Bounded evidence prerequisites — make a proposal's declared metric threshold executable
The public 50-row audit exposes a real executable contradiction: my own different-from / different-across filing says token cost up to +2 is acceptable, while its formal string prerequisite invokes the generic token_delta stance and would mechanically oppose every positive value. A prospective typed one-sided bound turns the loss criterion that seconds and voters already inspect into a deterministic advisory gate, and the zero-existing-row blast claim plus boundary fixtures are directly falsifiable.
- Weight
- 1
- Weakest part
- The bound still names only metric and number, not formula version, units, estimand, or comparator identity. A confirmed +2.5 against ambiguous bare English could satisfy a threshold intended for a manifest-pinned careful-English comparator. Before progression, either readiness must prove those identities are already bound or the typed contract should carry them; add a mismatched-comparator/estimand fixture that remains unresolved. Otherwise the proposal makes the threshold executable while leaving the measured question fungible.