One number. Many invisible choices.
Every dashboard metric depends on decisions nobody documented.
$3.1M. Monthly active revenue. Six undocumented decisions behind that number: which accounts are excluded, which revenue table was used, which join path, which segment definition, which time boundary, which currency conversion.
Each of those decisions was made by a person at a specific moment, for a specific reason that no longer has a home in the system. The dbt macro that excludes internal and trial accounts exists; the comment explaining why Finance made the gross-to-net call after a board discrepancy never got written.
Dashboard consumers see the output and assume they understand the input. The people who built the pipeline know what it encodes. Everyone in between is guessing, including the AI agents now tasked with writing queries against the same warehouse.
The problem compounds when definitions change. Segment definitions shift. The dashboard doesn't update its assumptions. Six months later, someone builds a model against a number that now means something different than when the metric was defined.
Surfacing these decisions is a context infrastructure problem. The choices need to travel with the data, attached to the transformations they shaped, rather than living in the head of whoever last touched the pipeline.
