LLumen

Ethical Design

System Tradeoff Studio

Set thresholds under uncertainty: false alarms vs missed harm, trust vs safety theater.

System Tradeoff Studio

Decisions under uncertainty

You are setting the sensitivity of a school chat safety filter. There is no free lunch: catch more harm and you will flag more innocent messages.

Scenario

A district wants AI to flag potentially harmful student chat. True harmful messages are rare. False alarms annoy students and bury real alerts. Missed harm is dangerous. You must pick a threshold and defend it.

Filter strictness50
Lenient (fewer flags)Strict (more flags)

Harmful messages caught

70%

False alarms on OK chat

33%

Missed harmful messages

30%

Student trust (model)

64%

Safety board score (model)

69%