Prove high-quality pixels improve ad performance

## Context You support an ads platform where each advertiser has a **pixel**. Pixel quality varies due to missing/invalid signals, which can affect attribution/optimization and therefore business outcomes. You have two data sources: ### 1) Pixel quality / event health signals Aggregated daily (example columns): - `date` (UTC) - `pixel_id` - `valid_event_count` - `invalid_event_count` - `missing_event_count` - optional: `p50_latency_ms`, `dedupe_rate`, `coverage_rate` (share of sessions with any pixel event) ### 2) Ads performance Aggregated daily at pixel level: - `date` (UTC) - `pixel_id` - `ad_account_id` - `spend` (USD) - `revenue` (USD) — e.g., attributed conversion value - optional: `impressions`, `clicks`, `conversions` ## Questions 1) **Define a metric** (or a small set of metrics) that measures “pixel signal health/quality” for a platform. Specify: - primary metric(s) - diagnostic metrics - guardrails - how you’d handle missing data and very-low-volume pixels 2) **How would you prove (causally)** that higher pixel quality improves business outcomes (e.g., revenue, ROAS, conversion rate) rather than merely correlating with them? - Propose an experiment if possible (units, randomization, duration, primary metric). - If an experiment is not possible, propose an observational/causal strategy and discuss confounders (e.g., advertiser sophistication, spend, seasonality, campaign mix, regression-to-the-mean, selection bias). 3) Briefly outline how you would analyze the results and communicate a recommendation to product/leadership (including what could invalidate the conclusion).