fix(sep-2322): stop two scored scenarios going green on a server that rejects everything

[AmirK-S]

Summary

Against a server that answers every request with -32000 Bad Request: Server not initialized, two scenarios in the 2026-07-28 requirement set score fully green on main (#451):

• input-required-result-validate-input, 3/3: both checks accept any JSON-RPC error as proof that inputResponses were validated.
• input-required-result-unsupported-methods, 2/2: its MUST NOT is never exercised when tools/list and prompts/list are both rejected.

This PR adds a control to each, following the #248 policy:

• validate-input first calls test_input_required_result_elicitation without inputResponses and requires an InputRequiredResult, as A1 already does. If that fails, both checks are reported with untestableCheck at their existing WARNING severity.
• unsupported-methods requires at least one probe to return a result. A server without prompts can still reject prompts/list. If neither returns a result, the check is reported with untestableCheck at FAILURE.

The reason names what the server returned, for example JSON-RPC error -32000: .... The catch path of validate-input now emits both check ids, so a throw no longer drops sep-2322-error-on-protocol-error.

Scope

This does not add the expected-vs-emitted count that #451 stays open for, nor the checks.json persistence on throw that @mohammedmessaoudene is taking.

Tests

• New input-required-result-reject-all.test.ts: a local HTTP server that returns -32000 for every request. Both tests fail on main (expected 'SUCCESS' to be 'WARNING', expected 'SUCCESS' to be 'FAILURE') and pass here. It is a separate file so it does not collide with fix(scenarios): stop MRTR retries when no input is requested #498 in negative-mrtr.test.ts.
• CLI, --requirements 2026-07-28 against that server: 2 fully green scenarios on main, 0 on this branch.
• Against the everything-server, both scenarios give the same results as on main.
• npm run typecheck, npm run lint and vitest run are green (48 files, 624 tests).

Refs #451

AI disclosure

Per AI_POLICY.md: this change, its tests and this description were written primarily by Claude Code under my direction, and I reviewed the measurements. Review replies may be AI-assisted as well.

🤖 Generated with Claude Code

[pkg-pr-new]

Open in StackBlitz

npx https://pkg.pr.new/@modelcontextprotocol/conformance@537

commit: 5c63aa1