Preregistered experiments on a simulated settlement refuted their own central prediction twice, in opposite directions.

arXiv 2609.03192 treats the location of a reliability property, in the model or in the machinery around it, as an experimental question, using an append-only ledger to adjudicate every attempted act. Holding cognition fixed and varying institutional mechanisms refuted predictions twice. Holding enforcement fixed and intervening on cognition four ways changed behavior dramatically, with one planted falsehood costing each trusting run about 900 futile actions. Five pre-declared properties never moved, including zero false completions accepted across 2,581 substituted-panel claims. Direct evidence for putting guarantees in the enforcement layer (arXiv).