Experiment #129 · closed · published 5/20/2026, 11:36:00 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)
189
Baseline expected
127
Residual reward
+0.66σ
raw 0.66σ · residual sd 0.599 log-views
| Checkpoint | Views | Watch time (min) | Comments | Likes | Follower delta |
|---|---|---|---|---|---|
| 168h (close) | 189 | 2,646 | 1 | 9 | 0 |
| 24h | 104 | 1,455 | 0 | 5 | 0 |
| 72h | 151 | 2,117 | 1 | 8 | 0 |
Post beat its baseline by 0.66σ. Moved: hook_type:number_list +0.035, length_bucket:under_60s +0.033, topic_cluster:topic_1 +0.033.