Evacuation egress on ResPlan floors — emptying the rooms through the doorways

Question

When a furnished-or-not floor has to empty — every occupant leaving at once for a single exit — does the walker reproduce the classic egress signature (doorway queueing, a clearance time that grows super-linearly with crowd size), and can we watch the floor drain in the replay? This exercises the crowd simulator in the regime crowd-safety research cares about most, on real layouts.

What we already know

• jebrane2026_3e34 ↗ — flow peaks near 2 ped/m² and congestion grows nonlinearly past it; an egress through a single doorway is exactly where this bites. At intensity 2.0 (48 agents converging on one exit corridor) we expect a visible super-linear clearance time (criterion 4).
• maury2018_d24a ↗ — the free-walk band (criterion 2) and the fact that egress flow is exit-width-limited: clearance time should scale with population / effective doorway capacity, not with floor area alone.
• c-crowd-motion-mix established the groups scenario layer + the movement surfaces; egress is a one-group special case (spawn-everywhere, single goal), so this is config-only on top of the 2026-06-15 symbolic-region walker — no new engine code.
• zhong2022_7cb2 ↗ — clearance-time-vs-load and in/out-flow counts are the standard validation instruments for simulated egress; the criteria here are those instruments on furnished real floors.

Prerequisites

1. The rebuilt walk-notebook image with symbolic regions (floor_walk.py::_resolve_region, 2026-06-15) — spawn_region: all + goal_region: west are resolved per floor. Same prerequisite as c-resplan-crowd-geometry-gallery; GHCR :latest predates it.
2. No furnishing needed — egress is a bbox/region goal; seats are irrelevant (on resplan-12439 the seats are simply unused — pure egress, no dwelling).
3. Per-floor staging automatic via stage_floor_bundle.

What the supervisor does

1. Phase A — gate. One run on resplan-1374 (the most complex floor) at intensity 1.0, seed 0. Confirm surfaces + kinematic gate + that departed == spawned (the exit region is reachable from every room). A non-clearing gate on the hardest floor is a walkable-connectivity finding — fail fast.
2. Phase B — grid. Remaining cells (3 floors × 3 intensities × 2 seeds − gate). Label {campaign: c-resplan-egress, floor, intensity, seed}.
3. Phase C — synthesis. Per floor + intensity: clearance time and the load-monotonicity (criterion 4). The headline is the intensity-2.0 trajectory.html on resplan-1374 — the floor draining through its choke points.

Figure render request

resplan_egress — (a) clearance time vs intensity per floor (the super-linear curve), (b) the footfall heatmaps at intensity 2.0 side by side (doorway hotspots), (c) on-floor population vs time per floor (the egress decay curves) — plus a pointer to the headline replay.

Out of scope

• Tagged entrances. The exit is the west region proxy until transition subtypes (main_entry) are tagged in PostGIS; a real door-width study waits for that. State the proxy in the synthesis.
• Panic / emotional contagion / variable exit width. This validates baseline egress kinematics, not behavioural models — those are a follow-on once the baseline holds.
• Multi-exit. Single-exit only; multi-exit choice models are out of scope here.

Expected interpretation

1. All criteria met → "The walker reproduces full, wall-tight, load-responsive egress on real multi-room floors with visible doorway bottlenecks." Establishes egress as a first-class scenario for the crowd-safety thread and a stressor for downstream CSI (a draining floor is a strong non-stationary occupancy signal).
2. Clearance fails at high intensity on a floor → a bottleneck finding: the exit corridor's effective capacity is below the converging demand. Quantify it (clearance plateau / turned_away) and cite the replay — that's a result, not a failure.
3. Kinematic/connectivity gate fails → a geometry-import finding for that floor (a room not connected to the exit region in the largest-part walkable). Report per floor.