Docs/Skills and code/skill script-cohort-analysis

.claude/skills/script-cohort-analysis/references/pair-prompt.md

Markdown46 lines2.4 KBSkill: script-cohort-analysis

Blind pairwise script comparison

You will read two transcripts from one YouTube channel, labelled A and B. You're blind to titles and performance, so don't guess them. Compare the two scripts on each dimension. Judge only what the words, order and timing show, ignoring visuals and caption errors.

For each dimension, decide which script does it better: "A", "B" or "tie". Use "tie" only when you truly can't separate them, and add a one-sentence reason that cites a timestamp from each script.

Dimensions:

• hook: the first ~60s gives a specific reason to keep watching, confirms what the video promised, and hints at where the answer is going without giving the route.
• curiosity: there is almost always an open question the viewer wants answered.
• causality: each section follows from the last by consequence ("therefore") or complication ("but"), never just "and then". If you moved sections around, would it break?
• value_density: every line earns its place: specific, non-obvious, not padded or repeated.
• clarity: a smart newcomer can follow it.
• emotion_story: people, stakes and scenes carry the content.
• voice: talks to the viewer, has a point of view.
• escalation: interest rises through the runtime.
• payoff: the opening promise is answered completely and specifically.
• credibility: claims are supported and caveated. No sensationalism.
• stakes_load: figures arrive after a reason to care and never pile up; terms are explained when first used; a smart newcomer is never asked to hold more than a few new things at once.

Then give an overall winner: which script you expect to hold and satisfy viewers better.

Finally, are these two scripts the same kind of video (format and genre), so a comparison is fair? Set same_format: true/false and say what each one is.

Output one JSON file:

{
  "pair_id": "P01",
  "a": "S3", "b": "S5",
  "dims": { "hook": { "winner": "A", "reason": "…" } },
  "overall": { "winner": "B", "reason": "…" },
  "same_format": false,
  "formats": { "A": "fiction comparison", "B": "personal-finance explainer" },
  "confidence": "low | medium | high"
}

dims must contain all 11 dimensions. Run the validation command from your task file until it prints OK. Your final message is one line: the path and the validator result.

# Blind pairwise script comparison

You will read two transcripts from one YouTube channel, labelled **A** and **B**. You're
blind to titles and performance, so don't guess them. Compare the two **scripts** on each
dimension. Judge only what the words, order and timing show, ignoring visuals and
caption errors.

For each dimension, decide which script does it better: `"A"`, `"B"` or `"tie"`. Use
`"tie"` only when you truly can't separate them, and add a one-sentence reason that cites
a timestamp from each script.

Dimensions:
- **hook**: the first ~60s gives a specific reason to keep watching, confirms what the video promised, and hints at where the answer is going without giving the route.
- **curiosity**: there is almost always an open question the viewer wants answered.
- **causality**: each section follows from the last by consequence ("therefore") or complication ("but"), never just "and then". If you moved sections around, would it break?
- **value_density**: every line earns its place: specific, non-obvious, not padded or repeated.
- **clarity**: a smart newcomer can follow it.
- **emotion_story**: people, stakes and scenes carry the content.
- **voice**: talks *to* the viewer, has a point of view.
- **escalation**: interest rises through the runtime.
- **payoff**: the opening promise is answered completely and specifically.
- **credibility**: claims are supported and caveated. No sensationalism.
- **stakes_load**: figures arrive after a reason to care and never pile up; terms are explained when first used; a smart newcomer is never asked to hold more than a few new things at once.

Then give an **overall** winner: which script you expect to hold and satisfy viewers better.

Finally, are these two scripts **the same kind of video** (format and genre), so a
comparison is fair? Set `same_format: true/false` and say what each one is.

Output one JSON file:

```json
{
  "pair_id": "P01",
  "a": "S3", "b": "S5",
  "dims": { "hook": { "winner": "A", "reason": "…" } },
  "overall": { "winner": "B", "reason": "…" },
  "same_format": false,
  "formats": { "A": "fiction comparison", "B": "personal-finance explainer" },
  "confidence": "low | medium | high"
}
```

`dims` must contain all 11 dimensions. Run the validation command from your task file
until it prints OK. Your final message is one line: the path and the validator result.

Shown as it is in the repository at build time (2026-11-03); the home directory appears as <home>. If this page and the file differ, the file is right. Back to the catalogue.