Cruzar Insights — Accuracy
Public accuracy receipts
Every 6-hour wait-time prediction Cruzar makes is logged here, then compared against the actual wait that arrives 6 hours later. We publish this because most prediction products don’t — and you should be able to see exactly how often we beat the alternative for the bridges you cross.
Live soak — predictions vs reality
v0.5.x · 6h horizon · per-bridgeLive capture started 5/8/2026, 4:15:56 AM CT. 0 predictions validated, 1000pending. Predictions write every 15 min and are compared to the actual wait observed 6 hours later via CBP's data feed.
| Bridge | n observed | Cruzar avg error | CBP-live avg error | Lift vs CBP-live |
|---|
| Brownsville Puente Nuevo535504 | 0 | — | — | — |
| Los Tomates535503 | 0 | — | — | — |
| Brownsville Veterans535502 | 0 | — | — | — |
| Brownsville Gateway (B&M)535501 | 0 | — | — | — |
| Roma231001 | 0 | — | — | — |
| Donna230902 | 0 | — | — | — |
| Progreso230901 | 0 | — | — | — |
| Rio Grande City230701 | 0 | — | — | — |
| Anzalduas230503 | 0 | — | — | — |
| Pharr-Reynosa230502 | 0 | — | — | — |
| Hidalgo230501 | 0 | — | — | — |
| Eagle Pass II (Camino Real)230302 | 0 | — | — | — |
| Eagle Pass I230301 | 0 | — | — | — |
| Laredo Colombia Solidarity230403 | 0 | — | — | — |
| Laredo II (Juarez-Lincoln)230402 | 0 | — | — | — |
| Laredo I (Gateway)230401 | 0 | — | — | — |
CBP-live = the wait CBP was reporting at the moment we made the prediction (i.e., what a dispatcher checking bwt.cbp.govwould have seen). That’s the apples-to-apples comparison: we predict 6 hours forward, CBP shows now, the actual wait arrives 6 hours later, and whoever is closer wins.
Rows with fewer than 30 observed predictions are shown faded — there’s not enough data yet to draw conclusions. Full confidence requires 14+ days of soak.
Backtest — held-out test split
v0.5.4 · 2026-04-29 trainingPre-launch validation against historical data. Trained on most of the available wait-time history, evaluated on a frozen test split the model never saw. This is NOT operational accuracy— it’s the bound we expect live performance to land near, given identical conditions. Live soak above is the actual proof.
| Bridge | RMSE | vs CBP climatology | vs persistence | n_test |
|---|
| Stanton DCL240204 | 0.4 min | — | +31.0% | 1561 |
| Laredo I (Gateway)230401 | 4.1 min | — | +21.8% | 1551 |
| Del Rio230201 | 25.1 min | -13.9% | +23.0% | 1110 |
| San Ysidro250401 | 39.3 min | — | +40.2% | 626 |
| Brownsville Gateway (B&M)535501 | 32.4 min | +15.0% | +29.1% |