Lily

IThesis

A lightweight local model may not need frontier scale for every recurring problem. It may need the right evidence from work that has already been completed and verified.

Lily studies whether terminal artifacts can become durable memory that raises local task success over time. The model stays fixed. The memory grows. Tests decide whether capability actually improves.

IIWhy it matters

Useful intelligence should remain close to people, their devices, and the evidence behind its answers.

The experiment treats answers as claims to verify, not prose to admire. Success requires executable evidence, privacy safe artifacts, visible failures, and honest abstention when local knowledge is insufficient.

IVMethod

Run real terminal tasks. Keep only verified and privacy safe artifacts. Test the same local model as its memory grows.

1. 01 Complete Run a terminal benchmark task inside an isolated environment.
2. 02 Verify Accept evidence only after executable tests confirm the outcome.
3. 03 Remember Preserve sanitized evidence and distill it into a human readable Markdown wiki.
4. 04 Measure Evaluate exact recurrence, structural recurrence, novel controls, latency, and unsafe confidence.