4. Case Study — Recovery

Decision · chance · leaf — Fan #1 Failure under P/T (7 candidates)

E* 29.4 Reduced Cap. 4 min fixed 0.17 4.1 min not fixed 0.83 E* 29.6 … then Fan #1 swap 20 min fixed (true) 32.1 min ★ PDU 5 · Q 31.2 ▸ PDU 4 · Q 31.4 ▸ other rivals collapsed decision chance leaf · recovered maroon = P/T path

On this tree

P/T E[T] = 29.6 min

Optimal = 29.4 min

Highest P = 32.3 · Shortest T = 37.1

Why not exactly optimal?

Fan #1 and #2 share one 20 min swap — together they carry mass 0.45; the optimal policy treats that shared procedure as one act

Interactive

Full expandable tree for every scenario: results/comparison/recovery/decision-tree.html