Diagnose and repair

A machine is a tree of parts with a removal-order dependency graph, each part carrying a condition or fault state. A job arrives with symptoms; the player uses tests and tools to reveal hidden part states, removes covering parts in valid order, replaces or repairs faulty ones, reassembles, and verifies with a final test. In Car Mechanic Simulator 2018 and its 2021 sequel the player examines engine, suspension, and chassis, removes parts in a realistic order to reach a damaged component, buys a replacement or repairs it at higher levels, and uses a test track to find brake and suspension problems. PC Building Simulator applies the same loop to computers. Unlike diagnosis-treatment, the target is a machine whose state is fully deterministic once revealed. Inputs: fault state, symptoms, tools. Outputs: repaired machine, payment, and experience.

Methodical, calm problem-solving with a clear finish. Players narrow down causes from symptoms, enjoy the order of unbolting and rebolting parts, and see a fixed machine run again. Learning the real structure of an engine or PC is part of the appeal.

• how symptoms map to faults, and how many faults hide behind one symptom
• tool set and what each test reveals
• disassembly depth and how many parts block access
• repair versus replace economics
• job time limits and payment for partial fixes

• When the game highlights faulty parts directly, diagnosis disappears and only the removal chore remains.
• Deep disassembly for a single part becomes tedious once the novelty fades.
• Faults that do not follow from symptoms make diagnosis guesswork.

• Players build mental models of real machine structure and speed up dramatically with experience.
• Tool unlocks reshape the diagnosis strategy, since better tests skip disassembly.
• Coupled with economy and workshop progression, choosing which jobs to accept becomes a planning layer.

• Can fault states generated from a causal model of the machine, rather than authored per job, produce varied but solvable diagnoses?
• Could a language-model customer describing symptoms in their own words make diagnosis a real inference task?