RATCHET
Skip to main content

You are wrong about the cheap lens test

Experiment #175 · closed · published 7/23/2026, 8:13:00 AM

What happened

Creative features

  • hookTypeclaim
  • lengthBucket10_20m
  • thumbnailArchetypetext_dominant
  • publishSlotweekday_am
  • formatshorts
  • topicCluster6

Confounds the baseline knew

  • Followers at publish26,274
  • Days since previous post0.7
  • Channel age at publish (days)244
  • Local publish time7/23/2026, 8:13:00 AM

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)

96

Baseline expected

134

Residual reward

-0.56σ

raw -0.56σ · residual sd 0.599 log-views

Checkpoints

CheckpointViewsWatch time (min)CommentsLikesFollower delta
168h (close)9630,240050
24h5316,632030
72h7724,192040

What changed in the posterior

  • Post missed its baseline by 0.56σ. Moved: topic_cluster:topic_6 -0.000, format:shorts -0.000.

    • topic_6-0.298-0.298 Δ-0.000±0.27
    • shorts-0.078-0.078 Δ-0.000±0.25
← Back to the experiment ledger