This guide defines how empirical ant movement, pose, spectroscopy, and behavior datasets should enter Ant Stack analyses. Imported data must remain traceable, validated, and distinct from generated examples.
flowchart TD
Source["External dataset"] --> Provenance["provenance.yaml"]
Provenance --> Raw["data/raw"]
Raw --> Normalize["schema normalization"]
Normalize --> Validate["validation outputs"]
Validate --> Package["antstack_core analysis or cohereants"]
Package --> Claim["source-backed paper or docs claim"]
data/external/<dataset_id>/
provenance.yaml
raw/
derived/
validation/
README.md
Each dataset needs:
- dataset id and version
- source organization or publication
- acquisition date
- license or use constraints
- species, colony, and experimental context when available
- measurement modality and units
- processing steps
- validation command and output path
Trajectory rows should include frame, time_s, ant_id, x, y, and coordinate units. Pose rows should include body landmarks, confidence values, and pixel-to-unit calibration when available. Behavioral event rows should include event_id, ant_id, start_time_s, end_time_s, label, and label provenance.
- Preserve unmodified source inputs in
raw/. - Normalize into
derived/with explicit units. - Validate schema, missingness, coordinate ranges, timing monotonicity, and label consistency.
- Record validation output under
validation/. - Connect validated data to
antstack_core.analysis.empirical_data,antstack_core.cohereants.behavioral, or manuscript tables.
Current repository checks for data-facing contracts:
uv run run-all-antstack --config configs/run_all_antstack.example.yaml --tasks data,statistics,reports,validation --validate-only
uv run pytest -q tests/antstack_core/test_veridical_reporting.py tests/antstack_core/test_cohereants_behavioral.py
uv run pytest -q tests/antstack_core/test_docs_contract.pyA dedicated external-data importer should be added as a package-backed console entrypoint before any importer command is documented here.
- Do not treat generated examples as empirical observations.
- Cite external datasets or mark claims as assumptions.
- Keep units and species context visible in tables and captions.
- Prefer effect sizes and confidence intervals over unsupported qualitative claims.