Prediction Partition Stability
Abstract
A prediction partition unchanged by one extra readout is unchanged at every depth.
Theorem 1.1 (A one-step stable prediction partition is permanently stable).
Proof. Machine-checked in Lean as D5/S3/ObserverMemory/Prediction/PredictionPartitionStability.prediction_partition_stable_forever (✓ std3). ∎
Source. Repository-derived.
Commentary.
For a self-map F and readout q, ReadoutWord(F,q,m,y) records the readouts of y at update times zero through m. The hypothesis says that equality of these words is exactly the same relation at depths m and m+1.
The first conjunct proves that this depth-m relation is preserved when both states are updated by F. Iterating that congruence makes every later readout agree, while truncation gives the reverse implication. Thus the relation at every depth m+r equals the relation at depth m.
Repository search found the exact finite-word definition but no theorem containing both conclusions. Pinned Mathlib and Loogle found Function.iterate_add_apply, which the proof applies to shift readout coordinates. LeanSearch’s shaped endpoint returned HTTP 404 and supplied no result.
The theorem is general in both types and does not require finiteness. A constant Boolean readout gives a checked witness that the stabilization hypothesis is satisfiable on a nontrivial state carrier.
References
- Truth anchor:
D5/S3/ObserverMemory/Prediction/PredictionPartitionStability.prediction_partition_stable_forever - Dependency: D5/S3/ObserverMemory/Prediction/ConditionalEntropyStability