Technical debt is the gap between a system’s best possible performance and its real performance. Expedient compromises may enable early production delivery, but they accumulate problems that impose extra work on users, connected systems, and maintainers. In ML systems, changing interfaces, data flows, operating conditions, and available algorithms can widen this gap over time. section 10.2.2 section 1.3
Manage technical debt through explicit maintenance and delivery actions:
- Instrument the system with activity logs, performance monitoring, and alerts for the support group. These controls make troubleshooting possible and provide early warning of failures. section 10.2.2
- Benchmark an inherited model against newer techniques. Rapid ML research can leave older approaches brittle, temperamental, or outperformed; identifying replacement opportunities can reduce future maintenance effort. section 10.2.2
- Assign support responsibility, establish governance arrangements, review systems regularly, and keep a plan for failures and technical-debt remediation. section 10.8
An inherited ML system
A team taking over a production model first checks whether logs record system activity, whether performance can be understood, and whether failures trigger support alerts. It then benchmarks the model against newer techniques and records replacement work alongside governance and support responsibilities. section 10.2.2 section 10.8
