Corey James- 5conditions a tool must meet before it can be synthesised
- 8fields in every evidence-ledger entry, lineage among them
- 4result classes, assigned from observed evidence
- 6layers of governance around each action
Methods and Systems for Evidence-Gated Synthesis and Governance of Agent-Callable Application Programming Interface Tools, and for Reconciled Execution of Mutations Against Ephemeral Third-Party Application Programming Interfaces. U.S. Provisional Patent Application No. 64/148,105 (patent pending). Embodied in the API integration platform of FAFO Core and Function Forge.
Unconstrained AI invents tools or actions, acts outside intent, returns unverified results, and is hard to audit. This invention puts every action through one governed pipeline, from the intent that asks for it to an observation that must prove its result.
A tool only with positive observation
A tool may be synthesised only when five conditions hold. The one that does the work is the second: for every output field of the tool there must exist an observation in the evidence ledger, so a field the system has no way to observe blocks the tool.
1 every output field is defined · 2 every field has a positive observation method · 3 the action is within authorised scope · 4 policy checks pass · 5 the tool is registered and versioned
Output field to observationmessage_idstatusrecipientThe execution contract
Each action runs inside an execution contract that fixes what can be accessed, what can be modified and how results must be reported.
No action can run outside the contract, and the AI cannot change the rules.
An evidence ledger with lineage
Every action produces an entry in the evidence ledger: what was attempted, what happened and how it was observed, linked to the entry it came from.
Results from evidence, not model claims
Results are classified from observable evidence; what the model says about its own work plays no part. A work item counts as complete only when every one of its required output fields has a positive observation.
Retry within constraints, or a person
When verification fails the system retries within constraints or escalates to a person or another agent. Every failure is logged with its evidence and a reason code, and unverified results are never marked complete.
Governance in nested layers
Layered governance enforces safety, compliance and control across all actions, from the policy that says what is allowed to the audit that proves what happened. The guarantees: no unauthorised actions, no unobserved outputs, no silent failures, full auditability and replay.