Dark Debt

Definitions

1. Hidden technical debt in machine learning systems that's harder to detect than traditional code debt. Includes training-serving skew, undeclared data dependencies, feedback loops, dead experimental code paths, and entangled model features where changing one affects others unpredictably. From Google's influential 'Hidden Technical Debt in ML Systems' paper.

   In plain English: Hidden problems in AI/ML systems that are much harder to see and fix than normal code issues.

   Example: "The model silently depends on a feature that's derived from another model's output — that's dark debt. If the upstream model changes, ours breaks in subtle ways."

Related Terms

• Webhook vs. Polling
• Event-Driven Architecture
• blast radius
• ADR
• Client
• Decentralization
• Distributed Systems
• Future Proof