Forecast experiment (V11)

Read this first. This is an experiment about the index, not advice, and until the registered criterion resolves it should be presumed to show that salience does not forecast. The monitor itself makes no forecasts anywhere; this page exists to test — in public, against a dumb baseline, under a registration signed before any question opened — whether press salience contains any predictive information about its own future spikes. Nothing here is a probability of conflict, risk, or harm; the outcome variable is a property of press coverage.

The design, frozen on signature (2026-08-06): every Monday, five mechanical questions — "will channel C record at least one spike day within 14 days?" — each committed before its window opens. Two arms answer: climatology (the trailing 104-week frequency of that exact event) and a salience-informed logistic model on two registered inputs whose coefficients were fit once on pre-registration history and frozen. Grading is mechanical from the committed series; voids count and publish. The fork criterion is registered: after 26+ resolved weeks per channel, the salience arm must beat climatology's Brier by ≥5% relatively at p < 0.05, or the negative result publishes prominently — there is no third outcome.

The question tape

Newest first. Probabilities are recorded at generation and never edited; outcomes are graded by the registered spike detector when each window closes.