This is the third post in a set on disciplined diagnosis, after Root-Cause Analysis Under Uncertain, Heterogeneous Evidence and When Is the Evidence Enough?. Those built a method for weighing messy evidence and deciding when you have enough of it. To keep that method clean, they leaned on three assumptions — and real systems break all three.

The assumptions are easy to miss because they feel like common sense: that there is one fault to find, that the fault stays put long enough to measure, and that your sensors tell the truth. Drop any one and the picture changes. Two faults at once produce a symptom pattern that matches neither alone. An intermittent fault hides between your snapshots, so more data of the wrong kind never helps. And a lying sensor turns a healthy system into a phantom fault — or hides a real one.

This post takes the three cases one at a time, in deliberately simple language, and shows how each folds back into the sufficiency method rather than replacing it.

What it covers

Nine sections, about twenty-two minutes.

§ 1 — Three assumptions. The comfortable defaults, and the symptom that each failure leaves behind.

Part A · Multiple simultaneous faults

§ 2 — Symptoms add up. Why two faults look like neither, and why the unexplained leftover (the coverage check from the prequel) is your early warning.

§ 3 — Reasoning about combinations. Generating candidate explanations from conflicts, minimal diagnoses and the parsimony trap, and taming the combinatorial blow-up — with a worked example.

Part B · Intermittent faults & operating-condition coverage

§ 4 — Richness is coverage, not count. The operating envelope, why a thousand idle logs can’t diagnose a fault that only appears on a hot uphill, and capturing the event.

§ 5 — Getting the data you actually need. Condition-coverage checks and targeted collection — driving the system to the corner where the fault lives.

Part C · Sensors that lie

§ 6 — Is the gauge broken? Telling a sensor fault from a system fault: hardware and analytical redundancy, plausibility limits, and parity/voting.

§ 7 — The augmented procedure. A single flowchart folding all three cases back into the sufficiency gate.

§ 8 — Practical notes. Rules of thumb for the hard cases.

§ 9 — References & further reading. Model-based diagnosis and fault detection, with pointers.

Read it

Open the monograph →

The monograph lives at its own URL in the warm-paper layout of the series, with symptom-superposition and operating-envelope figures, a sensor-validation block diagram, an augmented-procedure flowchart, three algorithms, worked EV examples, practical notes, and references.

Where it sits

The diagnosis series, in order:

I · The core method

  1. Root-Cause Analysis Under Uncertain, Heterogeneous Evidence
  2. When Is the Evidence Enough?
  3. Harder Cases in Diagnosis — (this post)

II · At fleet scale

  1. Fleet-Scale Root-Cause Analysis
  2. Fleet RCA, Made Rigorous

III · Proof, foundations & prevention

  1. Proving Cause with Experiments
  2. Diagnosis on the Vehicle
  3. The Data Plumbing Behind Fleet Diagnosis
  4. From Root Cause to No Cause

Full map: The Diagnosis Series index.

← Back to Autonomy