Memory

How has this person decided this before?

Answers two different questions, and keeps them apart on purpose.

likelihood is an observation about a person — how they have decided comparable requests before. Comparable means the same fingerprint bucket, not the same action type: three approvals of $5 say nothing about $500.

suggest_auto is a statement about a policy. It is true only where a rule the operator already wrote would auto-approve this exact case. It is never derived from likelihood, and a confident model never becomes new authority. It is false, with blocked_reason, when:

  • no rule covers the action (no_matching_rule);
  • a rule says a human decides (rule_requires_approval);
  • the amount is above the rule's own threshold (above_configured_threshold);
  • the derived risk tier is above the lowest (risk_tier_requires_step_up);
  • the action grants or destroys authority (action_is_sensitive).

The policy consulted is the calling agent's own, so a prediction cannot inherit authority from another agent that happens to observe the same person.

post/v1/peers/{peer_id}/predict-approval

Request

  • Base URL: https://api.1claw.co
  • URL: https://api.1claw.co/v1/peers/{peer_id}/predict-approval
  • Auth: HTTP bearer

Path parameters

peer_idstring uuid required

Request body

action_typestring required
payloadobject
effective_risk_tierinteger required

Example request

{
  "action_type": "refund.create"
}

Response

Prediction

likelihoodnumber nullable

Absent when there is no comparable history.

reasoningstring required
suggest_autoboolean required

Whether the operator's own policy already permits this.

blocked_reason'no_matching_rule' | 'rule_requires_approval' | 'above_configured_threshold' | 'risk_tier_requires_step_up' | 'action_is_sensitive'

Changes

No recorded changes to this operation across all 2 revisions of this API.