Failure modes
Type: kb/types/tag-readme.md
Failure-mode notes name recurring ways an agent-operated KB can fail despite having relevant artifacts on disk. Use this index when debugging why stored knowledge was not discovered, loaded, interpreted, or acted on.
Notes
- Knowledge storage does not imply contextual activation - core distinction between knowledge existing, being loaded, and actually affecting behavior
- Elicitation requires maintained question-generation systems - shows how inquiry processes decay when prompts stop generating discriminating questions
Related Tags
- LLM interpretation errors - adjacent error taxonomy for failures in how models interpret instructions and evidence
- Evaluation - methods for detecting whether failures are real and whether interventions improve behavior
Other tagged notes
- Abstract an experience into a lesson only when you can state where the lesson stops - Abstract an episode into a lesson only when you can state its boundary, else preserve the instance; an over-generalized lesson is one that drops the condition clause
- An LLM's generation confidence tracks typicality, not soundness - An LLM's next-token confidence measures how typical a continuation is, not whether it's true or valid; the two are decoupled, so soundness can't be read off confidence and needs a separate check
- Generality bought to avoid counterexamples is paid for in precision - Widening a claim's vocabulary to survive counterexamples raises universality by spending precision, so content stays flat — and the unreadability that follows is the symptom, not the price of rigor