quorum-judginglisted

# Judging and synthesis By this point the dossier is confirmed. Judging is not "check it again" — it is **deciding what to do about it**, under the constraints that actually govern. And synthesis carries the single most important rule in the pack: > **The only family-invariant across ~21 runs is an explicit dissent-honoring convergence step.** It > survives every specimen, including judge-less and verify-less runs. Everything else — phase count, > verify presence, judge count (0 to 3 observed), evidence mechanism — varied. --- ## Judges are constraints, not lenses A lens asks *how could this fail?* A judge asks *would this pass the thing that actually kills proposals like this?* Name the **2-4 hard constraints that would kill or approve this class of decision**, from the subject's real stakes. | Subject | Judges (constraints) | Not judges (those are lenses) | |---|---|---| | A process audit | automation architect · token economist · reliability judge | "thorough reviewer" | | A shipping decision | user impact · rollback cost · on-call burden | "QA" | | An architecture choice | migration cost · operational complexity · reversibility | "senior engineer" | | A board deliverable | credibility with the audience · defensibility of every number · does it prompt the right decision | "editor" | **If a proposed judge would have made a good miner, it is a lens.** Move it. --- ## Zero judges is a real answer The corpus observed **0 to 3 judges**. Use zero when nothing is cont