Company
Blog
Notes on deterministic verification, AI hallucination in RE, and what we find running the pipeline at scale.
01Recent writing
Why an LLM can’t grade another LLM’s RE report
A model that hallucinates an address will happily hallucinate a reason it’s right — the errors correlate. We walk through why independence is the whole point.
Zero false flags, measured
How the false-flag rate stays at 0 on real binaries: CONTRADICTED requires proof a claim is false; everything merely unprovable stays UNVERIFIED.
Grounding runtime output to a static address
When the agent runs a binary, every printed line is matched back to a defined string in the disassembly — tying behavior to ground truth.
Full posts are on the way — subscribe from Contact to get them.
More about groundre