Audit · DVAS · immutable evidence · replay

The transaction can be traced back to the evidence that produced it.

The transaction is a derived record. The evidence remains available: video, trajectory, plate images, axle and classification signals, LiDAR and RFID where present, timestamps, confidence, provenance, and model context.

01 · DVAS · one transaction, all evidence

One screen. Vehicle path, video, plate, class, confidence, transaction state.

A reviewer should see why a transaction exists without reconstructing it from separate logs. DVAS collapses investigation time and makes hard cases explainable.

DVAS · tx_2026_05_20_a8c2f · RESOLVED 14:33:05 EDT · 2026-05-20
SYNCHRONIZED VIEW · sidefire + plate + class
VEH · 0.92
cam_l2 · sidefire
traj_8f2e
plate · anpr_l2
BFK-2417
conf · 0.94
class · avdc
2-AXLE
conf · 0.92 · agreement: axle+vis+lidar
TRANSACTION RECORD
txn_id tx_2026_05_20_a8c2f
traj_id traj_8f2e
plate BFK-2417 · 0.94
tag 0x7F2A91C3 · matched
class 2-AXLE · 0.92
lane L2
time 14:33:05.142 EDT
model_ver avdc-2026.04-a
evidence_refs 5 items · stored
BOS · ACK
delivered · 14:33:06
REPLAYABLE
full evidence preserved
Schematic. Synchronized DVAS review surface for a single transaction. Vehicle path, plate, classification, model version, evidence references, and back-office acknowledgement, all visible without leaving the surface.
02 · Immutable evidence store

The evidence is the ground truth. The transaction is a view over it.

Legacy systems often treat raw events as intermediate inputs. Once a transaction is formed, the durable record may be the transaction and a few attached images. Tollscopic preserves the evidence beyond the transaction so the transaction stays explainable.

01
What gets stored

Video segments, trajectory records, per-device events, classification outputs, confidence values, model + scorer + configuration + schema versions, decision provenance, and back-office state.

02
How long

Retention is policy-driven by the operator. The system supports both short retention for cost-sensitive operators and long retention for audit-heavy contexts.

03
Why immutable

An evidence store that can be rewritten is not an evidence store. Audit, reconciliation, and replay all depend on the records being durable.

03 · Replay and model versioning

Improvements can be tested against historical evidence.

When a model, threshold, or configuration changes, historical evidence can be reprocessed to understand how the change would have affected real cases. Regression testing, auditability, and safer model evolution all live here.

01
Pick historical window

A day, a week, a set of disputed transactions, or a corpus of edge cases.

02
Apply candidate version

New model, new scorer, new threshold, new configuration. Pinned by version.

03
Compare and decide

Differences are surfaced as a diff. Reviewer decides whether to promote.

04 · Reconciliation

What the road saw. What was generated. What was acknowledged.

One evidence chain across the roadside, the transaction engine, the back-office adapter, and the back-office response. The system treats these as states of one record, not as separate logs to compare.

A · STATE
Observed

What the roadside captured: trajectories, plate reads, tag events, classification evidence.

B · STATE
Derived

What the transaction engine produced: a transaction record with class, confidence, and refs.

C · STATE
Delivered

What was published to the back-office: payload, timestamp, idempotency key, retries.

D · STATE
Acknowledged

What the back-office accepted, rejected, or routed to its own exception queue.

Disputes are not solved by faster billing. They are solved by better evidence.

The audit story is what a serious tolling operator measures vendors on. We are happy to walk through any disputed case end-to-end.