Machine Learning
When to use
Use this skill for machine learning when you need to select a model for a measurable prediction task. The expected result is a baseline, experiment plan, and error analysis.
Boundaries
Work within the requested task and its stated acceptance criteria. Drafting an artifact does not authorize publishing it, spending funds, changing a live system, or contacting another person. Identify any such action separately before taking it.
Inputs
Inspect the task, dataset provenance, labels, baseline, model constraints, latency, and cost. Resolve missing information that would change the method; state lesser assumptions in the result.
Method
- Diagnose. Define task, data rights, labeling quality, baseline, latency, cost, and the consequences of a wrong answer.
- Decide. Select an approach only after defining an evaluation set and failure costs.
- Produce. Build a baseline, experiment plan, and error analysis from the inspected material; keep assumptions distinguishable from observed facts.
Decision rules
- Reserve holdout cases before tuning. Compare model and non-model approaches and inspect error slices, not only averages.
- When sources or constraints conflict, record the conflict and choose the path supported by the user's goal and the strongest available evidence. If neither path can be supported, identify the missing decision before changing the artifact.
Domain rules
- Separate training, validation, and test data to avoid leakage.
- Report uncertainty, safety limits, and the cases where the model should abstain.
Verification
Test leakage and class imbalance. Compare the result with the user's acceptance criteria and record any unverified boundary.
Provide an evaluation set, baseline comparison, failure examples, abstention rule, and monitoring trigger.
Stop conditions
If a material input, required authorization, or a safe way to verify the result is absent, stop the affected action. Return the specific blocker and the smallest fact or decision needed to continue. Do not report an unrun check as passed.
Output
Provide a baseline, experiment plan, and error analysis. Include the decisive evidence and actual verification result. Name any artifact location and unresolved issue that affects its use.
