Name: wisp
Author: Kar

Question 1

Does wisp use an LLM to judge whether a claim is true?

Accepted Answer

No. Every check is a deterministic call to GitHub's own public REST API: does a claimed file exist, does a claimed companion gate/test file exist, does a claimed cross-reference file actually contain a claimed phrase. There is no subjective judgment anywhere in the verdict.

Question 2

What exactly does wisp verify, and what doesn't it verify?

Accepted Answer

It verifies that three specific, named claims about a PUBLIC GitHub repo are real: a source file's existence, a companion gate/test file's existence, and whether a named file genuinely contains a named phrase. It does not verify that code is correct, that a claim's substance is true, or catch every form of confabulation -- only these three checkable facts, scored honestly as PASS (3/3) or FLAG (anything less), with the failing checks named.

Question 3

Why GitHub's public API instead of checking local files?

Accepted Answer

A static page cannot read an arbitrary local filesystem, and trusting a claim's own author to self-report local state defeats the purpose. GitHub's public REST API is independently queryable by anyone -- the check is re-runnable by a third party, not just trusted.

wisp

What this actually checks