Start autonomous Belgium and North Sea RC

docs/BUILD_STATUS.md

@@ -1,38 +1,62 @@
 # GeoIntel Build Status
 
-Current preparation milestone: M7 Implementation Control Layer.
+Updated: 2026-07-17
 
-## Done
+## Current state
 
-- Product blueprint.
-- Data specifications.
-- Architecture specifications.
-- API/database/service documentation.
-- Codex build plans and prompts.
-- Operational readiness docs.
-- Autonomy pack.
-- M7 build control, regression traps and self-review layer.
+GeoIntel is an implemented map-first GeoAI workbench running as an all-in-one
+Unraid container with embedded PostGIS, FastAPI, React/MapLibre, local
+Ollama integration and optional local YOLO/PyTorch inference.
 
-## Ready for Codex
+The active release program is
+`docs/RC_ROADMAP_BELGIUM_NORTH_SEA.md`.
 
-Codex can begin with repository verification and backend foundation using the locked build sequence.
+## Product scope
 
-## Must Preserve
+- Target: all Belgian land and the separately labelled Belgian territorial
+  sea, EEZ and continental shelf.
+- Existing deep regression references: Mol and the Kempen transport region.
+- Required new golden areas: Wallonia, Brussels, a cross-region area, coast
+  and Belgian North Sea.
+- Architecture: federated authoritative providers with one coverage matrix;
+  no assumption that GRB, PICC, UrbIS and NGI are semantically identical.
 
-- GeoIntel is a GeoAI Workbench for the Kempen.
-- GRB-first reference strategy.
-- FastAPI + React + PostGIS.
-- API-driven frontend.
-- CRS-aware geospatial processing.
-- Fixture mode must be clearly labeled.
+## Implemented foundation
 
-## Known Limitations Before Code Build
+- Projects, Areas, Datasets, versions, PostGIS vector features and raster
+  artifacts.
+- Map selection, semantic metrics, historical comparison and exports.
+- Governed Flemish sources for buildings, context, orthophotos, terrain,
+  flooding, soil, nature, agriculture, population and policy rasters.
+- Detection persistence, local configured YOLO/PyTorch path, QA/QC and review
+  evidence.
+- Segmentation persistence foundation without fake production inference.
+- Local Ollama assistant grounded in persisted evidence.
+- Single-container Unraid deployment on port 1202.
 
-- Real GRB WFS integration still needs implementation.
-- Real YOLO/SAM inference should follow fixture boundary first.
-- Sentinel and LiDAR remain post-foundation roadmap items.
-- No production authentication in V1.
+## Release blockers
 
-## Next Recommended Codex Pass
+- Current production backup and isolated restore proof.
+- Fail-closed readiness and truthful runtime capability reporting.
+- Stale job/run reconciliation and correlated exception logging.
+- Temporal compatibility guard for detection QA.
+- Production secret/configuration parity, immutable images and rollback.
+- Full CI, reproducible dependencies and supply-chain evidence.
+- Critical response-model typing and real frontend/browser E2E coverage.
+- National and maritime scope/providers/golden areas.
 
-Run `prompts/codex/PASS_00_REPO_AUDIT.md`, then implement backend foundation according to `docs/12-build-control/BUILD_SEQUENCE_LOCK.md`.
+## Latest evidence
+
+Run:
+
+```bash
+python scripts/capture_release_evidence.py \
+  --output storage/release-evidence/rc-current/baseline.json \
+  --release-id rc-belgium-north-sea \
+  --live-base-url http://192.168.10.150:1202
+```
+
+The evidence manifest is runtime output and is deliberately ignored by Git.
+The first captured baseline reported one Alembic head (`202607160001`) and
+reachable live health/capability routes. Health truthfulness is an open RC-2
+blocker until readiness becomes fail-closed.