Definitions drift because the business changes and the reporting layer pretends nothing happened.
A new pricing model ships. Customer success changes the health score. Product adds a workspace type. Finance changes revenue recognition. Sales reorganizes territories. Each change may be reasonable. Together they make last quarter's metric definition less true than people think.
Semantic drift is meaning aging in place, not a simple data bug. The dangerous part is that dashboards often keep running. The number still refreshes. The chart still looks official. Nobody sees the definition quietly losing contact with how the business now works.
A semantic layer needs reconciliation rituals alongside definitions. When a definition changes, the company should know what changed, why it changed, who approved it, whether history was restated, and which reports or workflows are affected. If two definitions must coexist, the distinction should be named instead of hidden in filters.
This matters most around recurring reviews. Forecast, retention, pipeline, usage, and margin metrics need comparability. If a metric break is real, say so. Do not make operators infer it from a weird trend line and three Slack threads.
The practical pattern is a semantic change log. Keep it close to the definition. Include the business reason, effective date, downstream surfaces, and owner. Add a reconciliation note for the next review cycle.
Drift is unavoidable. Silent drift is optional. The semantic layer earns trust by making change visible before it becomes an argument.
This is part 6 of 10 in The Semantic Layer.
