Forecourt week · demo background
PathCI surface — this store’s forecourt patterns, not act-now. Not on the pitch path. This fold is for the presenter.
Talk track.
LiveOps Forecourt is what to do outside right now — blocked, bagged,
health. This page is the pattern altitude: which pump customers avoid,
what a fill really takes here, when the forecourt loads up, and whether
a bad hour was one stuck vehicle or real demand. No timers, no levers —
decisions, not interventions.
Claims.
The per-pump table, fill medians, utilization, and the hourly shapes
are measured numbers from a real pilot validation morning
(support/tasks/completed/forecourt-video-validation.md), shown here as one
Thursday. A fill is a merged occupancy episode ≥ 60s; a normal fill is
the pump’s own median. Excess counts only dwell past normal. High
excess on few fills = a stuck vehicle; high excess on many fills =
capacity. Pump 1 sits outside camera coverage and reads “no coverage”,
never zero traffic. Continuous presence tracking + sampled AI review.
Build readiness.
Everything on this page derives from shipped per-visit zone intervals;
the analysis exists as an offline script (forecourt_metrics.py). Not
built: this page itself, the avoided-pump flag history, and afternoon
coverage of the validation data (morning only).
Pump 5 does 2.6× the work of Pump 2.
| Pump | Fills | Dwell med | Dwell p90 | Occupied | Utilization | |
|---|---|---|---|---|---|---|
| Pump 1 | — | — | — | — | — | no camera coverage — not zero traffic |
| Pump 2 | 8 | 5m 45s | 9m 02s | 42m | 14% | slowest fills on the field |
| Pump 3 | 13 | 3m 58s | 10m 34s | 1h 05m | 21% | |
| Pump 4 | 10 | 4m 32s | 7m 50s | 52m | 17% | health-flagged 6 of 30 days — see LiveOps |
| Pump 5 | 21 | 4m 28s | 6m 03s | 1h 41m | 33% | first choice off the entrance |