Decifact
DECISION VERIFICATION LAYER
A decision is defined by its canonical content identity. Verification operates on the decision itself — not execution traces, receipts, or system-specific interpretation.
There is no shared method to determine whether two independently produced outputs represent the same decision.
Execution traces, logs, and outputs are not sufficient to define decision identity.
Without a stable invariant, cross-system verification is impossible.
Two systems can execute differently and still represent the same decision.
An output is a result. A decision is a structured artifact with invariant content.
Not execution traces. Not receipts. Not system-specific interpretation.
Execution details, evaluation methods, and context may differ if explicitly declared.
This establishes a stable boundary for comparison across systems.
A minimal verification interface for evaluating decision equivalence.
Designed to operate independently of execution systems.
This implementation is a reference realization, not the definition authority.
POST /verify
{
"decision_a": "...",
"decision_b": "..."
}
→ returns: equivalent | not equivalent
Ensure agents agree on the same decision even when execution differs.
Compare outcomes across heterogeneous models and runtimes.
Verify decision identity without replaying full execution paths.
Maintain consistent decision boundaries across distributed environments.
We are evaluating real-world use cases for decision verification.