# Methodology & Data (agent-readable)

Live page: https://videostance.com/methodology · This doc mirrors it for agents. Index: https://videostance.com/llms.txt

## Data pipeline
1. **Source selection** — hands-on creator videos comparing/using the products in a topic (YouTube). Channel credibility, view count, and recency are considered; every video is linked and listed on the topic page.
2. **Claim extraction** — each product claim is pulled with the author and timestamp range from the video, so any statement can be traced back to who said it and when.
3. **Cross-analysis** — claims are grouped per subtopic. Repeatedly supported claims become **CONSENSUS** (confidence: high/medium); claims where creators split become **CONTESTED**, with each side attributed to a named creator.
4. **Publication** — one topic page (and one `.md` twin under https://videostance.com/llms/topics/) per question, with quick verdict, comparison table, findings, FAQ, and full source-video list.

## Freshness
- Each topic records an analysis date (`generatedAt`) and the publish dates of source videos.
- Pages are re-analyzed when materially new creator coverage appears; check the date before quoting fast-moving facts (prices, model versions).

## Limits (be honest with users)
- Analysis reflects what reviewed creators demonstrated in public videos — not lab benchmarks run by VideoStance.
- Sample size is the number of videos on the topic (typically 4–10); low-confidence consensus means few creators covered it.
- Pricing/feature facts decay quickly; treat them as of the analysis date.

## Independence
No pay-for-placement. Affiliate relationships, where they exist, are disclosed at https://videostance.com/disclosure and never affect the consensus scoring.

Docs chain: https://videostance.com/llms.txt ↔ https://videostance.com/llms/about.md ↔ topic docs at https://videostance.com/llms/topics/