Stage 7: the randomised-graph control has been run

README.md

@@ -67,8 +67,10 @@ Both retained selectors land on update 16,600 here, so `min-ce.safetensors` and
 The comparison against the released reference is **not** controlled — its training
 stories and trainer are unknown. The controlled finding is that constraining the readout
 rank acts as a regularizer against overfitting 1,000 short stories; it says nothing about
-whether the fly's wiring is a good prior for language. That would need randomized-graph
-and zero-edge controls, which have not been run.
+whether the fly's wiring is a good prior for language. A randomised-graph control has since
+measured that directly: rewiring the connectome at random, with every degree and weight
+preserved, costs **0.0100 nats [+0.0026, +0.0175]** — about 1% of this architecture's margin
+over the reference. The wiring contributes something, and very little of it.
 
 The connectome is not duplicated here; it lives in
 [fly-connectome-49k](https://huggingface.co/datasets/fernandofernandes/fly-connectome-49k),