Docs Simulator Blog About Github ↗

Topology and Density (Supplementary)

Mean Attack Rate by topology and fleet density. Companion table to the main paper, showing layout as a second-order resilience lever beneath fault category and density.

This is the full Topology × Density table that was promised in the paper as supplementary material. The headline takeaway: layout is a second-order resilience lever. Density and fault category dominate.

Mean Attack Rate by topology × density

Attack Rate is the cumulative fraction of the fleet ever materially affected by any fault cascade across the 500-tick run. Lower is better.

TopologyLow densityDefaultHigh densityOverall
SD-w1 (Single-Dock, 57×33)0.57 (n=20)0.55 (n=40)0.49 (n=60)0.54
DD (Dual-Dock, 61×33)0.56 (n=40)0.50 (n=80)0.44 (n=120)0.51

Both topologies show Attack Rate decreasing with density. More agents in the warehouse means the same fault event touches a smaller fraction of the fleet. The mechanism is denominator inflation, not improved resilience per agent.

What this means

  • Topology shifts Attack Rate by about 3 points overall. Dual-Dock has a small advantage at matched density. Not enough to make topology choice a primary lever for fault response.
  • Fault scenario dominates. Within each row, across-scenario spread is roughly 0.8, dwarfing the topology effect.
  • Density dominates the topology effect. Going from low to high density on the same layout shifts Attack Rate by 8 to 12 points, several times more than swapping layouts at fixed density.

Methodology

  • 6 fault scenarios per cell (Burst-20%, Burst-50%, Wear-Medium, Wear-High, Zone-Outage 50t, Intermittent 80s80m15r)
  • 3 solvers per cell (PIBT, RHCR-PBS, Token Passing)
  • 30 seeds per (solver × scenario × density × topology) cell
  • Token Passing excluded from n=120 cells per its completeness envelope of 100 agents
  • Mean over all (solver × scenario × seed) within each density column

Source data: results/warehouse_single_dock_experiment_summary.csv and results/warehouse_dual_dock_experiment_summary.csv in the MAFIS repository.

Why this is supplementary, not headline

A paper has a page budget. We kept the main findings (RQ1: solver vulnerability profile, RQ2: paradoxical FT > 1 regime under RHCR-PBS) in the printed text and moved second-order analyses here. The framework is the same, the experiments are the same, the data is the same. Only the page count is different.