mesokit-schema

Shared data contract for the mesofield → datakit → databench pipeline.

This package defines the typed manifests that travel alongside data between the three repos. It is the single source of truth for:

• AcquisitionManifest — what mesofield writes at the end of a session: the list of producers that ran, the files they wrote, calibration constants, time bases, and session identity. Datakit consumes this instead of globbing the filesystem.
• DatasetManifest — what datakit writes alongside the dataset table: schema version, datakit version, content hash of the data file, hash of the upstream acquisition manifest, master TimeBasis, source-parser versions, declared columns. Databench validates against this on load and concatenates it into provenance.
• AnalysisDeclaration — what each databench analysis declares about its inputs and parameters, so a dataset can be validated against an analysis before it runs.

Plus shared building blocks: TimeBasis, ColumnSpec, BIDS path helpers, and a single SCHEMA_VERSION constant.

Install

From PyPI:

pip install mesokit-schema

From GitHub (latest main):

pip install git+https://github.com/Gronemeyer/mesokit-schema.git

Development

For local development across the pipeline, editable-install into all three envs at once:

./scripts/install-dev.sh                  # all envs
./scripts/install-dev.sh mesofield        # one env

The script editable-installs mesokit-schema plus the env's own package (mesofield, datakit, databench) into each matching conda env. Override layout via CONDA_ENVS_ROOT, DEV_ROOT, or DATABENCH_ROOT env vars.

Or install manually:

pip install -e /path/to/mesokit-schema

Versioning

The package exposes mesokit_schema.SCHEMA_VERSION (semver). Consumers compare against the schema_version field on each manifest and refuse to load anything they don't know how to read. Bump the major version on breaking changes; minor on additive fields.