Campaign session

01KXJQAMFKD1T05KX78TPR0VX6

cancelled 2026-07-15 11:06:55.347694+00:00 → 2026-07-15 11:24:17.532319+00:00 · 0 runs · supervisor: react-agent

“CANCELLED (0 runs, no compute) pending MCP-server reload — jamming fix (AVM model + give-up watchdog) is code-complete + unit-tested but the new WalkScenario fields (model/stuck_timeout_s/stuck_progress_m) are rejected by the live server's cached ParamsSchema. Reopen after restart.”

Archive snapshot, as of 22 h ago — the run corpus is rebuilt once a day, so this page is not a live reading. The fleet panel is the live one; it refreshes every 30 s.

Synthesis

Session cancelled — clean partial, 0 runs, no compute committed

This session opened to run the powered, rendered-evidence re-run of the continuity-prior CSI-density refutation with big crowds and the literature-grounded jamming fix the operator requested mid-session. It is cancelled (not failed) because a live-server infra constraint blocked launch before any run; the science work is intact and durable.

Durable work landed this session (outside the session record):

Diagnosed the crowd-jamming root cause: engine.py's loop removes an agent only on goal-arrival — no stuck handling — so a JuPedSim CollisionFreeSpeedModel deadlock in a narrow passage freezes a permanent static high-density blob into the density field (corrupting exactly the high-density regime this PINN needs).
Grounded the fix in vault literature (read in full): xu2019_05d4 (Anticipation Velocity Model / generalized CVM — ellipse + new direction sub-model kills the backward-movement/shaking deadlock and fits the congested fundamental diagram), helbing2000_a943 (faster-is-slower — don't over-drive v0), moussad2011_fa42 (beyond capacity flow stops; agents abandon blocked routes).
Implemented a parity-safe, config-gated fix: selectable model: cfsm|avm (default cfsm → CI parity holds; dense crowds set avm) in locomotion.py, a give-up-and-leave watchdog (stuck_timeout_s) in engine.py, wired through floor_walk.py, vendored into the walk-notebook image scripts, and added to WalkScenario (reused by exp-csi-crowd). 30 behaviour/engine/agenda unit tests pass; default path bit-exact.
Also extended continuity_density_pinn.py from 2-arm to the decisive 5-arm reduction (smoothing baseline + 3 FD closures + closure-invariance + advection index + per-seed scalars).

Blocker: the live MCP server caches the Pydantic ParamsSchema at import, so the new WalkScenario fields are rejected until it reloads. Resolution: restart the MCP server (or run in CI, which starts fresh); the big-crowd AVM corpus then runs cleanly. Reopen a fresh session then. All four criteria unresolved — a first-class partial.

Attached runs

The attached runs are not in the current atlas snapshot — rebuild via web --build-sim-atlas.