Answers are grounded in this paper and cite their sources.
When does a community of instruments settle a question about the world? We give an exact answer for finite models. A target is settled, or identifiable, exactly when every pair of worlds it tells apart is told apart by some calibrated, lawfully exported public test. It is then a unique function of the pooled readings that occur. Separation is not calibration: relabelled reports separate the same worlds and can defeat a fixed decoder. Agreeing instruments with disjoint declared ancestries can share one blind spot, and one added test separating every relevant pair repairs it. When instruments fail by source groups, identifiability survives the silent failure of any b groups exactly when every target-relevant pair differs in at least b+1 groups; lying groups need 2b+1. The public partition refines only finitely often, but a pause proves nothing. Two communities can translate their readings exactly when they draw the same distinctions. A panel can settle every question put to it without settling all questions. With positive price floors, forcing a fresh test at every stage outruns every sublinear budget, and cross-validating every pair of n or more instruments at stage n outruns every linear budget. All results are formalized in Lean 4.
No takes yet. Share an insight, caveat, or question.
Ioannis Tsiokos (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: