Compounding formulas and configuration management

In the compounding lab, we follow worksheets with the precision programmers reserve for config files — except a wrong value doesn't fail a build, it fails a patient. The discipline transfers to software configuration with surprising force.

What compounding worksheets encode

Every variable declared, nothing implicit. Active ingredient, quantity, solvent, procedure, equipment, expiration — stated explicitly. Compare to a config system where half the behaviour lives in environment differences nobody documented. The lab assumes the next person knows nothing; good config assumes the same.

The check is a separate step, by a separate person. Compounded preparations get verified against the worksheet before dispensing — verification is structurally separated from production. Software's equivalent — review, staging, canary deploys — exists for the same reason: the maker is the worst verifier of their own work.

Precision has tiers. Some measurements matter to the milligram; others are "q.s." (as much as needed). Knowing which values are load-bearing is the actual expertise. Config has the same tiers: one wrong cookie secret is fatal, one wrong analytics key is noise. Treating all config as equally critical (or equally casual) is how both labs and startups get hurt.

The habit I kept

I treat production config like a compounding worksheet: declared completely, reviewed separately, changed deliberately, with the blast radius of each variable known. The stakes in software are usually lower than a patient's life. The discipline, it turns out, costs nothing to keep at clinical standard anyway.