postgres logs reviewed across 06:09 → 08:39 UTC. Zero signal-9 / "Killed" / crash-recovery events. The last OOM crash was 2026-06-05 06:04:02 UTC (prior session). The pause-condition (any new OOM after 06:04) was NOT triggered.
Two graceful statement-timeouts were observed on v_rp_final_operating_acceptance_dashboard → fn_rp_ui_deploy_final_readiness_guard() → v_rp_ui_current_smoke_probe (at 06:23 and 08:25 UTC). These are the intended behaviour of the function-backed guard: it cancels on the 5s statement_timeout rather than OOM-crashing. Reading that dashboard requires the slow path (statement_timeout=0 via ssh), which returned cleanly.
v_rp_guard_safety_status= OOM_SAFE__NO_LIVE_SMOKE_COMBO_LANDMINE, live_crash_landmines 0, 7 function-backed guards, 10 slow-bounded views.v_rp_oom_landmine_detector: 16 SAFE + 7 SAFE_FUNCTION_BACKED + 10 SLOW_BOUNDED_NO_SMOKE = 33 RP views, 0 CRASH_LANDMINE.- All 16 new views in this macro read base tables + light views only; none references the smoke probe. No EXPLAIN of any smoke-combo view was run.
Dual-path verification
query_pg (read-only role, 5s timeout, LIMIT 500) used for fast reads; ssh → docker exec postgres psql -U directus -d directus used for the DDL apply and the slow function-backed dashboard. They agree. Example: scorecard via ssh = AUTOMATION_REALITY_VERIFIED_WITH_PARTIAL_SOURCES|94|PASS|9|2, identical to the RO read.
No-fake baseline (unchanged from SSOT)
births 1,205,440 (+21 background from 1,205,419; DDL-attributable 0) · trigger_guard_alerts 129 · axis_registry 2 (0 ACTIVE) · governance_object_ownership 0 · axis_assignment 25 · AX-PROCESS official RP 0 · PROC-OWN 5 requests / 0 votes · events active 30/52 · dot_config: real_run_enabled=false, execute_enabled=false, emit_enabled=false, operator_runtime_enabled=false.
Production health
11 containers up; incomex-nuxt nuxt-ssr-local:s174 up 17h (RP UI not deployed — operator-gated, unchanged); wf-universal-scanner.timer enabled/active (next 2026-06-06 04:10 CEST); postgres up 7 weeks healthy.