Validation methodology

Three rules govern the framework. Primary validation uses independently confirmed reference ages, not ages from the same model. Corrections are tested out-of-sample by leave-one-out, since a fit that only improves its own points is a curve fit, not a clock. Every claim is tagged [F]/[I]/[A], so confirmed facts are never merged with inference or assumption.

Three rules govern the framework. (1) Primary validation uses independently confirmed reference ages, not ages from the same model. (2) Corrections are tested out-of-sample by leave-one-out, since a fit that only improves its own points is a curve fit, not a clock (out-of-sample error ≈ in-sample error is the test). (3) Every claim is tagged [F]/[I]/[A], so directly confirmed facts are never merged with chain-dependent inference or assumption.

The framework is held to three rules, applied throughout this paper. First, primary validation uses independently confirmed reference ages—historically attested for radiocarbon, and for zircon the published, cross-anchored eruption ages—rather than ages produced by the same model. Second, corrections are tested out-of-sample (leave-one-out), because a correction that only improves the points used to build it is a curve fit, not a clock. Third, every quantitative claim is tagged [F]/[I]/[A], so that directly confirmed facts are never silently merged with chain-dependent inferences or with assumptions. Strong, independently convergent evidence is weighted above long or self-referential chains, and single-source reasoning is flagged where it occurs.