You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
feat(Conformance): leave the #278 include / sort refusal rules not decided on a probe that says nothing #352
The two #278 refusal rules (spec/query/include-unsupported-endpoint, spec/query/sort-unsupported-endpoint) and the seven #310 rules now share one probe driver (probeFirstUsable in src/internal/conformance/spec.ts), but they treat a probe that says nothing differently:
The feat(Conformance): probe include / sort on an endpoint that does not declare them (#261) #278 rules turn a probe that got no response into a finding: "?include=author on /articles/1 got no response". src/Conformance.test.ts pins this ("says a probe that got no response got none, not that it was not rejected"). A probe answered 5xx, 429 or 408 is a finding too, as "was not rejected".
#310 kept the #278 behaviour (unanswered: "finding") so that it would not change existing rules' outcomes outside its scope.
Proposal
Switch the two #278 rules to unanswered: "undecided". This is a one-line change per rule, plus updates to the pinned test and to the docs/conformance.md sentence that names the difference.
This is an owner decision: should a 500 for ?include= on an endpoint that does not support it be a failure (the server did not answer 400) or not decided (the server's answer says nothing)?
If the answer is "undecided", both rules return not-decided on a 5xx, 429, 408 or lost response, the test and docs/conformance.md are updated, and a changeset is added (patch).
Activity
thomasfosterau commentedon Oct 15, 2026
Owner decision, 2026-10-15 (session chat): a 5xx is a finding; 429, 408 and a lost response are not decided.
This applies to all nine refusal rules that share
probeFirstUsable, so they stay consistent: the two from #278 and the seven from #310.For the #310 rules this is a change: they currently leave a 5xx not decided. The changeset should say so.
This is queued behind PR #366, because both touch the kit's tests and
docs/conformance.md.Generated by Claude Code