Python packages

Access needed

The packages and notebooks are not yet publicly distributed. The steps below need a checkout of the UMP repository. Pages show recorded outputs, so you can follow along without running them.

Six packages carry the UMP contract into Python. Analysts mostly need two of them. The rest sit underneath, or serve the apps.

The prose and the published JSON Schemas are normative; the Python packages are informative reference implementations. Where they disagree, the spec wins and the package has a bug. The schemas page lists the published files.

Pick a package

Package	Import	What it is for	Who uses it
babelfin	import babelfin as fin	The analyst front door: estimands, numbers with intervals, gate checks, models, workflows and the core API from a notebook.	Analysts in notebooks and scripts
finprint	import finprint as fp	Records pandas prep as a transform recipe beside the output file, then verifies or reapplies it.	Analysts preparing data
ump-workflow	import ump_workflow as uw	Saves a fitted notebook model as a hashed bundle of data and source, and runs it on new data. Can serve one bundle over HTTP.	Analysts sharing models, and whoever runs them next
ump-spec	import ump_spec	The contract itself: Pydantic models, JSON Schemas, rule functions, the gate catalogue and Merkle maths.	Tool builders and every other package
ump-intake	import ump_intake as ui	Turns a client delivery (CSV, XLSX, connector rows) into a canonical weekly panel and ties it to the printed totals.	Data onboarding services
ump-sdk	import ump_sdk	Typed HTTP clients and a FastAPI app factory with the house conventions.	Service and adapter authors

How they fit together

Start here babelfin One flat namespace over the contract. Results render as cards in Jupyter. Data prep finprint No dependency on the rest, so it installs in any notebook. Share a model ump-workflow Never pickles. Code runs only after someone trusts its hash. Underneath ump-spec Pure contract code with no I/O beyond its own registry files.

Babelfin wraps ump-spec and calls ump-workflow for bundles. Finprint stands alone; babelfin picks it up through an optional extra. Ump-intake and ump-sdk serve the apps and adapters, and babelfin uses the SDK clients when it talks to a running core.

Install from the repository

The packages are not on PyPI yet. Each one is its own uv project (Python 3.12, own lockfile). Install editable from a checkout:

pip install -e 'packages/babelfin[recipe,plot]'   # pulls in ump-spec; recipe adds finprint
pip install -e 'packages/finprint[notebook]'      # finprint alone, with the IPython extension

Quote the extras in zsh, which treats [...] as a glob.

Next

• Work through the Learn notebooks, which use babelfin and finprint end to end.
• Look up exact signatures in the API reference.