t05 — Elevator wait
Supplied working report; original verdicts not validated.
Review status: Needs review. This is an archival source note, not a passed proof.
notes.md
# t05 — Elevator wait `code.py` runs an exact discrete-event simulation: N floors at 3.3 m, c cars of capacity 13 at 2.5 m/s, door dwell 2.0 s per stop plus 1.0 s per transfer. Passengers arrive as a Poisson process; origins are uniform over floors, destinations over the other floors. A free car answers the longest-waiting call, boards up to 13 in that direction and sweeps, stopping at destinations and hall calls ahead. Each configuration (N = 6, 12, 24, 36; c = 1–6) gets 40 replications of a four-hour peak at five rates: 120 cells, 21.5 million journeys, no external data. Capacity comes from a probe: HC = 300 × 0.8 × 13 / RTT, and LF = arrivals per five minutes / HC. Wait tracks LF steeply (207 s per unit LF), yet LF alone explains only a third of the variance across configurations. Each extra floor adds 4.3 s at fixed LF (95% CI 3.2–5.4). The decisive matched-capacity comparison at LF = 0.8 gives 132 s for 12 floors against 223 s for 36 (gap 91 s, CI 84–98). Holding achieved car load at 0.45 gives 68 s against 163 s, so the height residual survives the capacity definition. At fixed car count height dominates: 21 s at six floors rises to 2,156 s at 36 because LF climbs from 0.57 to 1.51. The rate ratio is a strong first-order driver but not a sufficient one; height adds a large residual through longer trips and deadheads.