Trenalia

Data validation

From 2026-11-04 to 2026-11-18 · computed 2026-11-18 04:30 UTC

How accurate Trenalia and its sources are, measured against what happened afterwards: every night, what the site showed is compared with the real arrival, measured by Trenalia from station boards and from the GPS position recorded throughout the journey. It is fully automatic: nothing is entered by hand.

11.1 minmean error of Trenalia's delay (official: 10.4 min)
38.1 %samples within ±3 min of the real arrival delay
297journeys with an observed arrival (36,201 samples)
66.8 %network alerts confirmed (4,356 checked)

1 · What happened at the end of the journey

Trenalia measures on its own when each train arrived: from the destination station's arrivals board (ADIF or RadarDeTrenes) when available, otherwise from the continuously recorded GPS position (entering within 700 m of the station, or an estimate from the last position and speed if it vanishes within 3 km). That arrival is compared here with Renfe's last published delay. Positive bias = Renfe reports more delay than observed.

SamplesMean difference (min)Agree within ±2 minRenfe bias (min)p90
All observed arrivals2975.153.9 %2.512.4
Destination arrivals board2315.549.4 %3.513
GPS: entering the station533.977.4 %-0.96.8
GPS: estimated when signal lost nearby132.438.5 %-2.44
Boards as the train passes each station26,1273.665.5 %1.48

2 · Delay shown vs. observed arrival

The toughest test: what the site showed at each moment is compared with the arrival measured by Trenalia (board or GPS), not with Renfe's figure. “Official” is Renfe's published delay; “Trenalia” the projected one. Gain = how much Trenalia reduces the official error.

SamplesOfficial error (min)Trenalia error (min)±3 min official±3 min TrenaliaBias (min)Gain
Total36,20110.411.144.9 %38.1 %-0.6-6.4 %
0–15 min before arrival3,6825.35.959.6 %54.8 %4.1-10.9 %
15–30 min before arrival3,9026.77.553.5 %47.6 %3.4-11.9 %
30–60 min before arrival7,2337.68.350.3 %44.1 %1.9-9.8 %
60–120 min before arrival10,56910.711.539.6 %33.3 %0-7.5 %
2–4 h before arrival10,81415.215.638.3 %29.5 %-5.8-3 %

By service

SamplesOfficial error (min)Trenalia error (min)±3 min official±3 min TrenaliaBias (min)Gain
Media Distancia and regional17,86614.414.839.3 %33.8 %-4.2-3 %
AVE8,1484.35.162.7 %52.9 %2.7-18.2 %
Alvia3,61513.113.731 %27.2 %-0.6-4.5 %
Avant2,8373.24.167.3 %55.6 %2.3-27.6 %
Euromed and Intercity2,3827.78.630.4 %26.7 %6.1-12 %
Avlo1,3538.110.327.3 %16.7 %10.3-27 %

By day

SamplesOfficial error (min)Trenalia error (min)±3 min official±3 min TrenaliaBias (min)Gain
2026-11-043,1907.27.754.5 %46.9 %-0.1-7.6 %
2026-11-051,3778.28.251.9 %45.9 %-1.4-0.1 %
2026-11-063,50588.546.1 %39.8 %2.5-6.1 %
2026-11-072,1569.59.536.3 %33.3 %-2.5-0.3 %
2026-11-083,8486.2745.5 %37.4 %2.7-13.4 %
2026-11-092,604910.541.3 %30.6 %4.6-16.4 %
2026-11-1047411.39.232.5 %33.3 %-5.1+18.3 %
2026-11-111,95913.415.241.8 %27 %-2.6-13.9 %
2026-11-122,59936.838.349.4 %42.4 %-23.4-4.1 %
2026-11-133,44698.847.4 %42.9 %-2.6+2.2 %
2026-11-143,0786.27.348.3 %40.5 %3-17.1 %
2026-11-151,6648.310.439.8 %31.4 %10-25.4 %
2026-11-163,9699.39.242.1 %35.1 %1.1+1.7 %
2026-11-172,3328.38.936.7 %37.3 %1.1-7.4 %

3 · Delay shown vs. Renfe's last figure

The same, taking as final the last delay Renfe published at the end of the journey (available for almost every journey, but not an independent measure). It shows whether Trenalia anticipates what Renfe will end up reporting.

SamplesOfficial error (min)Trenalia error (min)±3 min official±3 min TrenaliaBias (min)Gain
Total42,7099.410.144.8 %41.3 %-1.7-7.5 %
0–15 min before arrival4,2061.42.690.5 %79.3 %2.1-85.9 %
15–30 min before arrival4,4033.44.371.2 %59.9 %2.6-26.9 %
30–60 min before arrival8,3456.36.653.6 %50.1 %0.4-5 %
60–120 min before arrival12,38110.51134.7 %32.8 %-1.9-4.5 %
2–4 h before arrival13,37314.915.925.6 %25.7 %-5.5-6.3 %

By service

SamplesOfficial error (min)Trenalia error (min)±3 min official±3 min TrenaliaBias (min)Gain
Media Distancia and regional23,55512.213.339.5 %37.2 %-3.8-9.1 %
AVE8,4014.54.757.7 %53.9 %0.6-3.9 %
Alvia4,07610.39.935 %31.3 %-3.6+4.7 %
Avant2,8733.54.265.2 %57.1 %2.6-20 %
Euromed and Intercity2,4108.19.234.7 %31.1 %7.4-12.8 %
Avlo1,3944.74.761 %50.1 %0.1+0.5 %

By day

SamplesOfficial error (min)Trenalia error (min)±3 min official±3 min TrenaliaBias (min)Gain
2026-11-043,8226.96.654.7 %49.3 %-2.7+5.1 %
2026-11-052,1697.17.249.8 %44.9 %-1.2-2.2 %
2026-11-064,5058.38.642.6 %39.8 %-0.4-3.3 %
2026-11-072,7818.511.341.2 %35.9 %-0.3-33.7 %
2026-11-084,7797.67.442.6 %40.8 %-1.7+2.1 %
2026-11-092,9031011.135.3 %31.4 %3-11.2 %
2026-11-107627.314.648.6 %37.4 %8.4-99.9 %
2026-11-112,2437.2835.8 %36.2 %-0.1-11.8 %
2026-11-122,72033.534.154.7 %49.2 %-27.4-1.8 %
2026-11-133,6526.38.353 %44.1 %2-31.5 %
2026-11-143,7897.49.242.4 %38.7 %3.3-24.6 %
2026-11-151,9425.85.944.9 %42.4 %-0.1-1.1 %
2026-11-164,0129.38.144 %43.6 %-2.5+12.5 %
2026-11-172,6308.89.137.5 %40 %0.2-3.5 %

4 · Learned models

Trained every night on the history and evaluated on the last week, which is held out from training. They are only switched on if they clearly beat the baseline.

Delay correction (ETA)

rejected no mejora lo bastante a la referencia · trained 2026-11-18 04:30 UTC

Model mean error (min)2.348
Baseline mean error (min)2.272
Gain (fraction)-0.034
R²0.038
Training examples97,034
Validation examples52,966
Trees37
Validation from2026-11-12

False alarm filter for closing alerts

rejected AUC insuficiente · trained 2026-11-18 04:30 UTC

AUC0.732
Training examples3,260
Validation examples816
False alarm share0.918
Alerts that would be hidden (p ≥ 0.9)709
Precision when hiding0.989
Trees35

5 · Train closing alerts

There is no official data to compare them with, so we label what happened afterwards. If the gap kept closing, the alert was consistent with the data; if the train ahead was stopped at a station, it was probably a false alarm (normal operations).

Candidates evaluated2,378
Alerts shown2,044
Still closing 2 min later11 %
Likely false alarm (train ahead at a station)0.5 %

What happened afterwards

Automatic check 3 h after each alert shown: whether both trains finished their journey normally, whether Renfe or Adif published an incident in the area, and whether there were network alerts nearby. “No signs” is not official confirmation: official accident data is published by the CIAF.

No signs973
Incident in the area579
Mentioning collision or derailment0
Signs to review492
No data0
Reviewed by hand0

6 · Network alerts

A segment or stopped-train alert counts as confirmed if any train involved gained 3 min or more of delay 10 to 30 min later; a sudden delay jump, if it did not revert (it was not a data glitch).

AlertsCheckedConfirmed
Train stopped outside a station3,6993,33558.1 %
Sudden delay jump80678498.7 %
Segment with several trains losing time24023784.4 %
Total4,7454,35666.8 %

7 · Data quality

Days with trajectories15
Position samples per day102,671
Positions snapped to the track99.1 %
Estimated precision (median)189 m · p90 987 m

8 · Incident hypotheses

Trenalia opens a hypothesis when it sees, live, a stopped train or several trains in trouble on a section, and checks it by itself 3 h 30 min later against the actual delays. Every night it also re-simulates the previous day's situations with the engine to see whether it reproduces them. They are not Adif or Renfe notices.

Hypotheses checked live

SamplesConfirmedWith official noticePassenger votes
Line closure or interruption75577.7 %590
Stopped or broken-down train2,70278.2 %3390

Re-simulation of past situations

SamplesReproduced
Line closure or interruption4,7930 %
Stopped or broken-down train336.1 %
Slow running1693 %

Learned patterns: 1,713 · validated by the maintainer: 82 · proposed: 0 · retired: 317

What Trenalia tells travellers

Accuracy of the arrival times people see in “I’m on a train” and “Your train” (last 14 days): share of times the time shown was within ±5 min, by how long before arrival and the confidence given. Cases in brackets.

High confidenceMedium confidenceLow confidence
60 min before arrival50 % (8)—33 % (3)
30 min before arrival17 % (6)—100 % (1)
15 min before arrival17 % (6)—100 % (1)
5 min before arrival0 % (3)——
Arrived more than 15 min later than stated0 %—20 %

Only trains and stations are counted (all queries for the same train and station are merged), never people. The actual arrival is the last estimate in its final minutes or the moment the train was seen passing.

Free software (GNU AGPL v3)