Free Data Quality Scorecard

Seven yes/no questions. See how reproducible your AI or data workflow is and what to fix first. It runs in your browser and nothing you select is uploaded.

1. Run it twice

I ran the same input through the full workflow twice and the outputs were byte-for-byte identical.

2. Model steps

Any AI model output is schema-validated, stored, and never makes the final decision on its own.

3. Pinned versions

Models, prompts, libraries, runtimes, and reference data are pinned to exact versions, not "latest".

4. Time and randomness

Logic never reads the clock or uses unseeded randomness. Time and seeds are passed in as inputs.

5. Ordering

Data is sorted on a stable key before any order-sensitive step (dedupe, "first match", calculations).

6. External data

Inputs from live APIs, web pages, or shared sheets are snapshotted and hashed before use.

7. Evidence

Every run records input hashes, version info, output hashes, timing, and errors.

Fix first

Next steps

Read the full guide to the seven checks.

For a written, ranked report with specific fixes for your workflow, request a Deterministic Audit (from $499).

Email me this checklist