About the data
What finance-YouTube creators said about a hand-curated set of companies, over a 90-day window. Every kept row survived a grounding audit — channel, video, quote. Nothing is published yet: a human review step gates that.
About this universe
- 110tracked
- 84with evidence
- 98discovered
rule72 tracks a hand-curated universe of 110 companies — the companies it understands today, built from attributed creator evidence. This is not market-wide coverage. 84 of them carry grounded creator evidence in the current dataset.
Beyond it, 98 more companies surface in creator transcripts but aren’t tracked yet — what rule72 may understand next. Discovery is not coverage: a candidate becomes tracked only after an operator-gated review. See discovered candidates →
Provenance
snapshot 2026-06-25
- transcripts fetched
- 3947
- grounded evidence (kept)
- 912
- companies with sentiment
- 104
- published
- 0 — everything is review-pending
- synthesis model
- claude-sonnet-4-6
- generated
- Sep 1, 2026
- data source
- read-only rule72 API
Operator quality ledger (audit verdicts & quarantine) → · Consensus Engine status →
Known limitations
As of the 2026-06-25snapshot: YouTube-only sourcing with format and recency skew; a hand-curated probe universe of companies (not the market) over a fixed trailing window; residual guest/journalist attribution errors in interview formats; stance calibration on indirect macro takes is the largest remaining defect class; a handful of videos have no captions and are excluded. Every quote is the creator’s own words, shown verbatim — rule72 never paraphrases or rewrites them; timestamps are partial, and where an export carries no offset the receipt links to the video without a moment, never an invented one. This dataset describes creator speech — it is not a market model.