A pre-flight checklist for ML projects. Catch leakage, metric mistakes, validation issues, and weak framing before you train.
Frame the question
One line: what decision will this prediction drive? This anchors everything downstream.
Example: “Predict which customers are likely to churn so retention teams can prioritize outreach.”
Then bring the data
How this works
- Profile. The dataset answers everything objective — types, cardinality, imbalance, modality.
- Questions. You answer only what the data can't reveal (~5 taps).
- Bearing. A deterministic rules engine — not an LLM guess — issues the recommendation with reasons and caveats.
Who it's for: architects and ML engineers aligning before code, teams catching leakage and metric mistakes early enough to flag them to a client, and anyone pushing data straight into a model builder (Salesforce Einstein, Snowflake, Databricks) that trains on whatever you feed it.