Signal Detection, explained.
Signal detection theory separates sensitivity to a signal from the decision threshold used to report it, exposing the tradeoff between misses and false alarms.
Why it happens
When signal and noise overlap, no threshold perfectly separates them. Lowering the threshold catches more signals but also flags more noise. Better separation can improve both; merely changing the threshold usually moves along a tradeoff.
Detection separates the quality of evidence from the cost of acting on it.
Read the result
Adjust signal strength separately from the threshold. Track hits, misses, and false alarms together: a high hit rate alone does not show that a flag is reliable.
A worked example
An alert for unusual activity
A monitoring system assigns scores to both ordinary activity and genuine incidents.
Lowering the alert threshold catches weaker incidents and also admits more ordinary activity.
Whether that change helps depends on the frequency of incidents and the costs of investigating alerts versus missing them.
OPTIONAL DEEPER DETAILGo deeper: inside the model
Inside this model
Noise scores are uniform 0–100. Signal scores are uniform between max(0, signal strength − 25) and min(100, signal strength + 25). A score at or above the threshold is flagged. The chart shows exact hit and false-alarm probabilities.
Where this idea is useful
A practical use
Moderating content where a strict rule misses some harmful items and a loose rule flags harmless ones.
A common misconception
“More alerts mean a more accurate detector.”
Alert volume can rise simply because the threshold is lower. Sensitivity and decision policy are different properties.
What this explanation leaves out
- Uniform toy score distributions and equal cases are pedagogical, not a calibrated real detector.
Why can a detector with many hits still produce mostly false alerts?
If genuine events are rare, the much larger noise population can contribute many false alarms. You need the base rate as well as hit and false-alarm rates.
Which mistake is more costly here, and how common is the signal?
Further reading
Explore the original research or the teaching reference behind this experiment.