A ghost bus is a scheduled trip that never ran — the bus never departed, or left the garage but never appeared in the live feed for its assigned trip. A vanished prediction is a trip that started but whose arrival countdown disappeared mid-route, before the final stop, with no completion recorded.

The ghost rate is ghosts ÷ scheduled trips observed during the polling window. Days with partial collection (outages, the early dataset) show lower raw counts but a roughly comparable rate; the rate chart is the more reliable signal across time.

Each time a stop prediction disappears from the live feed, the tracker compares the bus's scheduled departure time for that stop to the moment the prediction died. Early means the bus reached the stop more than 1 minute ahead of schedule; late means more than 5 minutes behind; everything in between is on time — matching the FTA standard PRT reports against publicly.

The Timing stops only toggle filters to GTFS timepoint=1 stops — specific stops where PRT makes a firm schedule commitment and drivers are instructed to hold. The majority of stops carry interpolated times: estimates computed by dividing the gap between two real timing points proportionally across the stops between them. Timing-stop results are a stricter, more meaningful signal of whether PRT is actually keeping its schedule commitments.

When a prediction disappears, the tracker classifies the reason by the vehicle's reported state at the time:

Unexplained drop — the prediction simply stopped updating with no clear cause. The vehicle may have gone off-route, lost GPS, or the feed glitched.
Stopped mid-route — the vehicle switched to an out-of-service state before finishing the trip; likely a breakdown, pull-out, or driver change that wasn't recovered.
End of run — the vehicle reached a layover or terminal and the trip ended earlier than scheduled, or the prediction ran past the actual last stop.
Never left garage — the vehicle never appeared in the live GPS feed for this trip; the scheduled departure came and went with no vehicle observed. Similar to a ghost, but a prediction existed in the feed before disappearing.
Late join — the prediction appeared in the feed before the vehicle was tracked by GPS, so the trip largely ran; this is a tracking artifact and is hidden by default.

Vanish rate is the share of a vehicle's trips whose arrival countdown vanished mid-route — a proxy for that bus dropping out of service before finishing its run.

It is a trailing record: both the vanishes and the trip count are tallied per day, over only the days we still hold consistent data for. Older days fall off as their raw GPS rows are pruned — so a vehicle's history here reaches back a few weeks, not to the start of collection, and the window grows forward over time.

Each circle represents one stop. Color reflects the share of observed predictions that were late (>5 min behind schedule); size reflects sample volume. Early means the bus reached the stop more than 1 minute ahead of schedule; everything else is on time. Stops need ≥30 observations to appear.

The Timing stops only toggle shows only timepoint=1 stops from the GTFS schedule — the anchor points where PRT commits to a time and drivers are instructed to hold. Most stops have interpolated times (estimated from adjacent timing points) and are noisier signals; timing stops give a cleaner picture of real schedule adherence.

Two vehicles on the same route and direction are bunched when they are running within half a mile (2,640 ft) of each other along the route — measured by projecting each vehicle onto its route geometry, so it's distance in the line of service, not straight-line distance. The bunch rate is normalized: out of the snapshots where a route had ≥2 vehicles on the same pattern (a chance to bunch), it's the share where at least one pair was within that gap — so high-frequency trunk routes don't top the list on volume alone.

The hourly chart adds a second, complementary measure. Route-moments bunched is that same rate (a route counts if any pair on it is bunched). Buses bunched is vehicle-level: of all in-service buses, the share personally caught in a bunch — a lower, more literal "what fraction of buses are bunched right now."

A corridor is a stretch of street a bundle of routes share — Forbes/Fifth through Oakland, the East Busway, the Liberty Ave spine. Here two buses count as bunched when they're within half a mile along that shared trunk even if they're on different routes: the platoon a rider waiting at the stop actually experiences, where three buses arrive together and then nothing for twenty minutes. The cross-route bunch rate below is the percentage of exposed snapshots (moments the corridor had ≥2 buses on it) in which at least two buses of different routes were within half a mile — 0–100%, a signal the per-route view above structurally can't see.