Methodology · version 0.1
Evidence before confidence.
NarraScout is built around an inspectable chain: source asset → source span → observation → inference → recommendation → rewrite → version.
This is a creative-evaluation method, not a guarantee of audience response. A polished report can still contain a judgment you disagree with; the evidence lets you challenge it.
1. Start with the format
Feature screenplays, vertical drama and short video scripts are evaluated with different emphases. We do not force a short-drama checklist onto a feature film or treat a 30-second ad like a miniature three-act movie.
The input format determines the report dimensions. Revision order remains consistent: structural causality first, then character and emotion, then pacing, dialogue and presentation.
2. Separate fact from interpretation
| Label | What it means | What it cannot mean |
|---|---|---|
| Source text | A line or action directly present in the supplied script. | A claim about audience behavior. |
| Heard | Speech or sound supported by the supplied transcript or audio analysis. | Proof of what is visible. |
| Seen | A visual observation supported by an available frame. | Exhaustive coverage of every frame or cut. |
| On-screen text | Text actually detected in a visual source. | Text inferred from the narration. |
| Inference | A creative interpretation based on identified evidence. | A newly invented source fact. |
| Unavailable | A required input was not supplied or could not be verified. | Permission to fill the gap with a confident guess. |
3. Make every priority actionable
A priority finding includes an observation, an interpretation, an explanation of its creative effect and a concrete revision action. Severity labels communicate revision order:
- Blocker: prevents the story or message from working in its current form.
- Major: materially weakens motivation, structure, clarity or proof.
- Minor: creates a localized craft problem.
- Preference: an optional creative test, not an objective error.
4. Preserve strengths during revision
A rewrite starts from a selected finding and explicit protected elements. Long scripts are revised by scene or range rather than regenerated wholesale. Versions remain attached to the original project so a new draft can be compared and re-evaluated.
5. State the technical limits
Short-video analysis uses Narrascout's video engine on a complete 640p analysis video at the configured FPS, plus separately sampled review frames. It is not exhaustive frame-by-frame reconstruction; short shots, on-screen details and audio cues may be missed. Without the video engine, a transcription provider or user transcript is required for the explicitly labeled local heuristic fallback.
6. Keep performance data out of creative claims
Retention, CTR, conversions and ROAS depend on audience, placement, offer, spend and many factors outside the content file. Narrascout does not infer them from a script or video. A later analytics module must use real supplied campaign data before making performance statements.
7. Content, privacy and examples
Users must upload work they own or are authorized to analyze. Public examples are original fictional material unless explicit authorization is documented. Private projects are noindex. Local development stores data only in the project data directory; production retention and subprocessors must be configured before launch.