DriveStable Fleet · Engineering evidence
Inside our 25,000-organisation synthetic planning test.
When a Driver calls in sick, a vehicle goes off road or urgent work arrives, the plan needs to change quickly without creating another problem somewhere else.
Synthetic engineering test · No live customer fleets were used
At a glance
One hour of sustained modelled demand.
The isolated model represented 25,000 fictional fleet organisations becoming ready for replanning every minute.
Built for the moment the plan changes
We deliberately made the day harder.
The simulation introduced Driver sickness, vehicles going off road, delayed jobs, changing time windows, urgent work and temporary processing failures. Every fifteenth wave doubled the planning load.
All injected failures recovered on retry. The 500-worker tier was the first configuration to keep fleet-wide p95 processing below ten seconds across all 60 waves, including every doubled-load stress wave.
Validity came before speed. If work could not be covered safely within the configured limits, the planner had to surface it—not hide it inside an impossible schedule.
Modelled operational benefit
What could this mean for a working fleet?
In the ordinary-pressure model, measured across 12 synthetic 30-vehicle fleet-weeks, DriveStable Fleet's recommendations were estimated to:
These are modelled estimates—not promises. A supervised fleet pilot is required to replace the assumptions with observed handling times, accepted plans and operational outcomes.
A calmer way to respond
See the problem. Understand the response.
DriveStable Fleet is being built to help operators see what requires attention, understand the available response and approve a feasible change without rebuilding the working day by hand.
The aim is fewer manual checks, faster responses and a clear explanation of what changed, what remains uncovered and why.
Full methodology and resultsView the technical evidence
Test conditions
- 25,000 synthetic organisations every 60 seconds.
- 60 consecutive waves representing one hour.
- Two repetitions across four capacity tiers, producing 12 million modelled planning operations.
- 1.775 million vehicles, 2.081 million Drivers and 23.285 million scheduled jobs represented in each wave.
- Every fifteenth wave doubled planner service time.
- Temporary retries injected at 0.10% with a five-second delay.
- External routing and network access remained disabled.
Planning results
Every tier cleared its work before the next minute and every temporary failure recovered on retry.
| Capacity tier | Result across 60 waves | Worst-wave p95 |
|---|---|---|
| 250 workers | Cleared every wave; 56 of 60 met the target | 12.34 seconds |
| 500 workers | Passed all 60 waves | 6.07 seconds |
| 750 workers | Passed all 60 waves | 3.98 seconds |
| 1,000 workers | Passed all 60 waves | 2.97 seconds |
The smallest tier handled the sustained volume, but four doubled-load waves exceeded the strict ten-second target. It remains an investigation configuration rather than the recommended performance baseline.
Database boundary
A separate PostgreSQL 17 test created 25,000 isolated organisation records and used 32 concurrent claim lanes. It leased all work items in 5.84 seconds, with zero duplicate leases, zero lost items and every event attempted exactly once.
Full modelled benefit range
Each row covers 12 synthetic 30-vehicle fleet-weeks.
| Pressure | Extra jobs covered | Service hours protected | Estimated admin hours saved |
|---|---|---|---|
| Ordinary | 267 | 912.32 hours | 61.50–85.30 hours |
| Disrupted | 628 | 2,113.08 hours | 155.67–227.07 hours |
| Severe | 954 | 3,222.15 hours | 261.00–403.80 hours |
The estimates compare automated recommendations with leaving the original plan frozen after disruption. “Service hours protected” means scheduled operating time recovered through feasible reassignment—not shorter mechanical repair time.
What comes next
The next gate is a protected hosted-environment soak across the complete queue, database connection and persistence path. A supervised pilot will then measure real dispatcher handling time, recommendation speed, edits, overrides, uncovered work and service hours protected.
Help us prove it in the real world
Bring your fleet into the next test.
If you operate a working fleet and want to help validate the next stage of DriveStable Fleet, we would like to hear from you.
Register your interestEvidence note: This page reports isolated synthetic engineering tests completed on 19 October 2026. The results are not a guarantee of deployed production capacity, legal compliance or financial savings. No HERE calls were made. Hosted staging and a supervised real-fleet pilot remain required before production-scale or commercial benefit claims are made.