Accuracy Aggregates

GET /api/v1/accuracy-aggregates returns Brand FactCheck's window-level rollups: one row per measurement window and engine, summarizing the verdicts of that window's claims into an Accuracy Score.

Filters

Standard parameters (limit, offset, format=csv) work as described in Pagination, filtering, and errors.

Key fields

• measurement_window_id: joinable to measurement windows
• engine_id: the engine the row covers
• accurate_count, inaccurate_count, unverifiable_count: trusted verdicts in the window
• needs_review_count: low-confidence verdicts waiting for review; these are excluded from the score until confirmed
• claim_count: non-suppressed claims for the window and engine. Suppressed claims are excluded here, matching the score, so this count can be lower than the number of rows GET /api/v1/claims returns for the same window
• accuracy_score: accurate as a share of accurate plus inaccurate, on a 0 to 100 scale; null when the window has no scoreable verdicts
• updated_at: when the rollup was last recomputed

Rows are ordered by measurement_window_id descending with engine_id as the tiebreak, so offset pagination is stable. Deltas between windows are computed by comparing rows across measurement_window_id values.

Example

curl -H "Authorization: Bearer $AEO_API_KEY" \
  "https://app.citlyze.com/api/v1/accuracy-aggregates?engine_id=openai_web"

Use this resource to trend factual accuracy per engine window over window, and to feed Accuracy Score into BI dashboards next to visibility and sentiment.