What would prove this wrong?
The falsification number: under the exchangeability null the expected cross-class to same-class correction-exponent ratio is exactly 1.00, because architecture-blind averaging has no free parameter for composition class. The framework says materially above 1.00. Two predicted numbers for one measurable quantity. The ratio has not run on real data; the deciding registrations are drafted and awaiting human submission.
The load-bearing falsification number is the cross-class to same-class correction-exponent ratio. Under the exchangeability null its expected value is exactly 1.00, because architecture-blind averaging has no free parameter for composition class. The framework says materially above 1.00. Two predicted numbers for one measurable quantity, and the two predictions are as far apart as they can be: identity, or not.
An earlier candidate test, 'one number does two jobs, measured two independent ways, and the answers must agree', has been withdrawn in public as vacuous, before filing. The withdrawal carries its reason: if both exponents arise because accumulation is independence-limited, they share a law, agreement is forced, and a test that cannot fail is not a test. The replacement carries the unification instead: one accumulation law with two consequences, the conversion exponent and the correction ceiling, which asserts no independence and has nothing for the confound to bite on. The cross-class ratio on the register carries the load.
Standing bans apply. The book's printed R-squared is never evidence for the critical exponent being two; the book prints the square without deriving it, and the mechanism arrived eighteen months later from an author who knew the printed target. The cross-class ratio has not run on real data and is never described as measured. Preregistrations are drafted and awaiting human submission; nothing has been filed on the Open Science Framework yet. If the ratio runs and lands at 1.00, the framework has failed its stated condition and the record will carry the failure with the refuter's name on it.