Trust Score

Every AutoProctor report shows a Trust Score at the top. You will see it like the image below. A sample report can be found on this link. The Trust Score is a number between 0% and 100%. The lower the trust score, the more likely we think the candidate cheated on the test.

Of course, like all AI tools, AutoProctor isn't perfect. Even if a candidate's test attempt has a very low trust score, you absolutely must review the evidence to see if they actually cheated on the test. DO NOT go just by the Trust Score.

How is the Trust Score calculated?

As candidates take the test, AutoProctor monitors their environment real-time to see if they are cheating. Depending on the test's proctoring settings, it monitors:

their camera feed
their mic feed
the screen they are looking at, etc

You can imagine the AutoProctor software looking at this data very similar to how a human would monitor them. At the end of the test, our proprietary algorithm calculates a trust score based on all the evidence gathered. This trust score depends on the:

type of violation (Tab Switch leads to a higher trust Noise Detected)
frequency of violations (10 violations lead to a lower trust score compared to 5 violations)
duration of violations (Tab switch for 10 seconds leads to a lower Trust Score compared to 5 seconds)

What is a good Trust Score?

Of course, 100% is the ideal Trust Score. But, a rule of thumb is, if the Trust Score is lower than 85%, you may want to review the evidence.

This doesn't mean that if the Trust Score is 50%, the candidate cheated on the test. For example, if the candidate is attempting the test in a very noisy environment (say traffic noise), their Trust Score will be 0% because AutoProctor cannot distinguish between human and ambient noise. But, the candidate may not have cheated on the test at all!

You should always use the Trust Score as an indicator, but decide based on the evidence you see.

Updated on: 08/20/2024