model-first-reasoning

Model-first reasoning is a methodology that constructs an explicit, frozen problem MODEL before any reasoning or implementation, reducing hallucinations and misalignment. By defining entities, state variables, and constraints upfront, it prevents bugs through strict validation.

To build a formal model for a state machine, you detail entities, state_variables, actions, constraints, initial_state, and goal in a MODEL_TEMPLATE.json format. You then write a model.json file and run a validator to proceed with implementation.

The best way to handle unknowns during code generation is to stop the process until they are resolved. The model-first approach requires freezing the model after a MODEL AUDIT, ensuring no new entities or constraints are introduced during the implementation phase.

Yes, model-first reasoning applies to tasks involving state machines, constraints, and multi-step workflows across software engineering and design. It validates operability and consistency through a MODEL AUDIT before freezing the model for implementation.

You should not use a formal model approach when unknowns exist that cannot be resolved, as the methodology requires stopping until they are clarified. It also restricts adding new entities or constraints during the implementation phase after the model is frozen.