# Data dictionary

Runs: phase/provider/question/run/model/status/UTC timestamp/raw reference and immutable hashes. Citations: original URL, normalized URL, observed domain, resolved domain if available, answer-linked rank/title and native annotation positions. Domain leagues: occurrence counts/share and response presence with explicit denominators. Stats: 10,000-draw query-cluster intervals, matched overlap/stability, missing-provider and collection-readiness checks. YouTube characteristics: cached metadata, duration/age band, captions flag and retrieval timestamp; missing metadata remains missing. Classification: rule, confidence, review and reason. See CSV column names and method/limitations for interpretation.

`youtube-channels.csv` adds `responses_citing_channel` and `completed_response_denominator`; `unique_videos` deduplicates video IDs within a provider/channel. `kingy-youtube-citations.csv` reports unique videos, native citation occurrences and distinct answers for the confirmed Kingy channel, with all 2,400 completed answers per provider as denominator. `fetched_ts` dates every metadata snapshot.

## Top-100 human category review

Curtis reviewed the selected top 100 of 4,631 collection domains; the other 4,531 remain provisional. Category comparisons describe this selected subset, with all unreviewed citations pooled separately. Labels describe source type, not accuracy or quality.

The reviewed subset accounts for OpenAI 5,998 / 11,738, Claude 4,294 / 18,327 and Gemini 7,511 / 15,734 native citation occurrences. Selection used pooled citation frequency, not random sampling. The same domain-level conventions apply across providers. Whole-company labels include company blogs; shared hosting can contain many authors; academic does not guarantee peer review. Review approval does not validate factual support.

`domain-classifications.csv` records source labels, methods, review flags and reasons for all 4,631 observed domains. `domain-categories.csv` groups only human-reviewed labels and an explicit unreviewed pool; `share_of_reviewed` is blank for that pool. `human-category-decisions.csv` preserves top-100 approval provenance and the frozen review snapshot hash. `category-review-summary.json` records selection and coverage.