Evidence over confidence
A convincing summary is not a result. What ships is judged by reproducible artifacts — tests, diffs, residue — not by how sure an agent sounds.
NEO treats each AI-generated change as a bounded workflow with evidence before acceptance.
Intent, authority, paths, protected files, budget, and risk are approved before execution begins.
The change runs inside the approved surface, with custody boundaries and snapshots preserved as it moves.
A separate review path produces findings, rerun output, diffs, and a re-auditable acceptance packet.
A human makes the final call. The decision can keep, remediate, reject, or roll back the change.
Code is cheap. A model can generate plausible changes quickly, but speed does not make a change safe to accept.
NEO governs intent and authority before a change starts, custody and boundaries while it runs, and evidence and acceptance after the work is complete.
The product is built around the decision, not the agent's confidence: what matters is whether a responsible human has defensible grounds to keep, remediate, reject, or roll back the result.
A convincing summary is not a result. What ships is judged by reproducible artifacts — tests, diffs, residue — not by how sure an agent sounds.
The party that writes a change never grades it. Building, judging, and auditing are different seats, so no one can move the bar to fit their own work.
Every change is scoped before it starts and reversible after it lands. Authority to accept stays with a human; nothing consequential ships on autopilot.
We label what is proven and what is not. NEO is proven in founder-operated development today; commercial proof is earned through paid evidence, not claimed in advance.
NEO's posture is frozen judges, independent audit, and reproducible artifacts. It does not claim absolute correctness; it creates a durable record for a bounded acceptance decision.