Agents Commit Before They Can Be Corrected
The pillar. Why governance has to move to the point of commitment.
The runway
The pillar essay is post one. What follows alternates argument with proof, so the ideas stay varied and the demos land once their assets are clean.
The pillar. Why governance has to move to the point of commitment.
Action space first, governance bolted on after, and why that is the wrong order.
The type system and one inspected node, shown from the working kernel.
Hallucination of domain facts, and why retrieval is not enough.
The committed operation as the unit of audit, and why logs leave you doing forensic archaeology.
Governed double-entry, the case where ungoverned agentic action is obviously frightening.
Why "why is the system in this state" is unanswerable in most architectures, and why lineage has to be intrinsic.
The motor-control deep-dive behind the whole thesis.
The capstone: autonomy and governance as one boundary, not two systems joined by process.
Scaling, rising capability and retrofit as rearchitecture. The case for inevitability.