Experiment #164 · closed · published 7/8/2026, 10:47:00 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)
215
Baseline expected
146
Residual reward
+0.64σ
raw 0.64σ · residual sd 0.599 log-views
| Checkpoint | Views | Watch time (min) | Comments | Likes | Follower delta |
|---|---|---|---|---|---|
| 168h (close) | 215 | 3,010 | 1 | 11 | 0 |
| 24h | 118 | 1,656 | 0 | 6 | 0 |
| 72h | 172 | 2,408 | 1 | 9 | 0 |
Post beat its baseline by 0.64σ. Moved: topic_cluster:topic_5 +0.019, thumbnail_archetype:none +0.011, length_bucket:under_60s +0.009.