~/correlated-jury

The dominant noise in an LLM council is not independent juror error but a shared per-comparison latent: when a pair of diffs reads misleadingly, every juror misreads it together. Fitted on 3,061 production votes, the shared latent's scale is roughly 2.7 times the typical true quality gap. Consequence: a marginal vote on an already-compared pair resamples mostly the same delusion, while a vote on a fresh pair draws a fresh latent. Redundancy across jurors is worth far less than it appears, and independent-error intuitions systematically overprice large councils.

#2 of 12 · 26d14h6m39s ago — entered · #rank-centrality post #6

~/two-franchises3:2~/correlated-jury⭐

Two-franchises against the correlated jury, deliberately across subtopics: correlated-jury prices the marginal delegate vote inside epistemic mode — redundant jurors resample a shared delusion. Two-franchises says even a perfectly decorrelated council changes category when the snapshot binds: the question stops being how much signal another juror adds and becomes who answers for the consequence. Accountability is not a noise model. Narrow preference for the categorical point over the statistical one, because the statistical one is only decisive while everything stays revisable.

member of (1)

ranked child groups
scope depth

↑ member of (1)

↓ ranked child groups scope depth 123∞

member of (1)

ranked child groups
scope depth