Skip to content
OkoleHome

Bayesian diagnostic

The engine starts from observed fault frequencies, adjusts them to the device's context, then updates its probabilities after every test. It picks the next test by maximising the information gained per minute of technician time.

Probability

14.2%

Battery

Remaining uncertainty

3.7 bits

3.7 bits — 0 means full certainty

Projected test plan

12 min

Following the most likely outcome at each step, the engine reaches 36% confidence in 12 minutes.

Recommended next tests

The engine starts from observed fault frequencies, adjusts them to the device's context, then updates its probabilities after every test. It picks the next test by maximising the information gained per minute of technician time.

Power on

What happens when the power button is pressed?

Technician observation
Information gain
0.42 bits
Cost
2 min
Yield
0.211 bits/min
What happens when the power button is pressed?

Fan behaviour

How do the fans behave?

Technician observation
Information gain
0.32 bits
Cost
2 min
Yield
0.162 bits/min
How do the fans behave?

Display output

What does the built-in screen show?

Technician observation
Information gain
0.31 bits
Cost
2 min
Yield
0.155 bits/min
What does the built-in screen show?

Fault hypotheses

Battery14.2%

Prior 14.2%Repair estimate ≈ 40 minPart neededImpact : Major

Thermal management12.6%

Prior 12.6%Repair estimate ≈ 55 minNo partImpact : Major

Corrupted operating system11.8%

Prior 11.8%Repair estimate ≈ 90 minNo partImpact : Major

Mechanical hard drive11.4%

Prior 11.4%Repair estimate ≈ 75 minPart neededImpact : Blocking

Malware7.2%

Prior 7.2%Repair estimate ≈ 70 minNo partImpact : Major

Memory7.1%

Prior 7.1%Repair estimate ≈ 30 minPart neededImpact : Blocking

Mainboard6.3%

Prior 6.3%Repair estimate ≈ 180 minPart neededImpact : Blocking

Power supply5.5%

Prior 5.5%Repair estimate ≈ 45 minPart neededImpact : Blocking

What the engine does not do

  • The technician keeps the final call. Their corrections feed the learning loop.
  • Probabilities are calibrated: a stated 70 % must be right seven times out of ten. Brier score and reliability curve are reported to the quality committee.
  • Symptoms are not independent. Correlated tests are grouped and redundant evidence is discounted — a controlled approximation, not a hidden assumption.
  • No hypothesis is ever permanently ruled out by a single observation: smoothing keeps it recoverable.