Build a Video to Shorts Pipeline
Step 8 of 9 · About 5 minutes per asset
Score Every Asset Against an 85 Floor
Two rubrics, one loop, one rule: 85 or it does not ship. Three fix passes maximum, then it is killed or flagged. This is what stops the pipeline quietly degrading.
A scored, fixed, and gated set of finalists, plus an honest kill list.
Everything before this produced candidates. This step decides what actually ships. Without it, week three looks like week one with worse clips, because nothing was measuring the drift.
The shorts rubric
| Dimension | Weight | What it measures |
|---|---|---|
| Hook strength | 25 | Do the first two seconds stop the scroll |
| Standalone coherence | 25 | Can a cold viewer follow all of it |
| Payoff delivered | 20 | Does the thesis actually land by the end |
| Pacing and dead air | 15 | Any slack that should have been cut |
| Caption accuracy and readability | 10 | Do captions match the audio and stay readable |
| CTA clarity | 5 | Is the next action obvious |
The longform rubric
| Dimension | Weight | What it measures |
|---|---|---|
| Hook strength | 20 | The cold open from step 7 |
| Dead air remaining | 20 | What the rough cut missed |
| Pacing | 20 | Whether any saggy block survived |
| Audio levels and consistency | 15 | Anyone noticeably louder or quieter |
| Caption accuracy | 15 | Against the audio |
| Chapter quality | 10 | Titles and boundaries that actually help |
The loop
- Score. 85 or better makes it a finalist. Done.
- Below 85, identify the single weakest dimension and apply one targeted fix. Re cut, re order the hook, tighten, fix captions. Re render, re score.
- Three passes maximum. Still below 85 means kill it, which is the default, or flag it to a human if it is 75 or better and the idea is genuinely strong.
- Never pad. Adding length to a weak clip lowers the score, it does not raise it.
Create skills/shorts-editor-loop/SKILL.md and
skills/openmic-editor-loop/SKILL.md, then run the shorts one.
shorts-editor-loop scores a rendered short 0 to 100:
Hook strength (first 2 seconds stop the scroll) 25
Standalone coherence (cold viewer follows it) 25
Payoff delivered (thesis lands) 20
Pacing and dead air 15
Caption accuracy and readability 10
CTA clarity 5
Coherence is ALSO a hard gate: any orphaned reference is an automatic fail
regardless of total score.
The loop:
1. Score. 85 or better means finalist.
2. Below 85, name the weakest dimension, apply ONE targeted fix, re render,
re score.
3. Maximum 3 passes. Still below 85 means KILL by default, or flag to me with
the score breakdown if it scored 75 or better and the idea is strong.
openmic-editor-loop is the same loop for the longform episode with these
weights: hook 20, dead air remaining 20, pacing 20, audio levels 15, caption
accuracy 15, chapter quality 10. Route each weakness back to the skill that
owns it rather than patching the output.
Now score every rendered short in episodes/<date>/shorts/ and give me a table:
clip id, total, each dimension, verdict (finalist, fixing, killed), and for
anything below 85 the ONE fix you are applying next.
Be honest. A generous score here costs me the whole week.Questions
Why 85 and not 70?
Because 70 is the score of something that is fine, and fine does not get watched or shared. The number matters less than the fact that it is fixed in writing and applied every week regardless of how busy you are.
Is it really better to ship 4 shorts than 10?
Yes. Two great shorts beat six acceptable ones, because the acceptable ones train the algorithm and your audience that your clips are skippable. The quota is met by mining more ideas, never by lowering the floor.
Talk it through
Stuck on this step?
Free 30 minutes. Bring the error, the transcript, or the clip that will not pass the gate, and we will work through it.