PROCEED — release permitted.
NEEDS_REVIEW — do not auto-release; route to an operator.
SILENCE — release denied, execution withheld, audit preserved.
SemeAI Gate · release authority
Generation creates a candidate.
Generation creates a candidate.
The Gate decides release.
SemeAI Gate is a post-generation release-control layer. It evaluates a candidate against policy, visible evidence, and current context before the host product releases, reviews, or withholds it.
The release-control boundary
01 · InputCandidate
The model output enters as a proposal, not an authorized answer.
02 · ContextPolicy + evidence
Rules, business data, source support, risk, and context are evaluated.
03 · AuthoritySemeAI Gate
The Gate—not generation or metadata—makes the release decision.
04 · OutcomeSHOW / REVIEW / BLOCK
The host receives a restrained business-facing action.
05 · TraceReceipt
The decision and its audit evidence remain available for inspection and replay.
One decision, two semantic layers
SHOW maps to PROCEED.
REVIEW maps to NEEDS_REVIEW.
BLOCK maps to SILENCE.
SILENCE is controlled non-release, not deletion. The candidate, decision, and audit trail remain distinct. Payment, metadata, and generation are not Gate authority.
What the Gate is—and is not
Post-generation control
A release checkpoint used after a model creates a candidate.
Not a model or chatbot
It does not replace the foundation model, generation runtime, or operator.
Not universal certification
It is not an alignment replacement, universal safety proof, correctness guarantee, or production certification.
See the principle operating on bounded public evidence.
The Repository Evidence Benchmark has its own display-only Presentation Gate. It demonstrates inspectable decision flow; it is not SaC/PoR runtime release authority.