TrueVectors
TrueVectors is the instrument behind the public audit record — the same neuro-symbolic engine that produced every published finding. Authorized enterprise evaluations are delivered CI-first; public software distribution is closed. The TrueVectors Auditor checks an agent's claims against the realized result and proves what it can; the TrueVectors Hub keeps every verdict — proved, corroborated, or model judgment — in a durable system of record.
InputAgent outputcode · data · a recorded session
The instrumentTrueVectors Auditor
Neural proposesSymbolic proves
checks claims against the real resultOutputEvidence-stamped verdict
The recordTrueVectors Hubdurable, queryable system of record
TrueVectors Auditor
Proves what an agent actually did
Point the Auditor at code, a data artifact, or a recorded session. A neural proposer reads intent; a symbolic engine proves or refutes it against the realized result — and can overrule the model outright. Every finding is stamped proved, corroborated, or model judgment, reported misses-first, with a plain-English verdict and an exposure-gated "value of this audit". Explore the Auditor →

Every verdict traces to the lines that disagree
Here a migrated COBOL→Python interest job breaks its legacy control total by $0.26 — and the Auditor localizes it to a single line: the port rounds ROUND_HALF_UP where the mainframe truncates. A defect that passed code review and unit tests, proved deterministically against the real output. No confidence theater. See public audits →

TrueVectors Hub
A durable system of record
Every audit the Auditor runs is persisted to the Hub — what the agent claimed, what was proved, what your team confirmed or dismissed. A behavioral audit trail across your whole fleet that survives model upgrades, stays on your machine, and proves diligence when someone asks. Explore the Hub →

See it prove something
TrueVectors is an independent research project in early access. Reproduce the public evidence, or request an authorized evaluation.
TrueVectors verifies what your agents actually did. Trust through proof.
