Computer Science > Artificial Intelligence
Title:Observer-Conditioned Planning for Intermittent Printer Faults
Abstract:A printer that jams for a user may nevertheless produce a normal diagnostic page when a technician arrives. We formulate the sequence as a finite-horizon partially observable planning problem in which the printer's objective is to withhold user jobs while passing service tests. The visitor is either the user or a technician, and the printer receives a noisy cue only after choosing whether to stage the obstructing sheet in an internal buffer. A staged sheet can be released or returned to the paper path after the cue. We solve the model exactly and compare immediate commitment, waiting, and conditional jam restoration. Under a balanced prior, an 80% accurate cue and a stipulated loss matrix, staging has expected loss 2.90, compared with 3.40 for waiting and 5.00 for immediate commitment. The technician sees a jam in 20% of technician visits under the staged policy, while 80% of returning users encounter a jam. We characterize the prior range favoring staging and examine cue quality, restoration cost and retained evidence. All results are exact expectations for a synthetic model, not observations of physical printers.
Submission history
From: Elin Morland[v1] Fri, 20 Nov 2026 21:16:41 UTC (139 KB)