How it worksArchitectureScoreboardFAQVerify a binary
Developers

Documentation

Verify an AI report against a binary in one request, or let groundre write its own analysis. Everything is deterministic and every verdict carries evidence.

01Quickstart

Send a binary plus the claims an AI made about it; get back per-claim verdicts with evidence. safe_to_proceed is false if any claim is CONTRADICTED — gate an action on a single boolean.

curl -F binary=@suspect.exe \
     -F 'claims=[{"id":"c1","type":"address","cited_address":"0xdeadbeef"}]' \
     https://api.groundre.dev/v1/verify

02Two ways in

  • Verify a report — POST /jobs with a binary + an AI report (JSON/MD/PDF/text).
  • Analyze (no report) — POST /analyze; groundre writes its own multi-level safety report.
  • Investigate — POST /jobs/{id}/investigate runs the binary in a gated sandbox, then re-verifies every observed claim.

03Verdicts

Each claim is labeled by a type-specific verifier reading ground truth:

  • VERIFIED — proven against the binary (with the address/xref/section).
  • UNVERIFIED — could not be proven either way; withheld rather than guessed.
  • CONTRADICTED — proven false (e.g. an address in no mapped section).
No LLM decides a verdict. An LLM may reshape prose into typed claims; the judgment is pure code.
More for developers
Python SDKCLI referenceSARIF & STIX exportAPI status