Reference
Citable reference pages on AI output verification: each frame defines one concept precisely — its structure, its invariants, and the tests that separate the real architecture from look-alikes.
Frame 01
Every discrete factual assertion checked against an approved source corpus before delivery — decomposition, per-claim attribution, and selective refusal, with five structural tests any vendor claim can be measured against.
Frame 02
In clinical, legal, or regulatory settings, unsupported answers must be structurally blocked, not made less likely — three architectural tiers, a binary pass/fail criterion, and five measurable thresholds for any system that claims to prevent hallucinations.