Grouped citation intelligence

Grouped citation intelligence by url / domain / host with per-model breakdown, citation rate, and avg citation position. Counts and citation rate include visible citations and background source references. Average position ignores rows with position=0. Owned and competitor source matching honor the project's exact-subdomain setting. Filter vocabulary aligns with source_type returned by the API. Unavailable page content keeps the cited URL and citation metrics. Page metadata/content and unknown brand_mentioned/competitor_mentioned return null. content_gap_status is content_unavailable (or missing_page_cache) without usable content, and mentions_not_processed when the page has usable content but its mention analysis has not completed for this project. Mention fields (brand_mentioned, competitor_mentioned, brand_position, brands, mentions, content_gap_status) describe the analysis for the requesting project: a page not yet analyzed for it returns null (never false) brand_mentioned, competitor_mentioned and brand_position, empty brands and mentions arrays and page_cache.mentions_processed false. page_cache.mentions_processed is true exactly when brand_mentioned is not null. An analyzed page keeps its last result until a newer analysis replaces it. status_code shows a saved successful response or observed 404/410; other crawl failures and error_message are hidden. last_crawled_at dates the saved copy. Domain/host crawled_urls_count counts usable copies; brand_mentioned_urls_count and competitor_mentioned_urls_count count only URLs analyzed for the project and are null when none is. Pass key to read one group; brand_kind filters by the brand focus of the prompts, like the source page. citation_series (with include_series=true) holds the answers citing the URL, domain or host under the same filters: granularity (day, week or month, from the account tracking cadence, with weeks when a daily account spans more than 60 days, starting on Monday), total (equal to total_responses), by_model (answers per AI model, largest first, which can include enabled models outside the seven the model filter accepts) and data, one { date, value } point per period that holds a filtered answer, so a zero means answers ran and none cited it. Periods date each answer by its executed_at in Europe/Madrid time, the column from and to filter.