AI ML Problem Framing

Operation

Return an actionable plan for the requested outcome.

Goal

Determine whether the requested AI/ML/DS behavior is a well-defined product or decision problem before selecting a model, dataset, retrieval system, or evaluation stack.

Framing method

1. Define actor, decision/action, unit of observation, prediction/generation time, allowed context, desired outcome, and unacceptable failure.
2. Name the simplest non-ML/manual/rule baseline and why ML or a model is warranted.
3. Define target/label or acceptable output, feedback delay, data availability at decision time, and leakage boundary.
4. Specify offline metrics, slices, uncertainty, thresholds/abstention, latency/cost/privacy/safety constraints, and online product outcome.
5. Choose human-in-the-loop, fallback, monitoring, rollback, and learning/feedback ownership.
6. State feasibility blockers rather than inventing data quality, model capability, or production volume.

• Ground repository facts with exact path:line[-line] anchors plus symbol when possible.
• Separate desired business outcome, observable user behavior, statistical objective, operational constraints, and assumptions.
• Compare the model approach with a deterministic/search/manual baseline. Preserve an explicit research or model choice; explain its tradeoffs rather than silently replacing the user's objective.
• Do not invent data availability, labels, consent, latency, cost, or deployment guarantees.

Handoff

Return the decision contract, decision-time data boundary, baseline, acceptable output/failure, evaluation design and unresolved feasibility gates. Feed the selected feature/pipeline workflow only after the user requested implementation; framing alone does not select a provider, purchase compute or start training.