Published
nomue Protocol opens public discussion on an exact-value Welch successor
Public comment is open on a proposed bounded successor for Welch checks, with exact-value evidence, explicit domain limits and a Contract-bearing Record format.
Public comment is open on a bounded Welch successor
nomue Protocol has opened public discussion on a proposed successor for exact-value Welch checks. Researchers and developers can read the proposal, inspect its public record and fixtures, and comment on Issue #372.
The proposal is intended to make a claimed Welch result easier to check without treating rounded binary64 calculations as the authority for its numerical meaning. It is a discussion draft: it does not change Release 1, add verifier support, adopt a Protocol version, or authorize issuance.
One numerical authority, carried by an inspectable Record
The proposed successor places the numerical meaning of a Welch claim in one versioned semantics document. A separate Contract identifies the exact numerical rule, and the Record carries the identity, evidence and report fields needed for an independent checker to determine what it can check.
This separation is deliberate. The Contract and Record identify the claimed procedure; the numerical semantics define the quantities. A checker would preserve independent outcomes rather than turning a tool failure or resource limit into a statistical failure. The proposal describes the candidate interface and its tests, not a currently available verifier.
A narrow domain with visible limits
The initial proposal is deliberately bounded: it covers supplied independent two-group Welch Records with finite, supported quantities under the stated group-size and magnitude limits. Its exact checks use the supplied binary64 values as evidence, while retaining a defined tolerance only where the proposal calls for a rounded reported probability.
Records outside that domain are not reclassified as statistical failures. The proposed handling keeps the reason visible, including domain exclusion, computation error or resource limit. Nonfinite computation is outside the initial domain; it is reserved for a later, explicit extension rather than being silently folded into a pass or fail result.
Release 1 remains unchanged
Release 1 remains the current public Protocol and verifier scope. The proposal records historical differences rather than claiming to reproduce or revise Release 1 behavior. In particular, some inputs that Release 1 could pass would be unsupported by the proposed successor's narrower initial domain. That is a disclosed boundary, not a retroactive judgment about an earlier Record.
The public discussion therefore does not alter how to use the current Release 1 Welch Record verification. The capability map remains the source for what is available today.
What feedback is useful now
The proposal includes representative Records, report examples and machine-readable bindings so reviewers can test the boundaries between record structure, numerical semantics and reported results. Comments are especially useful on the stated domain, exact-value comparison rules, reason codes, evidence identity, and whether the proposed limits are clear enough to implement and review independently.
This public discussion follows preparation and review work, but it does not close the remaining Research Gate or design holds. Any candidate changes still require review before a freeze, adoption, publication or issuance decision.
How to participate
Comment on Issue #372 with the passage concerned, a counterexample or test case, and a proposed correction where possible. Link to reproducible public evidence when it helps assess a numerical or interoperability claim.
The discussion opened at . The earliest decision is . That date is not automatic adoption, a Protocol release date, or an availability commitment.