Fleet Console
Oversight of an AV test fleet, made visible before it fails.
- My part
- Proposed the concept and did its UX research, design and prototyping
- When
- Self-initiated at Woven by Toyota; an NDA re-creation
- Status
- A concept. Nothing shipped, no outcome metrics, outcome internal
- Built on
- Plumb, an original design system, and a three-surface prototype: desktop, iPhone and in-vehicle
Interactive prototypeOpen the interactive prototype (opens in a new tab)
The problem

Test fleets grow faster than their oversight
Leads can't see oversight debt, and escalation hangs on one phone call. Public reporting on AV test programs describes this; I used it as design pressure, not evidence.
Two users
One drives. One is the safety net.
Operator, in the car
Flags an issue in a glance, then triages it later with data.
Lead, remote
The human safety redundancy: follows up incidents, audits footage, answers emergency calls.
The insight
Organize around “what needs me now”, not the data
Navigation is verbs and queues; objects sit one hop below. I rejected object-first and time-first.
Decisions
Six calls that shaped it
Workflow-first IA
Chose verbs and queues over object lists, because the hardest jobs are “act now” or “see now”.
Now owns debt
Chose a plain sum with its arithmetic shown over a composite score, because leads must reproduce it by counting.
Escalation rota
Chose a rota that re-routes on timeout over a fixed on-call field, because that field is a single point of failure.
Protocol-gated resolve
Chose five unskippable steps over “not applicable” toggles, because those hand judgment back to a tired lead.
No green
Chose shape plus label over color-only dots, because hue never carries meaning alone.
Preview width
Chose dropping the Vehicle column over horizontal scroll, because Status, Owner and Age are the triage signals.
Three surfaces
Each surface gets the job it's best at



The system

Plumb: color is meaning, never the only signal
There is no green: severity uses shape plus a label. Two density modes, light and dark.
Proof and limits

What's real, what's assumed
- The cross-surface hand-off check passes.
- The 2-second glance floor is measured, not eyeballed.
- Placeholders: Escalation rota timeout (5 minutes per link) and protocol wording.
- No outcome metrics exist; the outcome is internal and I can't share it.
Next: run first-click, timed-triage and debt-trust tests with real safety owners. Try the live prototype (opens in new tab).