Predictive Completion Monotonicity
Abstract
Predictive completion preserves the refinement order of readouts.
Theorem 1.1 (Predictive completion preserves refinement).
Proof. Machine-checked in Lean as D5/S3/ConceptDynamics/Sufficiency/PredictiveCompletionMonotone.predictive_completion_monotone (✓ std3). ∎
Source. Repository-derived.
Commentary.
The premise says that equality under the finer readout r implies equality under q. This is exactly inclusion of the two current readout kernels.
The all-iterate congruence kernel is monotone under relation inclusion. Quotient equality for r therefore yields the corresponding kernel relation for q, and the quotient soundness theorem concludes.
No inhabitedness, finiteness, or update assumptions are used; the empty, singleton, constant-update, identity-readout, and zero-step examples are checked in the Lean module.
Lemma 1.2 (The refinement premise is necessary).
Proof. Machine-checked in Lean as D5/S3/ConceptDynamics/Sufficiency/PredictiveCompletionMonotone.refinement_hypothesis_is_necessary (✓ std3). ∎
Source. Repository-derived.
Commentary.
On Bool states with identity dynamics, a constant Unit readout makes the two quotient classes equal, while the identity Bool readout keeps true and false distinct.
Thus the conclusion fails when the relation-inclusion premise is removed. This concrete counterexample is the required audit of the only non-definitional hypothesis.
References
- Truth anchor:
D5/S3/ConceptDynamics/Sufficiency/PredictiveCompletionMonotone.predictive_completion_monotone - Truth anchor:
D5/S3/ConceptDynamics/Sufficiency/PredictiveCompletionMonotone.refinement_hypothesis_is_necessary - Dependency: D5/S3/ConceptDynamics/Sufficiency/MinimalPredictiveCompletionQuotient