The gradual decline in model output accuracy, reasoning quality, or performance over time caused by changing real-world data, unvetted training updates, or server context clutter.

A serious technical term used in executive retrospectives to describe an app's answers getting worse. It drapes software performance drops in scientific decay language, making bugs sound like inevitable physical degradation.

• Scientific Decay Language: Borrowing physical nuclear or biological decay terms to explain why software answers got worse.
• Maintenance Excuse: Using "model degradation" to explain why a team failed to audit prompt templates or update databases.
• House-Cleaning Reality: Like a vacuum losing suction because the bag is full and calling it "vacuum degradation."

Flowers for Algernon / Sci-Fi Drama classic scene where scientists observe declining maze test performance, tracking degradation in cognitive skills.

The drop in software answer quality or accuracy that happens over time as data changes.

“Our automated testing suite detects model degradation before it impacts client workflows.”

“We set up monthly accuracy checks to catch drops in answer quality caused by outdated policy documents.”