Python Model Selection

Process

1. Check whether the right answer is actually no extra construct yet.
2. Choose Enum for closed symbolic values and named states.
3. Choose dataclass for structured in-memory data; prefer frozen=True unless mutation is intentional.
4. Choose ABC for explicit nominal contracts or shared base behavior; choose Protocol for structural compatibility.
5. Put branching examples, anti-patterns, edge notes, and split signals in examples.md.

Examples

• Positive: Use Enum for named states, @dataclass(frozen=True) for immutable structured data, and Protocol only when unrelated classes just need the same callable shape.
• Negative: Force ABC before there is any contract pressure, use dataclass for a behavior-heavy service object, or turn a plain boolean flag into Enum without a semantic reason.

Outputs

• a review-ready model-selection rule set or skill draft
• a construct-selection matrix with positive and negative patterns
• local examples for normal cases and edge cases

Boundaries

• Do not define naming policy, type-hint syntax policy, or control-flow rules.
• Do not define framework-specific schema or validation choices.
• Do not turn architecture-specific interface patterns into a universal Python rule.

Local references

• examples.md: construct-based examples, anti-patterns, edge notes, and split signals