Every public prediction Magna Bureau issues — dated, probabilized, and scored against reality. Open by default, audited quarterly, never edited after issue. A forecaster that hides its misses is selling something else.
| Forecast | Probability | Resolves | Judged by | Status |
|---|---|---|---|---|
The US Federal Reserve will leave the federal funds rate unchanged at its October 2026 FOMC meeting. | 80% | 2026-11-04 | Federal Reserve FOMC meeting statement | Open |
OPEC+ will hold November 2026 production quotas unchanged at its October ministerial meeting. | 75% | 2026-11-04 | OPEC official press release | Open |
Brent crude front-month futures will settle above $75 a barrel on 2026-11-05. | 60% | 2026-11-04 | ICE daily settlement data | Open |
US CPI for October 2026, released in November, will print below 3.0% year-on-year. | 65% | 2027-01-03 | US Bureau of Labor Statistics CPI release | Open |
Nvidia's quarterly earnings reported in November 2026 will show year-on-year growth in data-center revenue. | 85% | 2027-01-03 | Nvidia quarterly earnings release | Open |
Brazil's October 2026 presidential election will require a second-round runoff. | 70% | 2026-11-04 | Tribunal Superior Eleitoral official results | Open |
Each morning edition carries two or three forecasts drawn from the day's reporting — falsifiable statements with an explicit probability, a resolution date (30, 90 or 365 days), and a named resolving source. The probabilities come from the Bureau's internal forecast desk; the prose may explain a number, it never manufactures one. No forecast is ever published at 0% or 100% — certainty is not a forecast.
An issued forecast is never edited. If the desk's view moves, the move is logged as a new event with its reason; if a statement needs correcting, the correction is appended and the original stands. Every event is hash-chained to the one before it, so any silent alteration — a nudged probability, a backdated resolution — breaks the chain and the quarterly audit refuses to score.
Each quarter the ledger is scored three ways: hit rate (share of resolved forecasts called correctly), Brier score (the mean squared error of the probabilities — 0 is perfect, 0.25 is a coin-flip), and calibration (whether events called at 70% happen about 70% of the time). Misses stay on the page, in full, forever. That is the whole point.
