Four steps sit between a link you paste and a report you can read. The first two run before you pay anything.
1. Discovery#
You paste a video, a channel, or a playlist. We resolve it to a concrete list of videos with their durations, using the official YouTube Data API. Results are cached, so a channel someone else already looked up costs nothing to look up again.
Nothing is fetched or generated at this stage — this step only establishes what could be built.
2. Quote#
The video list determines the price. Duration and count go in, a number comes out, and that number depends only on the listing — never on what happens to be cached. A short single video comes back free; anything larger comes back with a price and a Build Refinery button.
3. Extraction#
Transcripts are fetched with timestamps, and claims are pulled out of them video by video. Per-video work is reusable forever: two people researching the same channel never pay to read it twice.
4. Verification#
Extracted claims are cross-checked by a different model family than the one that produced them. Anything unsupported is excluded from the body of the report and moved to an appendix.
This step is not a premium feature. It runs on every report, including free ones, because a report that states unverified claims about real companies and people is not something worth shipping. What changes between tiers is the verifier, not whether verification happens.
What comes back#
A structured corpus: entities, the relations between them, contradictions where sources disagree, and the evidence behind each claim — rendered as a browsable wiki or a PDF.