How we measure whether we're right
We'd rather show you our method than a flattering headline number. Here's how accuracy is defined, how we track it, and what we can — and can't yet — claim.
Calibration, not a single score
A prediction is a probability, so the honest test isn’t whether any one ticket confirmed — it’s whether the numbers hold up in aggregate. If we label a batch of tickets 70% likely to confirm and close to 70% of them do, the model is well calibrated. That’s the standard we hold ourselves to, and the one worth reporting.
The figures we’ll publish here
These are the numbers that belong on this page. We’re holding them back until enough verified outcomes are in to make them meaningful — an empty box is more honest than a premature statistic.
Verified outcomes
—
Tickets whose final status we've recorded after the chart was prepared.
Calibration
—
How closely stated probabilities match real confirmation rates.
History coverage
—
Share of predictions backed by real outcomes rather than an estimate.
A closed loop, from prediction to outcome
Every prediction is recorded
When we show a probability, we keep the inputs, the number and the confidence band alongside it.
We log what actually happened
After departure, once the reservation chart is final, the real outcome for each combination is collected and rolled up.
We compare predicted against actual
Grouped by probability band, we check whether tickets we called ~70% likely confirmed roughly 70% of the time. That's calibration.
The results feed back in
Those same verified outcomes replace earlier estimates in the engine — so predictions get sharper as real data accumulates.
How to gauge a single prediction
Until the aggregate figures are ready, every result already carries its own reliability signal: the confidence band. It tells you whether a number rests on a strong run of history or on a broader estimate, so you can weight it accordingly.
A High band is backed by a solid sample of exact matches; Estimated means there’s no history yet and the number is a model estimate. The methodology page explains what sits behind each one.
Transparency
See exactly how a number is reached
Accuracy starts with an honest method. The methodology page walks through the four steps and the confidence behind each.