Experiment #130 · closed · published 5/21/2026, 7:00:00 AM
Creative features
Confounds the baseline knew
Raw views mostly measure how big the channel already was. The baseline predicts what this channel “should” have earned given the confounds above; the residual is how the post actually landed relative to that prediction — measured in standard deviations.
Observed (168h)
55
Baseline expected
111
Residual reward
-1.18σ
raw -1.18σ · residual sd 0.599 log-views
| Checkpoint | Views | Watch time (min) | Comments | Likes | Follower delta |
|---|---|---|---|---|---|
| 168h (close) | 55 | 28,875 | 0 | 3 | 0 |
| 24h | 30 | 15,881 | 0 | 2 | 0 |
| 72h | 44 | 23,100 | 0 | 2 | 0 |
Post missed its baseline by 1.18σ. Moved: length_bucket:20m_plus -0.052, thumbnail_archetype:none -0.049, topic_cluster:topic_2 -0.044.