Independence clustering of evidence

Source: imported from the ChatGPT second opinion. Probably the highest-value import — this menu was otherwise silent on correlated evidence, which FLF explicitly calls out.

Tag each evidence item with an independence_cluster and its shared dependencies (dataset, method, authorship, citation lineage), so reviews and dashboards can flag "12 items, 3 plausibly independent clusters." Ten papers are not ten pieces of evidence if they share the same dataset, method, assumptions, or citation lineage.

It is the evidence-side analogue of the reviewer-side logic in error correction works with above-chance oracles and decorrelated checks. Its link-level expression is correlated-with / independent-of edges (see dialectical link vocabulary); its review-level expression is the correlation-double-counting gate (see epistemic review gates).