geointel/docs/SPRINT_BOARD_M4.md

GeoIntel M4 Sprint Board

Sprint 0 — Repo Bootstrapping

• Verify repo root structure.
• Add backend package skeleton.
• Add frontend package skeleton.
• Ensure Docker Compose starts PostGIS and Redis.
• Add .env.example parity with backend settings.
• Add smoke command documentation.

Acceptance:

• A new developer can run the documented startup commands.
• Missing optional GIS binaries degrade gracefully with explicit warnings.

Sprint 1 — Backend Foundation

• FastAPI app factory.
• Health endpoint.
• Settings loader.
• Structured error response middleware.
• Database session management.
• Alembic migration baseline.

Acceptance:

• /health returns app, db, redis and storage status.
• Failed validation returns the standard error envelope.

Sprint 2 — Domain Models

• Project model.
• Area model with PostGIS geometry.
• Dataset model.
• AnalysisRun model.
• Detection model.
• Segmentation model.
• Metric model.
• Export model.

Acceptance:

• Migration creates all core tables.
• Geometry columns use SRID 31370 internally where applicable.
• API schemas never expose internal file-system paths unless explicitly an export path.

Sprint 3 — Dataset Manager

• Dataset upload endpoint.
• Dataset metadata extractor dispatcher.
• Raster metadata extractor.
• Vector metadata extractor.
• Dataset list/detail endpoints.
• Dataset status transitions.

Acceptance:

• Uploading GeoJSON fixture produces vector metadata.
• Uploading placeholder/demo raster either extracts metadata or returns a clear unsupported-fixture status.

Sprint 4 — Area & Map Workspace

• Create area from polygon.
• Validate area geometry.
• Calculate area in square meters.
• Return areas as GeoJSON.
• Frontend map shell.
• Draw polygon flow.

Acceptance:

• A user can create a project and add at least one polygon area through the UI.

Sprint 5 — Raster/Vector Labs

• Raster metadata page.
• Vector metadata page.
• Vector clip operation.
• Raster operation contract endpoints.
• Operation job creation.

Acceptance:

• Vector fixture clipped to an area returns a new derived layer record.

Sprint 6 — AI Pipeline Shells

• Model registry seed.
• Detection run endpoint.
• Segmentation run endpoint.
• Deterministic demo-mode inference.
• Store detections/segmentations as geometry outputs.

Acceptance:

• Detection run on demo area creates predictable detections with classes and confidence.
• Segmentation run creates polygons and area metrics.

Sprint 7 — QA/QC

• Reference dataset selection.
• IoU matching.
• Precision/recall/F1.
• False positive/false negative outputs.
• QA dashboard panel.

Acceptance:

• Demo detections compared to demo reference produce stable metrics.

Sprint 8 — Exports

• GeoJSON export.
• CSV metrics export.
• Report stub export.
• Export list/detail page.

Acceptance:

• User can download detection results as GeoJSON.

Sprint 9 — Codex RC Pass

• Run tests.
• Update TODO.
• Update CHANGELOG.
• Remove dead placeholders.
• Verify all pages have empty/loading/error/success states.