Evidence for
your data.

A check that says “pass” proves nothing.

Most data tools answer one narrow question — did this table pass? — and hide everything behind a boolean. When a pipeline breaks downstream, or an auditor asks what data a model was trained on, there is no record of what was checked, why it failed, or whether the exact dataset can be identified again. Data moves faster than the evidence about it — so bad data ships, and no one can prove what happened.

Profile → validate → diff, with proof.

import proofframe as pf

profile = pf.profile(df)                # canonical pf-fp-v1 fingerprint
report  = pf.validate(df, contract)     # bounded, row-level evidence
receipt = report.sign(secret_key)       # Ed25519 signed receipt

print(profile.fingerprint)             # pf-fp-v1:9a3f…e2
print(report.passed, report.violation_count)  # False  3

One engine you can trust in the pipeline.

7.6M real Bitcoin rows, validated in 1.742 s.

ProofFrame ran an exact, row-level contract over 7.6 million real Bitcoin market rows — the whole dataset, not a sample — and emitted a canonical fingerprint plus a signed receipt. Read the writeup, or run the notebook yourself on Kaggle.

Install it, run it, prove it.

ProofFrame ships to both registries under Apache-2.0 — a pure-Rust crate for Rust users, and ABI-stable wheels for Python.