Authority Transition Record

An agent’s activity log tells us what it did. It does not necessarily show what a risk-owner knew when deciding whether that agent should be allowed to do more next.

The separation that matters

An Authority Transition Record keeps two facts distinct:

1. Evidence at T0: material actually available when the owner made the decision.
2. Post-T0 outcomes: what happened later, which may inform the next review but cannot be relabelled as evidence available earlier.

Later success cannot retroactively justify an earlier decision. Later failure cannot be represented as something that was already known.

One record, one boundary

The format records a current boundary, a proposed future boundary, the accountable exposure owner, T0 evidence, known unknowns, the decision and later observations. Its most useful conclusion may be: the record does not support expansion.

What this is not

This is not a runtime permission system, an autonomy score, a certification, an insurance product, a performance guarantee or an instruction to deploy an agent. It is a working research artefact for making one decision inspectable.

The test

The format is useful only if a real decision owner says it makes a concrete authority-transition decision easier to inspect, compare or reverse. If it does not, it should be discarded or changed.

Open question

For organisations that already permit agents to act within constrained scopes: which part of a decision to widen that scope is hardest to reconstruct honestly?