data-explorer

Data Exploration

Every read takes a projectId and returns what that project's scope covers. Two levers decide whether you get a useful answer or a wall of noise: picking the right reader, and controlling which fields come back.

Discover before you filter

Label dimensions vary by project — they depend on what has been analyzed. Call list_labels({ projectId }) to see which dimensions exist and their top values before naming one in a filter; a dimension the data doesn't carry silently matches nothing. get_table_data({ projectId, listDimensions: true }) answers the same question from the pivot side. aggregate({ projectId, groupBy: ["brand"], measures: [{ field: "*", fn: "count", as: "n" }] }) tells you the exact entity names the scope tracks, which is what feedNames and brand filters match against.

Choosing a reader

analyze is the primary content tool. Four distributions:

• "balanced" (default) — equal representation per feed, a mix of top and recent. The honest "what does this scope look like" read.
• "top" — highest engagement per feed; set sortMetric to change which metric.
• "recent" — newest per feed.
• "exhaustive" — no sampling. A deterministic ranked page over the full filtered set with totalEstimated and nextOffset, so you can page the whole thing. Pair with sortBy, including the lift keys (likesMultiple, viewsMultiple, …). This is the leaderboard and "give me all of X" mode.

The sampled distributions read the creative feeds — brand and influencer — unless you name feedTypes yourself, so raw search and discussion items don't crowd a "what hooks are working" sample. Exhaustive reads exactly what you filtered, nothing added.

Two other modes: mode: "semantic" finds posts by meaning from a natural-language query (it honors platform, feed type, dates, and includeComments; narrow further after the results come back), and passing itemIds runs a by-id deep dive that returns full creative, labels, and performance for specific posts.