# Method

Dream-RSI combines a fixed evaluator and keep-only-if-better hill climb with append-only attempt records, fresh agent processes, independent review, paired confirmation, and correctness gates. Usage signals help rank candidate levers. The loop also qualifies its own metric before trusting it.

Each completed experiment is recorded as a world. A deterministic replay compares search configs using only recorded branch and refinement nodes. For one world:

```text
V = g - 0.02 * N + 0.01 * N / max(1, rounds)
```

`g` is normalized gain over baseline, `N` is the number of revealed attempts, and `rounds` is the number of search rounds. Replay scores are averaged across completed worlds. A config is eligible only when its replay coverage is at least 0.6 and its mean value beats the incumbent by more than 0.01. The incumbent is always a candidate. Agent output is config data and is never executed.

The improvement path stays fixed around that search-policy layer: qualify the metric and baseline first, keep candidate experiments bounded, promote only after paired confirmation, correctness gates, and adversarial review, then verify the applied patch independently. The dream step can change how later attempts are allocated; it cannot change the evaluator, safety gates, or acceptance rules.

Inspirations: autoresearch contributes the fixed metric and hill climb; RSI-Master contributes append-only records and adversarial review; Memento contributes fresh processes and inherited notes; MASS contributes a separate judge; usage-log funnels contribute lever evidence; Dream-RSI contributes world recording and replay.