Admissibility-Gated Output
Generate Evaluate Release / Withhold
The model produces an answer. The runtime decides whether to release it.
The language model never sees the verdict. It does not retry. It does not optimise against the gate.
The caller sees the verdict and decides whether the candidate output is released to the user.
User LLM produces prose + structured claim FCAL evaluates the claim, returns terminal verdict Caller SURVIVE releases prose · otherwise withholds it
Try a scenario
Or ask a quick check
FCAL does not steer the model. FCAL does not retry the model. FCAL returns a terminal admissibility verdict; the caller decides whether output is released.  What this proves
Enter · Shift+Enter for new line Turns: 0 / 12  ·