The Tabularium

Criteria are fixed at publication. Every Watch Ahead call states its expectation, its window, and the counter-signal that would overturn it, in the edition where it appears. Those words are the resolution criteria; they are never revised.

The counter-signal is the trigger. If it fires within the window, the call is a Miss. If the expectation holds, Hit. Where the evidence cannot fairly decide, the verdict is Non liquet — the finding Roman jurors returned when a case could not be resolved — and it is excluded from the record.

Scoring. A hit counts one, a miss zero; non liquet is excluded from the denominator. No call is ever removed, renumbered, or regraded.

Confidence is logged with every call — in the brief's fixed confidence language (almost certain, very likely, likely, roughly even, unlikely…), set when the call is made. It is shown so the record can demonstrate skill, not a padded win rate: a wall of “almost certain” hits shows nothing, while called “roughly even” outcomes that land are where judgment shows. As calls resolve this is made explicit: the calibration table above reports, for each confidence word, how often those calls actually came true versus what the word claims; past roughly thirty resolved calls, a Brier score over the logged probabilities reduces it to a single number. Those readings feed back into how later calls are written.

Against the market. Where a call lines up with a liquid prediction market, its implied odds at the time of the call are shown beneath the confidence and tracked live (“at call → now”), and the record is scored against the market's too — a check on beating the crowd. It is a secondary lens: most calls have no aligned market, and one is never forced.