01KV6J79N9MD3NVHG7Y0EED29F

c-resplan-crowd-geometry-gallery — session 01KV6J79N9MD3NVHG7Y0EED29F (refresh)

Re-run of the six-floor ResPlan grid (6090 tiny → 1374 largest × 2 intensities × 2 seeds = 24 runs, all gate-passed) after three fixes from the prior session (01KV6EFS...): agent_clearance_cm 5→2 (the safe band), a tight 30 s arrival window (overlapping crowd, not a trickle), and walls/doors drawn in the replay + heatmap.

C1 (movement surfaces) — MET. All 24 runs emit trajectory.html, footfall_heatmap.png, speed_vs_time.png + parquet. Replays now render walls as dark segments (doorways = the gaps) and show 3–6 concurrent agents per floor (was 1–2) — a readable crowd.

C2 (kinematic + walkable gate) — MET on all 24. n_outside_walkable == 0, mean_walk_speed 1.2 m/s everywhere.

C3 (traversal closes at i1.0) — MET on all 6 floors. through-traffic departed == spawned == 12, turned_away == 0 on every floor at intensity 1.0. The prior session's resplan-16157 near-trap (11/12, a 147 s wedge in an over-eroded 32 cm-wall doorway) is resolved: at clearance 2 the doorway is passable and 16157 clears 12/12, mean traversal 35 s → 12.6 s.

C4 (geometry→flow signal) — MET. Mean per-agent traversal time rises with floor size: Spearman(bbox_area, mean_traversal_s) = 0.886 (p = 0.019) across the six floors (6090 5.6 s → 1374 15.0 s), now clean (no outlier). Congestion onset is also visible: at 2× load, full clearance degrades on the busier floors (10425/16157/6090: 22–23/24 still on-floor at sim end) while the others clear 24/24.

Verdict. The floor-agnostic crowd scenario generalises wall-tight and in-band across the ResPlan corpus, traversal cost tracks geometry, and the one prior exception is fixed. Supports the Indoor Crowd Modeling generalisation claim. Attached runs: 24 (phase refresh2). Per-floor replays at sim://run/<id>/artefact/trajectory.html.

Critic notes (self-review)

• agent_clearance_cm is a fragile knob, not a solved parameter. 2.0 works across these six floors, but the JuPedSim segfault at clearance 0 (and 3) on resplan-10425 is non-monotonic — a different floor could crash or wedge at 2.0. The result is "clearance 2 happens to clear these six," not "the walker is robust." The real fix is a pre-flight walkable-QC (min-passage width + degeneracy repair); until then every new floor needs a gate run.
• Still free-flow. mean_walk_speed is 1.2 m/s on every run, including 2× load — these apartments don't reach a congestion density. C4's congestion clause is supported only by the non-departure counts at 2× (agents still in transit at sim end), not by any speed slowdown. Genuine FD-regime crowding needs far higher agent counts than a residential floor admits.
• Spearman on n=6 (p=0.019) is directionally solid but a small sample; driven partly by 1374's size.
• 2× non-departures unverified. The 22–23/24 at intensity 2.0 are reported as congestion but not root-caused — could be mild doorway queueing or a residual near-trap under load. Worth a glance before citing as a clean FD signal.