Tollscopic binds evidence to the vehicle path before it creates the transaction.
Most tolling systems still match device events using time windows. That works until a real road gets messy. Tollscopic uses the physical trajectory as the organizing object: track the vehicle first, then ask which evidence belongs to it.
Timing windows fail exactly when transactions matter most.
Roads do not produce clean event sequences. The hard problem is not collecting signals — it is deciding which plate read, tag read, axle event, and classification output belong to the same vehicle.
The plate camera sees the wrong vehicle in the window where it expected to see the right one.
Tag and plate reads arrive with overlapping timestamps. Timing windows do not know which is which.
A vehicle drifts across lane markings. Lane-keyed correlation cannot decide where the evidence belongs.
A plate read fails. A tag is absent. Timing windows still try to assemble a transaction from whatever they have.
The vehicle path is the correlation key.
Tollscopic tracks each vehicle through the toll zone as a physical object. Plate, tag, axle, LiDAR, and video evidence then attach to the vehicle that was actually present at the relevant position and time.
Classification is a fused decision with evidence behind it.
Tollscopic combines available signals — axle evidence, vehicle dimensions, visual classification cues, LiDAR measurements where configured, and trajectory context — and outputs class, confidence, and provenance. Never a black-box label.
{
"txn_id": "tx_2026_05_20_a8c2f",
"class": "TRUCK_3AX",
"confidence": 0.92,
"provenance": {
"axle": "vision · 0.95",
"dimension": "lidar · 0.88",
"visual": "vlm · 0.91",
"trajectory":"edge · 1.00"
},
"fallback": null,
"model_ver": "avdc-2026.04-a"
} Five stages. The transaction is a derived record at the end of the chain.
The original evidence remains available for audit. The transaction itself carries the references back to it.
Each device produces timestamped evidence with confidence and provenance.
Evidence attaches to the vehicle path that produced it.
Classification is decided from the fused evidence on the trajectory.
Transaction record carries class, identifiers, evidence refs, and confidence.
Adapter posts the transaction to the back office. State is tracked.
The back office stays where it is. The roadside logic does not depend on it.
Transactions publish through a stable adapter. The operator's BOS / OBO remains the source of truth for billing. Tollscopic can change roadside logic, models, and devices without forcing the back office to change.
HTTPS · HMAC-signed · idempotent. The default integration for modern BOS.
Kafka topic. At-least-once delivery, dedupe on transaction id. Operators who already run an event-driven back office.
For operators with non-modern interfaces. The roadside logic stays the same; the adapter normalizes the output to whatever the back office expects.
See it on a real toll zone.
Bring us the zone, the devices, and the back office. We will show how trajectory binding handles the cases your current system treats as exceptions.