Second-Rater Spot-Check of AI Scoring and B1/B2 DriftSecond-Rater Spot-Check of AI Scoring and B1/B2 Drift
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Second Rater notes #1: an AI scorer can agree with examiners on average and still be wrong in one predictable place.
In this illustrative Spot-Check (fictional app, synthetic data), the scorer matched a blind second rater on 11 of 20 workplace emails. Seven of the nine disagreements went the same way: the AI scored higher, mostly at B1, where long, fluent emails skipped a required recommendation and still earned B2.
The fix is not a new model. It is a task-completion gate, a re-anchored B1/B2 boundary and length-matched probes in the regression set.
If you run an AI tutor or scorer and want to know where yours drifts, the Second Rater Spot-Check takes five days.
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started