4. Case Study — Recovery

Same KG scores, four orderings — Fan #1 Failure

Candidates at the first decision (after 60 s gate)

HypothesisPTP/T
TCCS Aux. Fan #1 Failure ★0.372200.019
PDU 5 Failure0.05660.009
TCCS Aux. Fan #2 Failure0.164200.008
PDU 4 Failure0.04860.008
TCCS Aux. Fan at Reduced Capacity0.03040.007
Reduced Main Cabin Fan #1 Capacity0.03040.007
TCCS Filter Clog0.123180.007
Trace Contaminants0.123200.006
Fuel Cell #1 and PDU Failure0.056140.004

Diagnosis is held fixed (knowledge graph). Only the recovery ranking changes.

What each ordering tries

OrderingSequenceTTR
P/TFan #1 (first)21 min
Highest PFan #1 (first)21 min
Shortest TReduced Cap. → PDU → … → Fan #161 min
RandomPDU → Trace → … (timeout)—

After the timed gate, Fan #1 leads on both P and P/T, so those two orderings recover on the first try. Shortest T still burns cheap wrong procedures; this random seed never reaches the true swap in time.