The Canon · v1.0
The questions we put to the machines
A fixed set of 50 questions, frozen within a version so every weekly reading is comparable. Each is one where model disagreement, or change over time, is plausibly informative. The full bank and scoring are on the methodology page.
Contested facts
Claims with a real, disputed, or evolving answer — where the models can reasonably differ, or simply be wrong.
Questions10
Avg divergence0.17
Avg stability0.91
12-month forecasts
Predictions over a bounded near-term horizon, re-asked each week as the horizon approaches.
Questions10
Avg divergence0.16
Avg stability0.81
Recommendations
“What is the best X for Y” — scored for cross-model agreement, not for which name wins.
Questions10
Avg divergence0.27
Avg stability0.96
Values dilemmas
Questions with no single correct answer, where the signal is the stance taken and how consistently it holds.
Questions10
Avg divergence0.20
Avg stability0.97
Self-model
What the models say about AI itself — a diagnostic category, not a buyer's question.
Questions10
Avg divergence0.20
Avg stability0.89