Send A Public Proof-Surface Test Case

The most useful examples are boring enough to reopen.

A claim was made. Some evidence exists. Some checks ran. Something changed. Some part of the story is still unsupported.

That is the test case.

Good examples

• A public repo change where an agent says a task is fixed, but the supporting checks are thin.
• A release note, benchmark, or demo claim that sounds cleaner than the evidence trail.
• A RAG or agent workflow where citations, tool calls, or retrieved chunks are not enough to support the final answer.
• A CI or eval run where tests passed, but the conclusion being drawn from the tests is too strong.
• A trace that explains what happened during a run, but not what a reviewer can safely believe afterward.

Do not send

• Secrets, credentials, keys, tokens, cookies, or private URLs.
• Private logs, customer data, client names, medical data, financial data, or personal identifiers.
• Exploit details, active vulnerability instructions, or anything that would create a disclosure problem.
• Anything that requires me to access a private account, private repo, private document, or private workspace.

Useful format

If you reply to a public thread, GitHub discussion, forum post, or social post, this shape is enough:

Claim:
What did the agent, repo, release, benchmark, or workflow claim?

Public evidence:
What public URL or artifact should be inspected?

Checks:
What check already exists, or what check should be replayed?

Weak point:
Where do you suspect the final summary is doing too much work?

Desired output:
Do you want a MATCH / DRIFT / UNVERIFIABLE packet, a critique of the packet shape, or both?

What I will try to return

• The claim, separated from the evidence.
• The checks that can be rerun without private access.
• The action or route state that matters for review.
• A narrow verdict: MATCH, DRIFT, or UNVERIFIABLE.
• The unsupported remainder left visible instead of hidden in polished prose.

Boundaries

This is not a safety certification, compliance claim, security audit, or endorsement.

It is a small public test of proof-surface tooling: can another person reopen the claim and see what the evidence actually supports?

Start here if you want the longer framing: The Summary Is Not The Record.

Use this tiny runnable example if you want to see the packet shape first: Crucible Cleanroom Verdict Packet Demo.