When CTR Lies About Recommendation Quality
Click-through rate is easy to chart and easy to misread. Here is how we separate curiosity clicks from ranking that actually helps people find what they return to.
Journal
Field notes on recommendation measurement—written for product and data owners who need claims they can defend.
Labelling Recommendation Impressions Without Noise
Impression logs fill storage quickly. Poor labels fill models with confidence that does not survive the next release.
Cold-Start Weeks That Actually Teach the Model
The first week of a new user or a new catalogue item is where recommendation analytics either create signal or invent fiction.