v0.5.0

What's Changed

• chore: establish v0.5.0 hardening governance by @GeoGeekLab in #10
• fix: enforce staged request constraints by @GeoGeekLab in #12
• fix: harden antimeridian and polar coverage by @GeoGeekLab in #13
• fix: harden raster planning semantics by @GeoGeekLab in #14
• fix: harden provider and federation networking by @GeoGeekLab in #15
• fix: version manifests and make replay completeness-aware by @GeoGeekLab in #17
• test: expand deterministic hardening regression corpus by @GeoGeekLab in #18
• chore: harden v0.5.0 release and supply chain by @GeoGeekLab in #19
• ci: make v0.5.0 release branch self-create tag by @GeoGeekLab in #23
• ci: gate release on clean PyPI install by @GeoGeekLab in #25

Full Changelog: v0.4.0...v0.5.0

v0.4.0 — Geo Task Intelligence

v0.4.0 — Geo Task Intelligence

STAC is syntax. Dataset choice is semantics.

STAC Scout v0.4.0 adds a deterministic Geo Task Intelligence layer: users can describe the geospatial problem they want to solve, and Scout derives explicit, inspectable data requirements before touching a catalog.

What's new

• Geo Task Intelligence

  • Versioned task profiles for vegetation, wildfire, flood, surface water, snow, land-cover change, urban change, and terrain analysis.
  • Explicit separation between required and preferred measurements.
  • Task-aware temporal strategies such as single-window, before/after, and static analysis.
  • Rule-level rationale, sources, and provenance.

• Task-aware request enrichment

  • TaskAdvisor enriches ScoutRequest without silently overriding explicit user constraints.
  • Missing comparison windows remain explicit follow-up requirements.
  • Incompatible modality defaults are skipped instead of producing impossible band requirements.

• Intent → task semantics

  • IntentDraft can carry a task_type.
  • Language models may classify the task, but deterministic rules own the geospatial requirements.
  • No invented coordinates, dates, bands, or hidden constraints.

• Cross-provider federation

  • Provider registry and adapter architecture.
  • Conservative dataset identity with exact / probable / local matching.
  • Provider failures remain visible instead of disappearing from results.

• Live verification

  • Item-level availability checks.
  • Geodesic AOI coverage.
  • Asset and band inspection.
  • Explicit PASS / FAIL / UNKNOWN constraint semantics.

• Access planning & provenance

  • Measurement → asset resolution.
  • Windowed transfer estimates.
  • odc-stac loading recipes.
  • Replayable manifests with Item-set drift detection.

CLI highlights

stac-scout tasks
stac-scout task-profile wildfire_impact
stac-scout advise request.json --task wildfire_impact

stac-scout federate request.json
stac-scout verify request.json --provider earth-search --collection sentinel-2-l2a
stac-scout plan request.json --provider earth-search --collection sentinel-2-l2a
stac-scout replay scout.manifest.json

Quality

• Python 3.12 / 3.13
• 74 tests
• 90%+ coverage
• strict mypy
• ruff lint + format checks
• deterministic evaluation corpus
• wheel/package-data verification

Philosophy

metadata ≠ evidence
similarity ≠ identity
availability ≠ suitability
confidence ≠ proof

Find it. Verify it. Plan it. Replay it.