The test_mrtr_no_result_type probe in mrtr-client.ts deliberately omits resultType from its final result, and draft-result-fields.test.ts pins that omission as the check's stimulus. Problem is the referee only ever negotiates 2026-07-28 (server/discover returns supportedVersions: [DRAFT_PROTOCOL_VERSION]), and the released 2026-07-28 text scopes the absent-means-complete bridge to earlier revisions: "For backward compatibility with servers implementing earlier protocol versions, which do not include resultType, clients MUST treat an absent resultType as complete." The same spec says the result MUST include resultType, and the repo's own draft.schema.json has it in Result.required.
So the probe is an invalid stimulus on the very wire it negotiates, and a strict 2026-07-28 client fails sep-2322-default-result-type-complete for correctly rejecting a non-conformant reply. Downstream confirmation: cloudflare/agents baselines this scenario with the comment "Alpha.10 omits resultType after negotiating 2026-07-28. The final draft requires servers implementing that revision to include it; the absent-means-complete bridge applies only to earlier revisions."
Related: sep-2322.yaml still quotes the pre-final sentence without the version scoping ("If resultType is not specified, clients MUST assume a default value of complete for backwards compatibility"), which is probably where the check's shape came from. And the reason wire validation never flagged the probe is the inline-mock bypass from #418.
Following the #447 principle of running each revision's requirements at its own wire, I can see two fixes: gate the absent-resultType probe to runs where the referee negotiates a pre-2026-07-28 wire, or keep the 2026-07-28 wire and have the probe send resultType, retiring the default-check there. Since the omission is test-pinned as deliberate I did not want to just send a PR flipping it. Which way do you want this to go?
The
test_mrtr_no_result_typeprobe in mrtr-client.ts deliberately omitsresultTypefrom its final result, and draft-result-fields.test.ts pins that omission as the check's stimulus. Problem is the referee only ever negotiates 2026-07-28 (server/discoverreturnssupportedVersions: [DRAFT_PROTOCOL_VERSION]), and the released 2026-07-28 text scopes the absent-means-complete bridge to earlier revisions: "For backward compatibility with servers implementing earlier protocol versions, which do not include resultType, clients MUST treat an absent resultType as complete." The same spec says the result MUST include resultType, and the repo's own draft.schema.json has it inResult.required.So the probe is an invalid stimulus on the very wire it negotiates, and a strict 2026-07-28 client fails
sep-2322-default-result-type-completefor correctly rejecting a non-conformant reply. Downstream confirmation: cloudflare/agents baselines this scenario with the comment "Alpha.10 omits resultType after negotiating 2026-07-28. The final draft requires servers implementing that revision to include it; the absent-means-complete bridge applies only to earlier revisions."Related: sep-2322.yaml still quotes the pre-final sentence without the version scoping ("If resultType is not specified, clients MUST assume a default value of complete for backwards compatibility"), which is probably where the check's shape came from. And the reason wire validation never flagged the probe is the inline-mock bypass from #418.
Following the #447 principle of running each revision's requirements at its own wire, I can see two fixes: gate the absent-resultType probe to runs where the referee negotiates a pre-2026-07-28 wire, or keep the 2026-07-28 wire and have the probe send resultType, retiring the default-check there. Since the omission is test-pinned as deliberate I did not want to just send a PR flipping it. Which way do you want this to go?