RATCHET
Skip to main content

Live rebuild of my audio chain

Experiment #147 · closed · published 6/14/2026, 1:13:00 PM

What happened

Creative features

  • hookTypedemo_first
  • lengthBucket10_20m
  • thumbnailArchetypetext_dominant
  • publishSlotweekend_pm
  • formatlist
  • topicCluster5

Confounds the baseline knew

  • Followers at publish23,194
  • Days since previous post1.2
  • Channel age at publish (days)205
  • Local publish time6/14/2026, 1:13:00 PM

The residual reward

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)

183

Baseline expected

159

Residual reward

+0.23σ

raw 0.23σ · residual sd 0.599 log-views

Checkpoints

CheckpointViewsWatch time (min)CommentsLikesFollower delta
168h (close)18357,645190
24h10131,705050
72h14646,116170

What changed in the posterior

  • Post beat its baseline by 0.23σ. Moved: topic_cluster:topic_5 -0.001, format:list -0.001, publish_slot:weekend_pm -0.001.

    • topic_50.0090.007 Δ-0.001±0.29
    • list-0.035-0.036 Δ-0.001±0.28
    • weekend_pm0.2440.243 Δ-0.001±0.27
    • demo_first-0.089-0.090 Δ-0.001±0.26
    • text_dominant0.0500.049 Δ-0.000±0.25
    • topic_6-0.425-0.424 Δ+0.000±0.31
    • 10_20m0.0750.075 Δ-0.000±0.25
    • vlog-0.146-0.146 Δ+0.000±0.26
    • weekday_pm-0.016-0.016 Δ+0.000±0.24
    • weekend_am0.0950.095 Δ+0.000±0.29
    • interview0.2260.227 Δ+0.000±0.25
    • 4_10m-0.099-0.098 Δ+0.000±0.26
    • story_cold_open0.2390.239 Δ+0.000±0.26
    • 20m_plus-0.092-0.092 Δ-0.000±0.27
    • claim-0.174-0.174 Δ+0.000±0.26
    • topic_30.2570.257 Δ+0.000±0.26
    • commentary-0.069-0.069 Δ+0.000±0.27
    • number_list-0.114-0.114 Δ+0.000±0.26
    • topic_1-0.099-0.098 Δ+0.000±0.25
    • question0.1450.145 Δ-0.000±0.27
    • before_after0.1010.101 Δ+0.000±0.28
    • topic_40.2460.246 Δ+0.000±0.26
    • contrarian-0.022-0.022 Δ+0.000±0.27
    • tutorial0.0090.009 Δ+0.000±0.25
    • weekday_am-0.130-0.130 Δ+0.000±0.26
← Back to the experiment ledger