Pre-Registered Predictions

Before each gate, the program files the numbers it expects to measure, so demonstrations are judged against commitments rather than hindsight.

Predict, then measure

Pre-registration means writing down the predicted result of a test, with its uncertainty, before the test is run. The breeder program applies this discipline to every physics gate. The value is simple: a prediction filed in advance cannot be adjusted after the fact to match whatever the machine did.

What gets registered

• G1: energy confinement time and the resulting Q_sci at design current and field
• G2: achieved peak field, conductor temperature margin, and quench behavior
• G3: neutron yield, tritium breeding ratio, and recovered tritium mass

Why it matters for credibility

Fusion has a long history of results explained after the fact. Reporting measured-versus-predicted, with both numbers visible, lets outside reviewers see whether the design models actually predicted the machine, or only rationalized it. It also makes a miss informative: a confinement time below prediction tells the program exactly where the model was wrong.

Handling misses honestly

A pre-registered prediction that the machine fails to meet is not hidden. It is reported, and it feeds directly into the NOAK design revision. The program treats a documented miss on FOAK as the intended function of a first-of-a-kind unit: to find where reality departs from simulation while the stakes are contained.

All current numbers, Q_sci 3.076, 16.84 T, TBR across 1.1/1.5/1.8, are predictions in this sense. They are simulation-based design values, registered as the targets the built machine will be measured against.