RATCHET
Skip to main content

What to make next

Each brief is drawn from the model rather than from its current favourite, so the schedule keeps testing instead of converging on one answer too early.

reading the live Worker · 0 of 20 exploratory ·

  • Make a list format with a question-based opener, a text-dominant thumbnail, 10-20 minutes, posted weekend afternoons, around topic cluster 3.

    predicted +0.21 [-0.66, 1.07]

    Scores highest under the current posterior draw; the deciding features are topic_cluster:topic_2, hook_type:question, publish_slot:weekend_pm.

    exploit · blocked

  • Make a list format with a question-based opener, a face-reaction thumbnail, 10-20 minutes, posted weekday afternoons, around topic cluster 4.

    predicted +0.56 [-0.29, 1.41]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, topic_cluster:topic_3, thumbnail_archetype:face_reaction.

    exploit · blocked

  • Make a list format with a question-based opener, a face-reaction thumbnail, 10-20 minutes, posted weekend afternoons, around topic cluster 6.

    predicted +0.60 [-0.25, 1.44]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, thumbnail_archetype:face_reaction, publish_slot:weekend_pm.

    exploit · blocked

  • Make a shorts with a cold-open story, a before/after thumbnail, under 60 seconds, posted weekday mornings, around topic cluster 8.

    predicted +0.37 [-0.41, 1.16]

    Scores highest under the current posterior draw; the deciding features are topic_cluster:topic_7, hook_type:story_cold_open, length_bucket:under_60s.

    exploit · blocked

  • Make a shorts with a cold-open story, a before/after thumbnail, under 60 seconds, posted weekend afternoons, around topic cluster 5.

    predicted +0.54 [-0.27, 1.35]

    Scores highest under the current posterior draw; the deciding features are hook_type:story_cold_open, topic_cluster:topic_4, length_bucket:under_60s.

    exploit · blocked

  • Make a shorts with a cold-open story, a before/after thumbnail, under 60 seconds, posted weekend mornings, around topic cluster 2.

    predicted +0.42 [-0.41, 1.25]

    Scores highest under the current posterior draw; the deciding features are hook_type:story_cold_open, publish_slot:weekend_am, length_bucket:under_60s.

    exploit · blocked

  • Make a shorts with a contrarian take, no thumbnail portrait, 1-4 minutes, posted weekend mornings, around topic cluster 2.

    predicted +0.16 [-0.68, 1.01]

    Scores highest under the current posterior draw; the deciding features are publish_slot:weekend_am, thumbnail_archetype:none, hook_type:contrarian.

    exploit · blocked

  • Make a shorts with a cold-open story, no thumbnail portrait, 1-4 minutes, posted weekend mornings, around topic cluster 6.

    predicted +0.49 [-0.41, 1.39]

    Scores highest under the current posterior draw; the deciding features are hook_type:story_cold_open, publish_slot:weekend_am, topic_cluster:topic_5.

    exploit · blocked

  • Make a shorts with a cold-open story, no thumbnail portrait, 1-4 minutes, posted weekend mornings, around topic cluster 6.

    predicted +0.49 [-0.41, 1.39]

    Scores highest under the current posterior draw; the deciding features are hook_type:story_cold_open, publish_slot:weekend_am, topic_cluster:topic_5.

    exploit · blocked

  • Test draft with question opener

    predicted -0.02 [-0.76, 0.71]

    Scores highest under the current posterior draw; the deciding features are topic_cluster:topic_2, hook_type:question, thumbnail_archetype:face_reaction.

    exploit · blocked

  • Make a list format with a question-based opener, an object-hero thumbnail, 10-20 minutes, posted weekend afternoons, around topic cluster 1.

    predicted +0.05 [-0.80, 0.91]

    Scores highest under the current posterior draw; the deciding features are thumbnail_archetype:object_hero, hook_type:question, publish_slot:weekend_pm.

    exploit · blocked

  • Make a list format with a question-based opener, a text-dominant thumbnail, 10-20 minutes, posted weekday afternoons, around topic cluster 2.

    predicted +0.30 [-0.48, 1.07]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, thumbnail_archetype:text_dominant, length_bucket:10_20m.

    exploit · blocked

  • Make a list format with a question-based opener, a text-dominant thumbnail, 10-20 minutes, posted weekday afternoons, around topic cluster 2.

    predicted +0.30 [-0.48, 1.07]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, thumbnail_archetype:text_dominant, length_bucket:10_20m.

    exploit · blocked

  • Make a commentary with a numbered list opener, a face-reaction thumbnail, 10-20 minutes, posted weekday evenings, around topic cluster 5.

    predicted +0.09 [-0.70, 0.88]

    Scores highest under the current posterior draw; the deciding features are topic_cluster:topic_4, publish_slot:weekday_late, thumbnail_archetype:face_reaction.

    exploit · blocked

  • Make a commentary with a question-based opener, a face-reaction thumbnail, 20+ minutes, posted weekday evenings, around topic cluster 8.

    predicted +0.29 [-0.48, 1.06]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, topic_cluster:topic_7, publish_slot:weekday_late.

    exploit · blocked

  • Make a commentary with a question-based opener, a face-reaction thumbnail, 20+ minutes, posted weekday evenings, around topic cluster 8.

    predicted +0.29 [-0.48, 1.06]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, topic_cluster:topic_7, publish_slot:weekday_late.

    exploit · blocked

  • 20 minute videos are dead for good

    predicted -0.05 [-0.78, 0.67]

    Scores highest under the current posterior draw; the deciding features are topic_cluster:topic_2, hook_type:question, length_bucket:20m_plus.

    exploit · blocked

  • Make a shorts with a question-based opener, a face-reaction thumbnail, under 60 seconds, posted weekday afternoons, around topic cluster 6.

    predicted +0.47 [-0.22, 1.17]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, length_bucket:under_60s, thumbnail_archetype:face_reaction.

    exploit · proposed

  • Make a tutorial with a question-based opener, a face-reaction thumbnail, under 60 seconds, posted weekend mornings, around topic cluster 1.

    predicted +0.58 [-0.17, 1.33]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, publish_slot:weekend_am, length_bucket:under_60s.

    exploit · proposed

  • Make a vlog with a question-based opener, no thumbnail portrait, 4-10 minutes, posted weekend afternoons, around topic cluster 6.

    predicted +0.33 [-0.47, 1.13]

    Scores highest under the current posterior draw; the deciding features are hook_type:question, publish_slot:weekend_pm, length_bucket:4_10m.

    exploit · proposed

Exploratory briefs are capped. An unconstrained sampler would happily spend a real channel on experiments, which is optimal for the model and bad for the creator.