Evidence
Measured, not claimed.
"Governed AI" is a claim any vendor can make. Below is what OneAI Labs can actually show for it — the same figures OneForge's own control plane would return if you queried it yourself.
Snapshot, not a live feed
These are point-in-time figures from OneForge's control plane, refreshed by hand — not a real-time API connection. Building a genuinely live feed is real engineering work, tracked separately from this page, and we would rather show an honest snapshot than a fake ticker.
As of: 2026-09-24
Append-only. Nothing edited, nothing deleted.
A gate that never rejects is not a gate.
Reversibility proven, not promised.
Platform evidence model
Every layer proves a different fact.
A single green dashboard cannot prove an agent system. OneMission proves durable work state, the verifier proves outcomes independently, OneField preserves scoped evidence, and OneForge proves what was allowed to reach production.
Durable execution
Mission, task, lease, retry and approval history belongs to OneMission.
Independent verification
Signed verifier results are kept separate from the actor that produced the work.
Shared reality
Tenant-scoped context, consent and receipts converge in OneField.
Controlled evolution
Evaluations, release gates, canaries and rollback history belong to OneForge.
What one record actually looks like
The figures above aren't a marketing number. Here's a real one from our history.
This is a real training run from OneForge's control plane, not a staged example. The candidate scored below the production baseline. The gate rejected it. Production never changed. It predates a database reset, so it will not always match whatever the live count above happens to show right now — the counter is real either way.
RUN / QWEN3-8B
Candidate
0.5667
assistant-v6
Baseline
0.6167
production
Production version unchanged.
The model failed. The system didn't.
Execution log
The same discipline, applied to the whole platform
OneForge and TheOne both publish exactly which parts of themselves are verified, which are implemented but not yet accepted, and which are only planned — see their maturity ladders.
Your AI should not stop at a demo.
Route intelligence through Core, govern it with TheOne, persist the work in OneMission, execute through OneClaw and keep verified outcomes in OneField.