jeafreezy · Oct 2, 2026 · Oct 2, 2026
diff --git a/.github/workflows/CI.yml b/.github/workflows/CI.yml
@@ -23,6 +23,8 @@ jobs:
           toolchain: '1.88.0'
       - uses: Swatinem/rust-cache@v2
       - run: cargo test --workspace --exclude superstac-python --locked
+      - run: cargo test -p superstac-cli -p superstac-engine --features geoparquet --locked
+      - run: cargo check -p superstac-python --no-default-features --locked
 
   wheels:
     name: Wheel (${{ matrix.target }})
@@ -61,6 +63,11 @@ jobs:
           python -m pip install --no-deps --force-reinstall dist/*.whl
       - name: Test the installed wheel
         run: python -m unittest discover -s crates/python/tests -v
+      - name: Execute the GeoParquet notebook against local fixtures
+        if: matrix.target == 'x86_64-unknown-linux-gnu'
+        run: |
+          python -m pip install pyarrow nbformat nbclient ipykernel
+          python scripts/test-geoparquet-notebook.py
       - uses: actions/upload-artifact@v4
         with:
           name: wheels-${{ matrix.target }}

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -7,6 +7,23 @@ releases.
 
 ## [Unreleased]
 
+### Added — v0.3.0
+- Optional Rust/CLI GeoParquet backend and GeoParquet-enabled Python wheels.
+- Scoped, named, resumable STAC metadata ingestion with progress and storage caps.
+- Snapshot and automatic API fallback modes, freshness checks, and gap-aware scope unions.
+- Per-catalog sorting, local collection summaries, row-group pruning, and a reproducible benchmark.
+- Explicit acquisition-window incremental overlays, atomic compaction, and cleanup with reader protection.
+- Sync/async Python ingestion, backend configuration, maintenance, and type hints.
+- Automatically versioned Parquet provenance and search diagnostics.
+- A GeoParquet Jupyter/Colab notebook, release highlights page, and documented local benchmark.
+
+### Changed — v0.3.0
+- Dataset manifests use version 2; legacy version-1 inventories remain readable.
+- Release publishing includes the GeoParquet crate before the engine; CI checks both feature modes.
+
+### Fixed — v0.3.0
+- Forward sort fields to live STAC searches instead of silently discarding them.
+
 ## [0.2.0] - 2026-09-27
 
 ### Added

diff --git a/Cargo.lock b/Cargo.lock
diff --git a/README.md b/README.md
@@ -24,9 +24,12 @@ normalized to canonical names — regardless of which catalog they came from.
 > **Status: alpha.** APIs and YAML schema are not yet stable. Pre-1.0; expect
 > breaking changes.
 
+**[What’s new in v0.3](https://spatialnode.com/superstac/docs/releases)** · [Try GeoParquet in Colab](https://colab.research.google.com/drive/1JIx8j0Vi7jypKe5Q6klgCzYQMFHMaKGc?usp=sharing) · [Local search benchmark](https://spatialnode.com/superstac/docs/guides/benchmarks)
+
 ## Features
 
 - **Federated search:** Query multiple STAC catalogs concurrently with spatial, temporal, collection, and item filters.
+- **Reusable satellite-data inventories:** Save metadata for your area of interest and explore it repeatedly without querying every provider again. Fall back to live catalogs when saved coverage is missing or stale.
 - **Collection discovery:** Find available collections and the catalogs that serve them.
 - **Consistent names:** Map provider-specific collection IDs and asset keys to canonical names through configurable aliases.
 - **Deduplicated results:** Merge items across catalogs by item ID, with source provenance available in Rust and CLI results.
@@ -69,6 +72,7 @@ cargo add superstac-search
 cargo add superstac-engine
 cargo add superstac-cli
 cargo add superstac-config
+cargo add superstac-geoparquet
 ```
 
 ### From source
@@ -106,7 +110,9 @@ superstac --json search -c landsat-c2-l2 -l 10 | jq '.metadata'
 superstac collections microsoft sentinel-2-l2a
 ```
 
-Run `superstac --help` for the full surface.
+Run `superstac --help` for the full surface. For saved metadata inventories, build
+with `--features geoparquet` and follow the [GeoParquet guide](crates/geoparquet/README.md).
+Published Python wheels include this backend.
 
 ## Configuration
 
@@ -178,11 +184,18 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
 | [`superstac-core`](https://crates.io/crates/superstac-core) | domain models, errors, storage trait |
 | [`superstac-config`](https://crates.io/crates/superstac-config) | YAML config loading |
 | [`superstac-search`](https://crates.io/crates/superstac-search) | federated search logic |
+| [`superstac-geoparquet`](https://crates.io/crates/superstac-geoparquet) | Reusable satellite-data inventories for repeated searches |
 | [`superstac-engine`](https://crates.io/crates/superstac-engine) | runtime (health, introspection, search orchestration) |
 | [`superstac-cli`](https://crates.io/crates/superstac-cli) | the `superstac` binary |
 
 ## Logs and debugging
 
+Include `superstac --version` or Python’s `superstac.__version__` in bug reports.
+Search diagnostics also include `metadata.superstac_version`. Provider requests
+identify the library as `superstac/<version>` through the User-Agent header.
+New ingested and compacted Parquet files record the writer version in their footer.
+
+
 Logs flow through `tracing`. The default level comes from `settings.log_level`
 in your config; override at runtime:
 

diff --git a/ROADMAP.MD b/ROADMAP.MD
@@ -3,29 +3,36 @@
 Our priority is reliable federated STAC search. These are areas of focus,
 subject to feedback and capacity, with no fixed delivery dates.
 
-Last Updated: September 28th, 2026
+Last updated: October 2, 2026
+
+## Implemented for v0.3
+
+- [x] [Reusable satellite-data inventories with GeoParquet](https://github.com/Spatialnode/superstac/issues/31): scoped ingestion with progress and resume, local search and API fallback, Python integration, scope unions, sorting, incremental refresh, compaction and cleanup.
+
+
 
 ## Short term
 
-- Fix search and iterator edge cases, backed by focused tests.
-- Improve configuration errors, logging, and documentation.
-- Support private catalogs through custom headers and environment-based credentials.
+- [ ] [Support private catalogs](https://github.com/Spatialnode/superstac/issues/32) through per-catalog headers and environment-based credentials, consistently across search, discovery, health checks and ingestion.
+- [ ] [Add AI/agent integration](https://github.com/Spatialnode/superstac/issues/22): define a supported tool interface for catalog discovery and bounded searches, with a working agent example. Existing Python APIs can be called by agents; a dedicated integration is still pending.
+- [ ] [Explore WASM bindings](https://github.com/Spatialnode/superstac/issues/21): validate a browser-compatible API search client first, including CORS and runtime constraints. Native filesystem and GeoParquet support should not be assumed to work in browsers.
 
 ## Near term
 
-- Add catalog preferences and improve health-based fallback.
-- Introduce SQLite persistence.
-- Improve pagination and result handling across Python, Rust, and the CLI.
+- [ ] Add explicit preferred-catalog ordering and failover between providers. Improve recovery of catalogs unavailable at startup. Health checks, retries, concurrent search and snapshot-to-API fallback already exist.
+- [ ] Implement SQLite persistence for catalog registration, provider configuration and settings. 
+- [ ] Expose resumable federated pagination and bounded/lazy result iteration across Python, Rust and the CLI. Provider pagination is already followed internally, but callers currently receive materialized, capped results rather than a federated cursor or lazy stream.
 
 ## Long term
 
-Explore broader authentication, additional storage backends, and ecosystem
-integrations as real use cases emerge.
+Explore OAuth/token refresh and asset signing, additional configuration storage
+backends, and ecosystem integrations as real use cases emerge. These extend the
+basic credential support above rather than duplicate it.
 
 ## v1.0 release
 
-- Stable public APIs and configuration format.
-- Well-tested core search, with predictable results and failure handling.
-- Clear documentation, supported-platform coverage, and migration guidance.
+- [ ] Stable public APIs and configuration format, with a compatibility policy.
+- [ ] Release confidence in core search, with predictable results and failure handling. Existing tests are a foundation; scale, interoperability and failure-recovery coverage remain ongoing work.
+- [ ] Maintain documentation, supported-platform coverage and migration guidance for every stable release. Documentation and packaging already exist; this is a continuing release requirement.
 
 The v1.0 release depends on core stability; longer-term ideas are optional.