skills/analyze-any-video/references/analysis-rubric.md

# Multimodal analysis rubric

Use only relevant sections. Scores are optional and must be justified with timestamped evidence.

## Meaning and audience

- Apparent objective, target viewer, promised value, context, prerequisites.
- Main claim, emotional proposition, desired viewer action.
- Clarity versus ambiguity; consistency between title/thumbnail/page and video.

## Opening and retention mechanics

- First visible and spoken information.
- Curiosity gap, stakes, specificity, novelty, proof, identity, tension.
- Time to value, setup burden, open loops, pattern changes, escalation, payoff.
- Never claim actual retention without analytics; call these `retention mechanics`.

## Visual track

- Shot type, composition, subject hierarchy, motion, lighting, color, continuity.
- B-roll relevance, demonstrations, graphics, typography, subtitles, OCR text.
- Branding, platform-safe zones, mobile readability, visual repetition.
- AI-media consistency, temporal artifacts, lip synchronization, impossible motion.

## Audio track

- Speech intelligibility, delivery, pace, pauses, emphasis, emotion, naturalness.
- Music role, level, transitions, sound effects, silence, clipping, noise, balance.
- Alignment between words, picture, text, effects, and musical changes.

## Structure and editing

- Premise, setup, development, examples/proof, turn, payoff, CTA.
- Average perceived shot length, transition purpose, dead time, cognitive load.
- Whether edits improve comprehension or merely create activity.

## Persuasion and credibility

- Evidence, demonstration, specificity, authority signals, social proof.
- Claims, omissions, exaggeration, manipulation, sponsorship disclosure.
- CTA clarity, timing, effort, trust requirements, next-step continuity.

## Accessibility and inclusion

- Caption presence and accuracy, contrast, text size, audio dependence.
- Language complexity, flashing/motion risks, descriptive sufficiency.

## Production reverse-engineering

- Distinguish observed production evidence from hypotheses.
- Offer multiple plausible workflows when tool attribution is uncertain.
- Describe transferable principles; do not facilitate copying protected creative expression.

## Improvement prioritization

For every recommendation state: timestamp/problem, proposed change, intended effect, evidence strength, effort (`low/medium/high`), and expected impact (`low/medium/high`). Do not present impact estimates as guaranteed performance.