# Bench

Runs a synthetic finite-state “puzzle belt” over a *batch* of boxes.

## Run

```bash
python -m pip install -r requirements.txt
. scripts/bench_env.sh
python bench/run_bench.py

# Bench

- `run_bench.py`: pure speed micro-benchmark (synthetic FSM)
- `run_curiosity_demo.py`: demonstrates **non-advancing PEEK** and **k-ary sequences**
  with two puzzle families:
  - **Informative**: `EAT` is valuable *after* `PEEK`, costly otherwise
  - **Uninformative**: `PEEK` yields cost but no benefit

Expect higher peek rates in the informative segments only.

# Bench

- `run_bench.py`: pure speed micro-benchmark (synthetic FSM)
- `run_curiosity_demo.py`: demonstrates **non-advancing PEEK** with **k-ary sequences**,
  logs a CSV of results per segment
- `plot_curiosity.py`: reads CSV and renders summary figures into an output directory

## Typical usage

```bash
python -m pip install -r requirements.txt
. scripts/bench_env.sh
python bench/run_curiosity_demo.py --out results/curiosity_demo.csv
python bench/plot_curiosity.py --in results/curiosity_demo.csv --outdir results/figs