# Novel Feedback v5 Qwen3-4B LLM Simulator

## Simulator Eval Protocol

The v5 simulator is a deterministic CPU/proxy-only stand-in for
Qwen3-4B-Instruct-2507 with a LoRA adapter. It does not load a model, download
weights, call a network service, or use a GPU. Public reporting is generated
from 960 synthetic examples shaped like the Qidian-WebNovel DTA
feedback loop.

## Contract

- base model: Qwen/Qwen3-4B-Instruct-2507
- adapter: LoRA
- dataset: Qidian-WebNovel DTA
- style: Claude Blog style
- uses_reranking=false

## Dynamic Metrics

- Cumulative Interest
- Preference Drift Alignment
- Reading Fluency Penalty
- Positive Expectation Violation Rate

The preferred simulator should keep strict JSON valid, complete the feedback
schema, separate same-text users, and emit scoped state updates with both
`scope` and `anti_scope`.