Does the brain's actual wiring at the learning site change how it ranks alloys?

EXPERIMENTS.md · lines 328–356

E10 — The wiring enters the causal path

Date 2026-09-12 · Question With the ranking stable (E9), does the specific measured wiring change it? · Provenance …/rewire_test.py

Method. Degree-preserving directed double-edge swaps confined to the KC->MBON block (61,210 edges). Each Kenyon cell keeps its out-degree onto MBONs, each MBON its in-degree from KCs, the contact multiset is preserved, and everything outside the block is untouched. 10 swaps per edge, five seeds. A global rewire was rejected as a control here because it destroys the mushroom body entirely, leaving nothing to compare against.

Result. 98.3% of block edges changed. Spearman(measured, rewired) over the 64 candidates:

Interpretation. Against seed-to-seed agreement of 0.9995 (E9), a rewiring effect of 0.868 is far outside the noise. Under the previous architecture, swapping to an entirely different-sized connectome moved the top selection probability by 1.2e-5; here, rewiring 61,210 of 25,582,938 edges measurably reorders the candidates.

The measured wiring is in the causal path. Most of the ranking (rho 0.87) still survives rewiring and therefore comes from degree structure, the encoder and the compartment readout rather than from who connects to whom.

Not claimed. That the wiring changes the ranking does not make the ranking better.