Why this site exists
AI systems preserve context, prompts, embeddings, model files, logs, and generated artifacts. Teams benefit from that memory while it stays current. When maintenance falls behind retention, the same material becomes a source of cost and uncertainty.
We define Cache Debt as a practical AI operations term for that gap.
What the term is for
The term gives engineering, platform, security, and operations teams a shared way to ask concrete questions:
- What does this system retain?
- Who owns the retained state?
- When was it last reviewed or refreshed?
- What depends on it now?
- When should it be removed?
What we are not claiming
CachedDebt.com does not present the term as an official framework or an established industry standard. The idea extends earlier work on hidden technical debt in machine learning systems, credited on the definition page. The aim is narrower: name a recurring maintenance pattern clearly enough to measure it and decide what to do next.
How we check claims
- Every external claim cites a paper or the tool's own documentation, listed on the sources page with the quoted passage.
- A script re-fetches every source and fails when a quote no longer appears on the page.
- Commands come from vendor documentation. Thresholds are this site's judgment and are labeled that way.
- Measurements come from the audit script, which only reads file sizes and dates. Each one states the machine it ran on. One machine is an example, not a benchmark.
Last reviewed: .
Corrections
Found a wrong path, a broken command, or a claim the source does not support? Open an issue on GitHub with the page and the evidence. Confirmed errors get fixed and noted in the commit history.
Start small
Run the audit script on one machine. Score the three largest areas. Fix the highest score first.