C-D24-PRUNING

Reachable-state pruning 55.3% VALIDATED-BOUNDED: on the actual reachable mod-84 distribution (not random candidates), the fixed refinement test-time cuts at levels 14, 28 and 42 cut 55.3% of states and the mandatory p=197 witness is cut at level 14. These fixed-level cuts are genuinely effective but give only a CONSTANT-factor reduction; the residual scale for the current architecture (about 2^83) is a size indicator, NOT a proof that every exact representation must materialize that many states

In plain language

The fixed level cuts really do remove more than half the reachable states, but only by a constant factor, so the search is still too big for the current setup; that size is an indicator, not a hard floor.

Exact statement

Reachable-state pruning 55.3% VALIDATED-BOUNDED: on the actual reachable mod-84 distribution (not random candidates), the fixed refinement test-time cuts at levels 14, 28 and 42 cut 55.3% of states and the mandatory p=197 witness is cut at level 14. These fixed-level cuts are genuinely effective but give only a CONSTANT-factor reduction; the residual scale for the current architecture (about 2^83) is a size indicator, NOT a proof that every exact representation must materialize that many states