Definition
The engineering posture for the point where optimization runs out. Optimizing a metric or reward function invites Goodhart drift, leaves out what the metric cannot see, and produces residual complexity.
Post-optimization rigor does not try to fix model incompleteness by adding more parameters or objective terms. It builds the institution around the model instead: independent review, measurement of the uncounted human subsidy, corrective standing for affected parties, and revocable authority grants.
Tests that show it is in place
- The deployment pairs every optimized model with independent infrastructure for correction and for challenges.
- Performance audits track externalized friction and absorption rather than relying solely on internal loss functions or proxy metrics.
References
- Core axioms · Doctrine
- Laws for engineering delegated intelligence · Doctrine