Epistemic review gates

Layer: assessment.

The review subsystem (note-gate pairs, freshness, acceptance state) supports new gates without new plumbing. Two flavors, kept separate:

Structural gates (deterministic, validator-checkable)

• Every claim has a source-span.
• Every contested claim shows at least one rebuttal edge.
• Every gap is linked from the claim it blocks.

Reasoning production is not reasoning evaluation argues the rebuttal check needs a decorrelated reviewer — don't let the pass that wrote the support also grade it.

Semantic gates (judgment, for the semantic review system)

Complementing the structural set — these are judgment calls, not deterministic checks:

• source-span-fidelity — does the claim accurately represent the cited span?
• scope-creep — a conclusion exceeds the population, timeframe, method, or source context.
• rhetorical-overweighting — prestige, vividness, or narrative force used as if it were direct evidence.
• missing-countercase — an assessment that does not represent the strongest serious opposing view.
• settlement-illusion — a debate called settled when only social/institutional closure has been shown.
• correlation-double-counting — evidence treated as independent when it shares data, authorship, method, or citation ancestry (see independence clustering).