Comment says "measured: N rows" but no code ever computes N
Claude Code and agents
Symptom
A code comment read "measured on <date>: N rows." From that one line grew a load claim ("we re-query N rows every cycle"), and a whole work track was planned on top of it, dry-run counting procedure included.
Measured for real: that re-query never existed. Zero queries anywhere in the codebase aggregate that number. The system looks at a different population, and the actual scan count that day was 0. The track was closed for having no target.
Cause
- A comment is a snapshot, not a value. "Measured on <date>: N" means *a screen showed N that day*, not *this code handles N*. Those are entirely different claims
- A comment sits next to the code, so it reads like a property of the code
- A number with a source attached is more dangerous. An unsourced number gets doubted; a number with a date and a source looks like a citation and makes the reader skip verification
Fix
- Before planning a track on a number, find the code that computes that number first. If you can't, it is not data, it is an observation log
- When leaving a measurement in a comment, say what it is a value of (eyeballed on an external dashboard is not our own aggregate)
- Pin load and scale claims in a contract. If a query producing that number appears later, the contract breaks and you find out
Verify
For every number in the claim, answer "which code path computes this value?" If the answer is "a comment," treat it as unsupported and measure again.
★ Before building a track on a number, find the code that produces it.