v0.3.2b3

Fixed

• Atmosphere.upload_blob() TypeError: The timeout heuristic passed timeout= to Client.upload_blob() which only accepts (data: bytes). Switched to the namespace method com.atproto.repo.upload_blob() which forwards kwargs through to httpx

Testing

• ATProto SDK signature compatibility tests: New test_atproto_compat.py with 7 tests that instantiate a real atproto Client (with ClientRaw._invoke patched) to validate method signatures without network I/O. Covers upload_blob, create_record, list_records, get_record, delete_record, and export_session

v0.3.2b2

Added

• Lexicon-mirror type system: StorageHttp, StorageS3, StorageBlobs, BlobEntry, ShardChecksum dataclasses that mirror ATProto lexicon definitions, with storage_from_record() union deserializer
• ShardUploadResult: Typed return from PDSBlobStore.write_shards() carrying both AT URIs and blob ref dicts
• Lexicon reference docs: Auto-generated documentation page for the ac.foundation.dataset.* namespace
• Example docs: dataset-profiler, lens-graph, and query-cookbook with plots and interactive tabsets
• Typed proxy DSL for manifest queries (foundation-ac #43)

Changed

• DatasetPublisher refactored: Extracted _create_record() helper, fixing a bug where publish() used dataset.url instead of dataset.list_shards() for multi-shard datasets
• PDSBlobStore.write_shards() returns ShardUploadResult instead of using a _last_blob_refs side-channel
• Blob storage uploads: PDS blob uploads now use storageBlobs with embedded blob ref objects instead of string AT URIs in storageExternal, preventing PDS garbage collection of uploaded blobs
• Replaced lexicon symlinks with real files
• Guarded redis imports behind TYPE_CHECKING in index/_entry.py and index/_index.py
• Standardized benchmark outputs to .benchmarks/ directory

Fixed

• publish() multi-shard bug: was passing single URL instead of full shard list
• Double-write eliminated in PDSBlobStore
• Lens-graph example: removed float rounding in calibrate lens that broke law assertions
• Unused imports and E402 violations in atmosphere module

Testing

• Strengthened weak mock assertions with argument verification across 4 test files
• Fixed misleading unicode tests: real emoji (🌍🎉🚀) and CJK characters (日本語テスト, 中文测试, 한국어시험) instead of ASCII placeholders
• Exact shard count assertions instead of >= 2 bounds
• Fixed self-referential assertion in test_publish_schema
• Added content assertions for empty/corrupted shard recovery tests

v0.3.2b1

What's Changed

• release: v0.3.2b1 by @maxine-at-forecast in #5
• release: v0.3.2b1 by @maxine-at-forecast in #6

Full Changelog: v0.3.1b1...v0.3.2b1

v0.3.1b1

What's Changed

Added

• Lexicon packaging: ATProto lexicon JSON files bundled in src/atdata/lexicons/ with importlib.resources access via atdata.lexicons.get_lexicon() and list_lexicons()
• DatasetDict single-split proxy: When a DatasetDict has one split, .ordered(), .shuffled(), .list_shards(), and other Dataset methods are proxied directly
• write_samples(manifest=True): Opt-in manifest generation during sample writing for query-based access
• Example documentation: Five executable Quarto example docs covering typed pipelines, lens transforms, manifest queries, index workflows, and multi-split datasets
• Bounds checking in bytes_to_array() for truncated/corrupted input buffers

Changed

• AtmosphereClient → Atmosphere: Renamed with factory classmethods Atmosphere.login() and Atmosphere.from_env(); AtmosphereClient remains as a deprecated alias
• sampleSchema → schema: Lexicon record type renamed from ac.foundation.dataset.sampleSchema to ac.foundation.dataset.schema (clean break, no backward compat)
• Module reorganization: local/ split into index/ (Index, entries, schema management) and stores/ (LocalDiskStore, S3DataStore); local/ remains as backward-compat re-export shim
• CLI rename: atdata local subcommand renamed to atdata infra
• Uniform Repository model: Index now treats "local" as a regular Repository, collapsing 3-way routing to 2-way (repo/atmosphere)
• SampleBatch aggregation uses np.stack() instead of np.array(list(...)) for efficiency
• Numpy scalar coercion in _make_packable — numpy scalars extracted to Python primitives before msgpack serialization

Fixed

• Schema round-trip in Index.write() — schemas with NDArray fields now survive publish/decode correctly
• Duplicate force-include in pyproject.toml causing PyPI wheel upload rejection

Full Changelog: v0.3.0b2...v0.3.1b1

v0.3.0b2

What's Changed

• Release v0.2.1a1 by @maxine-at-forecast in #1
• release: v0.3.0b2 by @maxine-at-forecast in #2

New Contributors

• @maxine-at-forecast made their first contribution in #1

Full Changelog: https://github.com/forecast-bio/atdata/commits/v0.3.0b2