Predictions stated before the substrate. Tested by the substrate.
External AI governance measurement operates in a pre-paradigmatic domain. There is no panel of human raters that could be the ground truth, no historical record of governance failures whose magnitude is independently quantified, no regulatory benchmark whose calibration could be borrowed. Criterion validity in the classical sense is structurally unavailable. The substrate validates instead by construct validity: it produces values that satisfy a network of theoretically derived expectations stated before the substrate is sealed. Architectural envelope predictions follow by construction from methodology choices and must hold. Pre-specified predictions are quantitative empirical claims registered before sealing and tested against the sealed snapshot. Post-validation observations surface patterns after sealing but cannot be confirmed or falsified because they were not pre-registered. For the v13.1.0 substrate, all architectural envelope predictions held and all pre-specified predictions confirmed.
The sealed v13.1.0-production-day-zero-full snapshot was preceded by eight pre-registered predictions across two classes. Architectural envelope predictions follow by construction from methodology choices: failure of an envelope prediction constitutes methodology failure. Pre-specified predictions are quantitative empirical claims registered before sealing: the methodology stakes its credibility on producing values within stated ranges. All eight predictions held against the sealed substrate. The v13.0.0 scanner that preceded this iteration produced a clean falsification on a different prediction class, which motivated the v13.1.0 corrective interventions documented in Stationary Sea Part 1, Section 5.4. A falsifying experiment produces more information than a confirming one.