Proof & Roadmap
What’s built, how we know it works, and what’s next
Every capability below was built against a written exit criterion and checked against known ground truth before it shipped.
How we validate
Test scenarios that know the right answer
Our test harness generates realistic traffic on real street networks, emits the same standard messages vehicles and signals broadcast, and records exactly what happened, so every result is compared with the truth.
- Scenarios that stress the systemRush hour, incidents and weather, signal outages, map revisions, delivery stops, late uploads and every kind of malformed message.
- Independent predictionsOur test harness re-implements the documented rules on its own and predicts every outcome, record by record, before the platform answers.
- Automatic comparisonScripts compare the platform’s API answers with the truth after every run, and any regression shows up immediately.
- 249 automated backend testsUnit, contract, integration and architecture tests run on every change, including tenant-isolation and privacy checks.
Results
Validation results by capability
Benchmarked against known ground truth in controlled test scenarios.
Connected-vehicle ingest
Messages from roadside units, normalized, pseudonymized and deduplicated.
| Check | Target | Result |
|---|---|---|
| Replaying the same data twice | 0 new rows | 0 new rows |
| Process killed mid-batch while draining 30,000 messages | Nothing lost or duplicated | Exactly 30,000 stored, all distinct |
| Sustained throughput | 5,000 msg/s | ≈19,500 msg/s on a laptop |
| Raw vehicle IDs anywhere in storage | None | None (every table scanned) |
Segment speeds & hard braking
15-minute speeds per road segment and direction; braking events detected from motion.
| Check | Target | Result |
|---|---|---|
| Observations matched to a road segment | — | 99.9% |
| Mean speed error per segment window | ≤ 1 m/s | 0.44 m/s (bias −0.02 m/s) |
| Derived hard-braking events | — | 1,019 of 1,019 found |
| Vehicle-reported hard-braking events | — | 591 of 591 found |
| Cells with fewer than 5 vehicles | Never published | Stored, never published |
Intersection performance
Signal metrics from probe vehicles and signal broadcasts, at four signalized intersections.
| Check | Target | Result |
|---|---|---|
| Stop-line passages detected | — | 97.0% within 3 s |
| % arrivals on green | ±5 pp | 4.85 pp mean difference |
| Stop rate | ±5 pp | 2.61 pp |
| Mean approach delay | ≤ 3 s error | 0.71 s |
| Red-light-running candidates | ≥ 80% recall & precision | 83% / 83% |
| Signal-data outage | Lag, never loss | Late data completes the same cells |
Identified telemetry & consent
Phones and telematics units uploading under a consent ledger.
| Check | Target | Result |
|---|---|---|
| Per-record outcomes vs. independent prediction | — | 56,936 outcomes, 0 differences |
| Consent revoked mid-run (16 people) | Ingest stops | Stopped at the next upload |
| Deletion requests | — | 32 of 32 completed, including compressed data |
| Cross-organization reads | Impossible | Refused both ways |
Trips, journeys & mobility
Trips detected from positions; endpoints protected for people without trip-history consent.
| Check | Target | Result |
|---|---|---|
| Trips reconstructed | — | 250 of 250 (start/end within 2 s) |
| Protected positions within 200 m of a home or work endpoint | None | None of 5,875 checked |
| Origin–destination and daily usage | — | Equal to ground truth |
Driving scores
Scores from events per 100 km, night driving and speeding, with weights set by each organization.
| Check | Target | Result |
|---|---|---|
| Factor rates for 241 people and vehicles | — | Match ground truth |
| Share of time speeding (fleet-wide) | — | 0.1512 vs. 0.1513 true |
| Scores vs. the published formula | — | Within 2 points |
Built & validated
Ten capabilities, each with a passed exit criterion
- 1Connected-vehicle ingestBasic Safety Messages from roadside units, decoded by USDOT’s open-source pipeline, unit-normalized, pseudonymized and deduplicated.
- 2Ground-truth test harnessRealistic traffic producing standard V2X messages with known answers, so every metric is checked, not assumed.
- 3Road analyticsMap matching, 15-minute segment speeds, derived hard braking and k-anonymity publication rules.
- 4Customer APIEntitlement-filtered GeoJSON and CSV with strict tenant isolation and an OpenAPI contract.
- 5Accounts & enterprise identityOrganizations, roles, API clients, enterprise SSO, optional MFA policy and an append-only audit trail.
- 6IntersectionsSignal phase and map ingest, versioned intersection geometry, signal performance metrics and broadcast health.
- 7Identified telemetryPhones and telematics units upload through an API or with device keys, under a consent ledger with export and deletion.
- 8Trips & journeysTrip detection, journeys, origin–destination flows and daily usage, with endpoint protection for people.
- 9Driving scores & retentionScores with weights each organization sets, and retention each organization chooses within bounds.
- 10Driver onboardingDrivers join with an invitation code or QR in one tap, and can stop sharing from the app.
Next
From validated platform to live deployments
What we’re building next. Customer needs shape the order.
New regions and partners
Connecting more roadside-unit networks, insurers and fleets.
Partner adapters
Consented automaker data (ISO 20078), telematics-platform integrations and bulk ingest.
Security hardening
Managed key storage for pseudonym and VIN keys, an independent penetration test and a privacy legal review.
Road weather
Wiper, temperature and traction signals from vehicles turned into road-condition products.
Known limits
What we tell every partner up front
Credible numbers come with their caveats. These are the ones that matter most today.
- V2X is a thin sample today
- Few production vehicles broadcast V2X yet, concentrated around instrumented corridors. V2X products are labeled as probe samples, never as traffic volume.
- Sparse cells stay private
- At quiet intersections a 15-minute window may hold few vehicles. Cells under five vehicles are suppressed rather than shown with misleading precision.
- Per-point map matching
- Positions are matched one at a time. Sequence-aware matching is planned for places where roads cross at shallow angles.
- Missing speed limits
- Where OpenStreetMap has no posted limit, speeding uses a 25 mph default until a better source is added.
Put Traffica to work
Get the platform, direct input into what we build next, and a team that answers with data.