The field kit, published.

The grant itself. Names AI and machine-learning training — pre-train, fine-tune, validate, benchmark, evaluate, generative models included — and grants sublicensing through multiple tiers. Clause 6 states plainly that a model already trained cannot be untrained.

One page, one person, signed before recording starts. Written in plain language because a release nobody understood is not consent. Records the fee paid, and states the withdrawal right and its limit.

Read on camera at the head of every take, tying a face and an asset ID to a spoken agreement. It survives a lost or smudged signature, and it is the evidence that holds when paper does not.

Every column, its type, and what an empty cell means. The live CSV export builds its header from the same schema this page documents.

A real asset's annotation, with the failure events in it rather than described in the abstract.

The structure a records delivery is transformed into. A specification Texa authors, not a corpus Texa holds.

What was changed, dropped or de-identified on the way through, recorded per field.