Track record
Updated live. We publish accuracy only after at least 30 predictions are scored, and we say so while we are still collecting.
Every leave-by answer logs the security wait we assumed. Once that time passes we compare it with the median MEASURED reading (airport feeds, ±10 min) for the same airport and lane. Predictions with no measured reading are counted as unscorable, not hidden. Numbers appear only once enough predictions are scored.
11745 predictions logged · 10695 scored against a measured reading · 1028 unscorable (no measured reading at that time; most airports have no live feed).
Scored predictions
| Scope | Scored | Mean error | Within 5 min | Within 10 min |
|---|---|---|---|---|
| All | 10695 | 2.9 min | 85.4% | 94.1% |
| Live-based | 5648 | 2.9 min | 84.8% | 94.2% |
| Estimated baseline | 132 | 21.5 min | 1.5% | 4.5% |
| Reported by travelers/agents (0 reports) | 0 | collecting | — | — |
For AI agents
Every leave-by and wait answer from the API and MCP server includes a trust object (confidence, basis, data age, abstain, guidance) and, for leave-by, a prediction_id. The same data is available as JSON at /api/v1/accuracy and via the MCP tool track_record.