CONCEPT Cited by 1 source
Metric definition drift¶
Definition¶
Metric definition drift is the anti-pattern where different teams independently maintain their own SQL/logic for the same business metric, resulting in subtly different definitions that produce different numbers for the same question. It emerges naturally as organizations scale: products launch, teams rotate, people leave, and the tribal knowledge of "what Metric ABC actually means" fragments.
Symptoms¶
- Different dashboards show different numbers for the same metric
- Stakeholders lose trust in data ("which number is right?")
- Outdated definitions silently creep into decision-making pipelines
- Reconciliation meetings consume engineering time
- ML models trained on one definition diverge from monitoring dashboards using another
Root causes¶
- No centralized version control for metric definitions
- No shared standard or single authority on what a metric means
- Team turnover erodes institutional knowledge
- Copy-paste SQL proliferates across codebases
Solutions¶
The standard remedy is a headless BI semantic layer or metric-as-code approach — define once, consume everywhere. Lyft's Metric Semantic Layer is a canonical example: a versioned Python package where each Golden Metric's SQL is defined once and distributed to all consumers.
(Source: sources/2026-06-10-lyft-metric-semantic-layer)
Seen in¶
- sources/2026-06-10-lyft-metric-semantic-layer — Lyft's motivation for building MSL was directly combating this drift