Every Number Has A Grain
Know The Grain Before You Trust The Number.
Illuminate The True Grain Of Every Metric — because a correct query can still produce the wrong metric.
Drag Event → User → Workspace → Customer → Market. Only Customer is the true grain of Active Customers.
Every product usage ping — logins, clicks, API calls. One person can fire forty events a day.
Distinct user_id — collapses events, but still counts every teammate plus test and duplicate accounts.
Distinct workspace / tenant id — collapses users, but dead trials and abandoned workspaces still count.
The billing entity in the contract table. This is the true grain of “Active Customers.”
Customer × market/region — re-inflated when one multi-region account joins to more than one row.
Four ways the same warehouse produced two, three, or four numbers, all from technically correct SQL.
orders ⨝ contacts ⨝ subscriptions — one customer counted many timesARPC rose overnight when orders were joined to contacts and subscriptions without a grain key. GrainLux names the true grain of each table and explains why the ratio ballooned.
COUNT(DISTINCT user_id).SELECT COUNT(DISTINCT user_id) AS active_customersTwo different grains sharing one metric name — Growth and Finance were never talking about the same thing.
WHERE plan = 'enterprise' -- after a LEFT JOIN duplicated invoice rowsPredicate pushed after a row-duplicating join keeps the duplicates and drops the discipline.
ALTER: renamed fct_billing_events → fct_billing_events_v2Correct SQL, correct deploy — the grain underneath moved anyway, and nobody downstream was told.
Not dashboards. Instruments for finding, defining, and protecting the true grain of a metric.
No warehouse connection required. We infer entity, grain, and fanout risk from the definition alone.
GrainLux is the AI Semantic Data Reliability Layer between your warehouse and every model, metric, and dashboard built on top of it — grain-aware, metric-native, explainable, and deterministic.