San José curb pilot exposes gaps between installed equipment and usable evidence

Engage the transportation director, curb operations staff, public works, IT, accessibility representatives and procurement lead. Ask which decisions need timely curb observations and which missed events could misdirect those decisions. Offer a bounded corridor validation and integration assessment before a broader equipment purchase. The value hypothesis is a more defensible operational dataset, not guaranteed revenue, reduced crashes or shorter journeys. Discover existing assets and their approval dependencies. A small locality should assess whether it can maintain and review the proposed sensor estate. The historical pilot illustrates questions to ask; it does not establish a current opportunity or justify extrapolating one observation session.

Start with a representative field protocol covering event duration, weather, lighting, user type and location. Compare independently annotated observations against timestamped platform outputs and report missed events separately from classification errors. Prerequisites include lawful sensor access, an agreed data model and operational APIs. Test latency, outages, duplicated records, stale curb rules and retrieval after a vendor change. Assess storage, bandwidth and retention across candidate edge, cloud and hybrid designs; the report is not a comparative hosting benchmark. A proof of value should demonstrate the required end-to-end fidelity before any agent is allowed to change allocations or initiate enforcement.

Assign the curb service manager accountability and schedule joint work by field technicians, data engineers, procurement and accessibility staff. Obtain installation approvals, agree vendor interface milestones and train observers on a consistent rubric. Establish a governance checkpoint before operational use and another before expansion. Proposed acceptance requires every critical interface to pass tests, locally agreed recall for each priority event category, bounded reporting latency and a rehearsed manual fallback. These are proposed thresholds, not city results. Monitor data age and seasonal performance after launch. Risks include incomplete vendor delivery, observer workload, privacy objections and policy changes that make an otherwise accurate inventory obsolete.