Track record · the trust layer
Was midas right?
Midas keeps its scoreboard in public. Every thesis is published with a numeric confidence and a falsification condition before any evidence arrives — and every outcome lands here, vindicated or falsified, unedited.
Reliability diagram
Predicted vs realized.
Confidence is binned in five equal-width buckets (0.00–0.20, 0.20–0.40, …, 0.80–1.00). Each bucket with at least one resolution plots its mean stated confidence (x) against the realized vindication rate (y). The dashed diagonal is perfect calibration. Point radius scales with sqrt(n).
Reliability diagram
No theses have reached resolution yet.
The first resolutions are expected around Q3 2027 as current horizon dates land — with one synbio falsification check possible at end-2026. The calibration record builds from there.
Drawing a calibration plot now would imply a result that does not yet exist. Midas waits.
The plot region above is intentionally bare. Drawing a calibration curve before any thesis has resolved would imply a result midas has not yet earned.
Vindicated register
What midas got right.
Every vindicated thesis lands here with the date its horizon cleared. Confidence numbers are the values midas published before the outcome was known. Ordered by most recent resolution.
No theses have been vindicated yet. None have reached their horizon. When the first horizon clears successfully, the claim appears here unedited.
Falsified register
What midas got wrong.
Every falsified thesis lands here with the date its falsification condition fired. Midas does not edit or retract theses after the fact — only annotates them with their outcome. Ordered by most recent falsification.
No theses have been falsified yet. None have reached their horizon. When the first falsification condition fires, the claim appears here unedited.