Kenshiki

Emission gate

Boundary Gate

Keep priors from leaking past proof.

The last separation between fluent generation and claims the system is willing to let reach a user. The Gate reads the Claim Ledger output and makes the release decision — not based on model confidence, but based on what the evidence supports.

Without this: the Ledger scores claims but nothing enforces the boundary. A response with one unsupported claim still reaches the user whole. The scoring becomes telemetry, not governance.

How the Gate works

Reads the Ledger's per-claim evaluation and makes a single deterministic decision about the entire response. Does not re-evaluate claims — enforces the boundary the Ledger defined.

  • Reads per-claim scores, attribution, and gate reason codes from the Ledger
  • Applies the output-state taxonomy based on aggregate claim support
  • Assigns one of five states before the response reaches the caller
  • Deterministic — same Ledger output always produces the same Gate decision

Who this is for

Governance runtime

executes automatically as the final step in the bounded-synthesis pipeline. No manual review, no override path.

The caller

receives a response that has passed the Gate — or an explicit state explaining why it was classified as partial, narrative, or blocked.