> ## Documentation Index
> Fetch the complete documentation index at: https://project-brain.rocheaiops.net/llms.txt
> Use this file to discover all available pages before exploring further.

# Self-improvement

> A future evaluation-and-rollback loop, not autonomous V0.1 behavior.

**Future only:** experience → observe → reflect → hypothesize → modify candidate → test → promote or rollback → repeat.

| Risk level | Future example                         |
| ---------- | -------------------------------------- |
| 1–2        | Observations and memories              |
| 3–4        | Preferences, routines, rules, skills   |
| 5          | Concepts and knowledge structures      |
| 6          | Algorithms and code                    |
| 7–8        | Training material and model adaptation |

Higher-impact changes require stronger evidence, approval, evaluation, monitoring, and rollback. V0.1 has bounded candidate/consolidation behavior, not autonomous production edits, schema changes, training, or LoRA. S.P.O.C.K. may later execute tested engineering candidates but does not own cognition.