Trust the Calculator

Short summary

An expert reflection on encoding domain knowledge into software: trust the artifacts money flows through (calculators, invoices, worked claims) over documentation (manuals, guidance prose). The author built a water-feature pricing engine calibrated against 132 real quotes (within 5% of actual charges) and a medical billing denial engine whose action accuracy jumped from 36.7% to 76.7% when anchored to real worked claims instead of guidance text. The core insight: every domain keeps two records—what it says about itself and what it paid to learn—and you should encode the latter.

• •Domain rules should be extracted from money-bearing artifacts (calculators, invoices, worked claims), not documentation
• •Pricing engine calibrated against 132 real quotes landed within 5% of actual shop charges
• •Medical billing denial engine accuracy improved from 36.7% to 76.7% by anchoring to worked claims instead of guidance prose

Generated with AI, which can make mistakes.