Tracker comparisons
Railwatch vs. Amtrak’s train tracker
You’re meeting someone at New York Penn Station. Their train is late. Before you leave, you want to know whether it’s moving, how the delay has changed, and when it should reach New York.
That’s the sort of check Railwatch is built around. Its train dashboard brings the next stop, delay, journey progress, and map together. Amtrak’s own tools cover much of the same ground, with the advantage of keeping status alongside tickets, trip changes, and official service information.
Our recommendation: keep Amtrak for managing the trip and following operational instructions. Open Railwatch when you want a dedicated screen for following the train.
Start with the station if you’re picking someone up
Open New York Penn Station’s Railwatch board and look under arrivals. Find the train number, check the date and direction, then open its train page. Long-distance services can have more than one departure still on the railroad, so the run matters as much as the number.
On the train page, check the arrival estimate for New York, the latest reported location, and the age of that report. Then look at the delay trend. A delay estimate that has held at 20 minutes across recent reports presents a different picture from one that keeps growing.
The chart follows recent delay readings for the currently tracked stop, so check your New York arrival separately. A growing delay is a reason to check again before leaving. Allow time for parking and reaching your meeting point instead of aiming for the precise minute on the screen.
Amtrak’s map also shows station arrivals and departures. Railwatch’s advantage here is the linked workflow: station board, individual journey, and the details behind its current status.
What Amtrak’s tracker already does well
Amtrak’s Track Your Train map is a good starting point when your question is geographic. Search for a train, station, or route and explore the service on the map. Amtrak’s map guide describes train speed, travel history, upcoming stops, and station arrival and departure information.
Keep it close when the situation changes. A marker and an ETA don’t explain every cancellation, boarding change, or issue with a reservation. Amtrak’s service information and station announcements are the sources to follow for those decisions.
The website map and the Amtrak app are also different products. The map’s help recommends a desktop or laptop for the best experience. The mobile app includes status in active trips and a station-to-station tracker, alongside tickets and trip management. If you’re already using the app for your ticket, start there.
Where Railwatch’s train page earns its place
For repeated checks, the Railwatch train view puts the journey first. The next stop, reported delay, speed, and signal age sit with the train’s progress. Scroll down for the stop timeline, map, and recent delay readings.
You can stay with one page as the trip unfolds. Early in the journey, the map helps you place the train. Near your station, the upcoming stop and estimated arrival become the details you care about. The timeline connects them to the rest of the run.
Railwatch also supports dated official schedules and attaches live status when it matches that run. If a stop has a scheduled time without a matching live report, read it as a schedule. An empty live field doesn’t establish that the train is on time.
That layout is a good fit for a quick phone check. If you prefer to explore the whole network on a larger screen, Amtrak’s map remains worth keeping open.
Why you may see two arrival estimates
Amtrak already publishes estimates for upcoming stops. Railwatch can put a separate forecast beside the Amtrak estimate, using current delay, live movement, and patterns from earlier trips.
We checked how those forecasts compared. Replaying 11,450 accepted archived forecast/outcome pairs from October 11–15 across 737 train runs produced 6.16 minutes of mean absolute delay error for Railwatch, versus 8.30 minutes for Amtrak’s contemporaneous estimates. That is 25.8% lower mean absolute error on this replay cohort. In plain terms, it measures the average size of the miss. The methodology and tradeoffs are below.
The Railwatch forecast is experimental and limited to supported stops scheduled within the next two hours. It isn’t available for every train or stop. Read it alongside Amtrak’s estimate and the latest status of your train.
If the estimates disagree, check the stop and the timestamps, then leave some room in your timing. Either estimate can change when something happens further along the line. For boarding, follow Amtrak’s instructions and recommended arrival time rather than planning around the latest predicted minute.
When you want the wider picture
A station pickup starts with the station board. Watching several trains on the same service is easier from a route page, where Railwatch combines the schedule, matching live information, and a fleet map.
Station pages also show summaries of recent recorded delays. They help you see what has been happening around the station, but don’t predict the outcome of your particular journey. Check the sample size and update time before drawing conclusions.
Both Railwatch and Amtrak offer maps and ways to explore trains, stations, and routes. Choose your starting point around the question: one train, one place, or the whole service.
Check the timestamp before trusting the number
When two pages disagree, first confirm they show the same run and stop. Then compare timestamps. A page can refresh while the upstream position remains old.
Railwatch shows signal age and data-quality notices beside the journey details. Read them, especially when a train appears stopped or an estimate seems surprising. Public train information can be delayed or incomplete.
Railwatch is independent of Amtrak. It can help you follow the trip, but it cannot change a ticket, hold a connection, or confirm that you have time to board. Keep Amtrak for those decisions and bookmark your train for the next status check.
How we checked forecast error
We replayed recorded live inputs through the unchanged alpha-2026-10-10 model, frozen on October 11 at 01:07 UTC. Observations ran from October 11 at 01:10 UTC to October 15 at 08:30 UTC. The first and last dates are partial. All 737 evaluated runs were separate from the model’s 6,595 fit and calibration runs.
The archive samples one early observation per run, stop, and forecast horizon. It filters invalid or stale inputs, requires a recorded actual outcome, and holds forecast observations back by 48 hours before evaluation. The holdback applies to observation time, not necessarily to the time the train arrived. We retained the archived features unchanged and compared both forecasts against the same actual outcome.
Of 11,467 valid, mature, previously unseen archived pairs, 11,450 were accepted by the replay. Fourteen were unsupported and three produced past estimates. This is coverage of that filtered archive, not of all trains or website requests.
- Mean absolute error: Railwatch 6.16 minutes; Amtrak 8.30 minutes
- Median absolute error: Railwatch 3 minutes; Amtrak 4 minutes
- 90th-percentile absolute error: Railwatch 13 minutes; Amtrak 18 minutes
Average error improved on all five observation dates, all three forecast horizons up to two hours, and all 22 routes represented in this replay. One small route/horizon group was slightly worse: Capitol Corridor, 0–30 minutes, with 64 pairs.
There was a tradeoff for delays of at least 30 minutes. Railwatch detected 69.6% of them, compared with Amtrak’s 53.0%. It also raised more false alarms: 1.40% of observations whose actual delay was under 30 minutes, versus 0.33% for Amtrak. These rates describe observations, not individual trips.
These results come from replaying recorded inputs, not from logs of forecasts actually shown on the website. Request-time freshness, status, and display mapping were not available in the archive. The short window cannot establish future or network-wide performance. We checked artifact integrity and the portable model against 2,395 stored reference fixtures; we did not retrain the model for this comparison.