Acceptance is not a grant
Why accepting a decision would never start work on its own, and what would have to happen between a yes and a run.
Design noteThis note describes how Assent is designed to work. Intent, not results.
- 01Evidenceevidence informs a decision
- 02Acceptanceowners accept a revision of it
- 03Publicationpublication records the accepted revision where the code lives
- 04A grantonly then can a grant authorize a run
Four things that are easy to blur
Evidence, acceptance, publication and permission to execute are separate in Assent’s design, on purpose. It would be simpler to treat an approval as the start signal. It would also turn prose, or a single yes, into open-ended permission.
So each step would be its own record: evidence informs a decision, owners accept a revision of it, publication records the accepted revision where the code lives, and only then can a grant authorize a run.
Source: Decision and assumption register
What approval binds to
An approval is designed to bind to exact content: the decision revision, its scope, the person approving and the policy and authority versions in force. Change a word and the approval would not carry over.
That would hold on every surface. Binding approval would need the exact revision on screen and a deliberate, authenticated action; a spoken confirmation alone would never approve. Agents and AI representatives would not be able to approve at all.
Source: Authority and enforcement
Publication comes before permission
The accepted decision, with a snapshot of the target contract, would go through Git review. A trusted reconciliation step would tie the merged artifacts to the exact reviewed revision; changed content or missing approvals would block it.
An implementation grant would require a committed publication for every decision it depends on. Even then, acceptance and publication wouldn’t grant execution by themselves.
Source: Product workflow
A grant is narrow, and it stays live
After publication, Assent would request a stage on the initiator’s behalf. Policy would check the actor, the decision revision, current policy, authority, scope and required controls, and issue a revocable grant only if every one holds.
The runner would never hold the grant as a key. The handoff it receives would only refer to it, and before each governed action the runner would check the grant’s live state. A copied handoff would authorize nothing.
Source: Product workflow
What this costs, and how we’ll know
More steps between a yes and a run is a real cost. The design bets that the separation is worth it; the first evaluation will count approval time as effort alongside everything else, so the cost will be measured rather than assumed away.
Source: Metrics and economic evidence
Design partners
Bring one recent cross-team change.
We reconstruct it with you before showing a solution: the decisions, their owners and the effort, agent operation included.