Fix day-2 cell rehydration and add record-level license default

[jsimonclark]

What

Two pre-0.8 package fixes (blind-review findings 3 and 4):

1. Day-2 workspace rehydration. In a fresh process, ws.add("test", cell="S1") now resolves a cell whose record sits in .battinfo/records/cell-instance/, instead of raising Could not resolve cell 'S1' .... The resolver lazily rehydrates the serial index from disk, and the failure message now names what was actually searched.

2. Record-level license. New ws.license("cc-by-4.0") sets a workspace default that is persisted in .battinfo/workspace.json and stamped onto every saved record, so records carry a license by themselves (FAIR R1.1) rather than only on the Zenodo leg.

Why

• The serial to record index was session-memory only; the resolver's "locally" branch never scanned saved records, so authoring a test the day after saving cells failed with a false error. Reproduced with a cold subprocess.
• Served and locally saved records carried no license (two reviewers, independently): license existed only if the author took the Zenodo path, so a record dereferenced at its identifier stated no reuse terms.

How

Rehydration:

• _resolve_cell now, on a miss, calls _rehydrate_cell_index() (a lazy, once-per-instance scan of .battinfo/records/cell-instance/, sharing the loader core with reload_cells) before falling back to the registry. Error text rewritten to name the saved-records directory and note when the registry is not configured.
• Protocol/spec references already scan disk (_resolve_test_protocol_ref, _resolve_channel); a cold-process test pins that so it never regresses to the session-only defect the cell path had.

License:

• ws.license(id|clear=...) mirrors ws.contributor/ws.project: setter/getter/clear, persisted in workspace.json. Known SPDX-ish ids are case-normalized; an unrecognized string is kept verbatim with a warning (never hard-rejected).
• Stamped through the existing PR Fix [unchanged] misreport with contributors; teach vent part type #324 pre-compare _record_stamp path so byte-identical re-saves still report [unchanged]; datasets carry it on the existing dataset.license slot (leaving an explicit ws.add(license=) untouched), other record types at the record top level.
• Emission reuses the existing dataset convention: dcterms:license in record_to_jsonld, schema:license/dcterms:license in the publication bundle. ws.zenodo/preview_jsonld fall back to the workspace default when no license= is passed.
• Gold-standard panel: a licenseless publication emits a publication.license_missing warning (not an error) nudging ws.license(...).
• Schemas: added a top-level license string to the cell-spec/cell-instance/test/test-protocol save gates (edited assets/schemas source of truth, re-synced the packaged mirror, ran web/scripts/sync-schemas.mjs). Datasets already accept a nested dataset.license.

Precedence (documented in the ws.zenodo and ws.add docstrings): explicit ws.zenodo(license=...) > explicit ws.add(..., license=...) > workspace default.

Public API added (for the docs agent)

• ws.license(license: str | None = None, *, clear: bool = False) -> str | None — setter (returns normalized id), getter (no args; returns/prints current or None), clear=True to remove. Persisted in .battinfo/workspace.json under license; stamped onto every record on ws.save().
• ws.zenodo(..., license: str | None = None) — was license: str = "cc-by-4.0"; now defaults to the workspace default, then cc-by-4.0.
• ws.add("test", ..., license=...) — unchanged, still overrides the workspace default for that record's datasets.
• Precedence: ws.zenodo(license=) > ws.add(license=) > ws.license() default.

Testing

• uv run pytest — 1621 passed, 1 skipped. uv run ruff check clean on all changed files (pre-existing notebook/demo lint in docs/ and demo/ is untouched and out of territory). uv run mypy — success, 69 files.
• New tests/test_ws_rehydration.py: cold-subprocess resolve of a cell serial saved in a prior process; fresh-instance resolve; truthful error text; cold-process protocol-reference resolve.
• New tests/test_ws_license.py: setter/getter/persistence/clear, case-normalization, unknown-kept-verbatim-with-warning, URL verbatim, stamping into every saved record + record_to_jsonld, no-license-means-no-stamp, explicit-add-overrides-default precedence, preview bundle carries the license, ws.zenodo default resolution + explicit override, missing-license warning nudge + no-nudge-when-present.
• Extended the PR Fix [unchanged] misreport with contributors; teach vent part type #324 stamp-order regressions in tests/test_pkg_silent_loss.py: identical re-save with license set -> [unchanged] (bytes + mtimes unchanged); changing the workspace license -> [updated].

Rehydration repro (registry_url=None), before vs after:

Before: session B (fresh process) -> ValueError: Could not resolve cell 'S1' in this session, locally, or the registry. Create it with ws.add('cell', ...), or check the serial.

After: session B (fresh process) -> resolves S1 and creates the test; an unknown serial now errors with ... in this session, the records saved under <root>\.battinfo\records (registry not configured). ....

Pre-existing flakes: none observed. Both full-suite runs (before and after adding tests) were fully green; the documented intermittent rehydrate/finalize flakes did not appear, so nothing needed diffing against main.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
batt-info	Ready	Preview	Aug 1, 2026 12:10am