UNI Gate Registry (rendered)

This is a rendered view of a ledger, a list added to and never edited, and it says so at the top. The ledger itself is an append-only file with one row per gate and a published schema. Nothing on this page may be edited directly. Changing a verdict means appending a new row and re-rendering, which means the history of a verdict is preserved rather than overwritten.

Two vocabularies are declared before any result appears. Verdicts are drawn from a fixed set of words and are never scored as a percentage. Evidence is also classified. It separates a result that was independently reproduced from one that was observed with an artifact, and from plain command output. It separates both from a security-relevant claim that is unproven, and from something that has not yet been established at all. That second vocabulary is the interesting one, because it lets a passing verdict still be marked as weakly evidenced.

A summary line gives the tally by name, and it is careful to say that the row count is larger because many rows are superseded revisions of the same gate. Stating both numbers, rather than one, is what keeps a backlog from being confused with the history of a backlog.

The rest is the table. Each row names a gate, the phase of work it belongs to, its verdict, its evidence class, and the file recording what was run, which carries the evidence. The gates span very different kinds of work: the mathematics and behaviour of the agents, the studio and broadcast surfaces, and network and naming changes. Then there is a long run of gates about honesty itself, whose names describe a display refusing to show a green it did not measure, or refusing to imply a claim no code enforces.

Several rows are pending, which the vocabulary defines as written down in advance but not yet run. Several are partial. One is a failure, and it is left in place rather than removed. That is the point of the page: the adverse rows are not filtered out of the view.