v0.10.0

Added

• CO-OPS observed water levels for one station, explicit datum and UTC interval, with raw CSV quality fields, response validation, and a reproducible manifest example.
• noaa:gsoy annual station CSVs with complete UTC year selection, shared NCEI geographic discovery, reproducible restoration, and a small manifest example.

Documentation

• Keep generated catalog files separate from handwritten documentation and combine dataset reference facts with usage guides in the local docs site.
• Make the dataset catalog easier to browse with Released, Source only, and Planned labels; explicit file formats and selection behavior; and automatic dataset navigation.
• Make the first-use walkthrough the site home, separate NOAA dataset guides from service research, and clarify source installation and reproducibility.

Development

• Add optional fast commit hooks, workflow linting, and PR-title validation; pin GitHub Actions to reviewed commits while retaining Dependabot updates.
• Allow focused manual live-test and notebook CI runs while retaining complete weekly coverage; summarize test failures, skips, and timings in Actions.
• Assemble release notes with Towncrier fragments, validate notes in PRs, and preview upcoming changes in the documentation.
• Generate the radar-site CSV with LF line endings so fresh worktrees stay clean, and apply lint fixes before formatting.
• Prepare releases on a branch before editing files, validate draft release PRs before review/merge, check navigation and notebook release notices, and provide guarded cleanup for merged local branches and worktrees.

v0.9.0

Documentation

• Default the documentation site to dark mode, with a light-mode toggle.

• Add a local documentation site with navigation, Mermaid diagrams, generated API/CLI references, saved notebook previews, and strict CI link checks. Separate generated dataset catalogs from provider access notes.

Added

• Per-dataset live checks, independent example jobs, retained test/coverage and
  failed-notebook diagnostics, and scheduled minimum-core-dependency validation.

• Shared adapter conformance scenarios for every available dataset, covering
  deterministic listing, fetch destinations, byte preservation, and client ownership.

• Documented L0–L4 test levels, responsibility-based suites, and just test-live
  selection; existing integration command and marker aliases remain supported.

• Pure select_by_time asset selection with explicit tolerance and nearest/prior
  direction, signed offsets, candidate counts, and an explicit no-match result.

• Explicit zero-based radar sweep selection with FetchedAsset.open(sweep=...)
  and an executed Storm Events / NEXRAD / GOES event-context notebook.

Fixed

• Require Typer 0.18 or newer; the old declared minimum failed to construct the
  CLI with modern union annotations and lacked current Click compatibility.

• Reject truncated S3 listings with missing or cycling continuation tokens instead
  of returning incomplete assets or requesting pages indefinitely.

• Reject NEXRAD decoder tables that misalign moment and coordinate records,
  including equal-length sweeps following a missing interior end marker; valid
  explicitly selected sweeps remain readable without silently dropping others.

• Reject out-of-range or non-finite point coordinates and negative/non-finite
  radii before bounding-box clipping; complete the manifest provider options for
  GSOM, GOES ABI, and Storm Events.

v0.8.0

Added

• Optional local NetCDF4 opening with eager xarray loading, CF decoding,
  closed file resources, source metadata, and an executed GOES infrared notebook.
• noaa:goes-abi CONUS Cloud and Moisture Imagery with explicit satellite/channel
  selection, precise scan times and sizes, and checksum-verified NetCDF downloads.
• Storm Events annual event-details gzip CSVs, selecting current creation-date
  revisions with exact source-byte provenance and locked restoration; local
  compressed CSV opening preserves event and geographic identifier strings.
• Executed Storm Events notebook with local Oklahoma/date filters, event-record
  and damage-rating plots, and source/reporting caveats.
• Optional NEXRAD Level II opening with xradar sweep DataTrees, native units and
  flag masking, source provenance, and an executed reflectivity notebook.

Changed

• Replace example analysis scripts with executed Jupyter notebooks containing
  compact data previews, plots, and source provenance. Add an optional examples
  environment, offline notebook checks, and isolated live execution/refresh commands.
• Pin the development interpreter to Python 3.14.7 to avoid an upstream crash
  affecting NetCDF decoding in older Linux uv Python builds.

v0.7.0

Added

• Opt-in GHCN station-search diagnostics with bounded response details, and
  independent live checks for geographic discovery and explicit-station downloads.
• noaa:gsom monthly station CSVs through NCEI, with geographic discovery,
  complete-month selection, reproducible restoration, and a pandas analysis example.
• Terminal progress for CLI fetch and pull, showing known asset sizes, HTTP download
  bytes and retry attempts, and validated cache hits. Disable with --no-progress;
  redirected output retains its existing format.

v0.6.0

Added

• Optional FetchedAsset.open() CSV reading through usdata[pandas], including
  ERDDAP units metadata, string identifiers/codes, explicit dtype/date/column/row
  options, and a fetch-to-analysis example. Opening leaves raw bytes and provenance intact.

Changed

• Test core-only and pandas installations in CI and installed-wheel checks.
• Check versioned source-only documentation notices during validation so release
  PRs cannot carry known stale availability wording.
• Focus v0.6 on optional CSV readers and practical usage; NASA, model output,
  climate grids, and remote caching remain follow-up work.

v0.5.0

Breaking

• Manifest pulls now fail when a required source resolves to no assets. Set
  allow_empty: true per source to permit an intentionally empty result.
• verify rejects a manifest edited since locking (exit 2), matching pull.
• NOAA adapters reject unknown provider parameters, empty explicit identifiers,
  invalid units, and conflicting or invalid radar selectors.

Fixed

• Generated provider and roadmap docs distinguish unreleased implementations
  from support included in the declared package version.
• GET metadata and download requests retry transient failures up to three attempts,
  respecting bounded Retry-After delays and preserving existing files on failure.
• Release publishing promotes the distributions from successful CI for the exact
  release commit, with version checks, instead of rebuilding.

Changed

• Narrow v0.5 to USGS, ERDDAP/CoastWatch, complete place lookup, and reliability;
  additional NOAA adapters and optional readers remain follow-up work.
• Enforce offline unit tests and smoke-test the installed wheel on Linux, macOS,
  and Windows. Add release-tool regression tests.
• Add a manifest reference and a small NOAA/USGS example with a live round-trip test.

Added

• noaa:coastwatch-sst: bounded ERDDAP gridded CSV subsets by location, UTC
  time, and variable, with optional spatial stride and reproducible restoration.

• Census 2025 lookup for 56 states/DC/territories and 3,235 counties/equivalents,
  including county/state names and quoted FIPS. Add --location as an alias for
  --state. Generated bounds replace the six approximate seed boxes, so
  geographic queries may select different stations; existing lockfiles stay pinned.

• usgs:water-daily: the first non-NOAA adapter, using the modern USGS Water Data
  API for paginated CSV downloads by site or bbox, dates, parameter codes, and
  statistic. Preserves units and quality metadata; supports cached fetches,
  manifest pull, locked restoration, and verification without new dependencies.

v0.4.0

Breaking

• Manifest and source fields reject unknown keys instead of silently ignoring them;
  provider-specific inputs remain supported under params.

Fixed

• Verify cached bytes and source provenance before reuse or fresh lockfile creation.
• Reject unsafe cache paths, including symlinks escaping the cache root.
• Preserve existing files after failed downloads or checksum mismatches; replace downloads,
  provenance, and lockfiles atomically using unique temporary files.
• Reuse provider clients across downloads and close internally owned clients on success or failure.
• Report invalid dates, radar IDs, manifest YAML, and lockfiles with clear CLI input errors.
• Infer multiple domains correctly when constructing a custom registry without a domain catalog.
• Gate package publishing on successful CI for the exact main-branch release commit.

v0.3.0

Added

• usdata pull fetches every source in a manifest and writes a lockfile pinning each asset's
  checksum and provenance; a later pull restores from the lockfile without re-resolving, and
  refuses (until --force) if the manifest changed. usdata verify reports missing or altered
  cached files. Python API: usdata.pull, usdata.verify.

• Datasets carry a status (available, stub, planned). search and info show it; fetching a
  planned dataset exits 3 with a clear message.

• Planned registry entries for USGS, NASA, EPA, FEMA, Census, USDA, and GOES so the roadmap is
  visible from usdata search and the docs.

• Provider metadata (name, homepage) in the registry; README gains a generated per-provider table.

• Datasets carry a domain (shared taxonomy across providers), since (version shipped) or
  target (planned phase or later). search and info show them; the roadmap lists
  datasets by target version, generated from the registry.

• 26 NOAA registry entries spanning 18 domains, each verified for anonymous access
  (except GHCN-Hourly, whose entry says its bulk path is unconfirmed), with target phases.
  The NOAA provider page gains a data-landscape table mapping domains to entries.

Changed

• usdata search hides planned datasets unless --planned is given.

Changed

• Dataset reference moved to docs/providers/: one page per provider with hand-written access notes and a generated dataset table, plus a generated index.

v0.2.1

Added

• Type information is shipped (py.typed).
• Contributor docs: CONTRIBUTING.md, SECURITY.md, a guide to adding datasets, issue and PR templates.

v0.2.0

First usable release.

• Curated dataset registry with keyword search filtered by provider, bbox, and time
• noaa:ghcn-daily: station observations via the NCEI search and data services
• noaa:nexrad-level2: Level II volume scans from the unidata-nexrad-level2 bucket, anonymous S3 over HTTPS
• Local cache with sha256 verification and a provenance sidecar for every fetched file
• CLI: usdata search, info, fetch (with --dry-run), and manifest validation in pull

Not yet implemented: ERDDAP adapter, pull downloads and lockfiles. See docs/roadmap.md.