01
The deployed system nobody names
Google decides which entities exist, through the Knowledge Graph; whose statements carry authority, through quality signals like EEAT; and, since AI Overviews, how the answer itself is framed. That last step matters: when the ranking layer writes the answer instead of listing candidates, the ranking function stops being a librarian and becomes an epistemology.
None of this is a criticism. It is an observation about scale: the most consequential machinery for deciding what gets treated as fact is deployed infrastructure, built by search engineers, running on trust gradients — and most discussion of AI knowledge acts as though it does not exist.
02
Trust is a gradient, not a gate
The formal picture I keep returning to: a belief about an entity carries belief mass, disbelief mass, and uncertainty mass that must sum to one. Independent confirmation moves mass from uncertainty toward belief; contradiction moves it toward disbelief. "Verified" is not a boolean anywhere that matters — it is a position on this gradient.
Wright's tier language makes the gradient legible: cryptographically anchored entities at the bedrock, platform-authenticated ones above, passively cross-referenced ones after that, and single-source fog at the top. The engineering question is never "is this true?" — it is "how much evidence, of what independence, from what sources, how recently?"
03
Why agents need it
An agent without belief grading treats a signed record and a stray forum post as the same kind of object. Its memory recalls both with equal confidence, its plans build on both with equal weight, and its errors compound quietly — which is the recall-debt failure and the unadjudicated-zero failure wearing a different coat.
The gradient is the adjudication layer for beliefs: it tells a system when it may commit, when it should hedge, and when a human should look before anything downstream inherits the claim.
04
Still open
The question I am working on is what a system should do at each position on the gradient — the policy layer above the arithmetic. That work is ongoing and belongs to a longer piece; this note records the position it starts from: veracity is not metadata sprinkled on a knowledge base. It is infrastructure, and the largest systems in the world already run on it.