Install from GitHub; ResMill is rule-based; current self-check result.

README.md

@@ -12,12 +12,12 @@ tags:
 
 The reference data that [ResBench](https://github.com/SciLM-ai/ResBench) scores
 generative reservoir models against. Every volume was produced by the ResMill
-process simulator (commit in `ENGINE.txt`) under the same rules as the
+rule-based simulator (commit in `ENGINE.txt`) under the same rules as the
 [SiliciclasticReservoirs](https://huggingface.co/datasets/SciLM/SiliciclasticReservoirs)
 training dataset. You do not need to download it by hand:
 
 ```bash
-pip install "resbench[download]"
+pip install "resbench[download] @ git+https://github.com/SciLM-ai/ResBench"
 resbench download        # fetches this repository into the local cache
 resbench score SUBMISSION --out results.json
 ```
@@ -39,4 +39,6 @@ binary sand (1) / mud (0), `int8`.
 
 Everything here can be rebuilt from the ResBench repository's `tools/`
 (command sequence in its SPEC.md) and ResMill at the commit in `ENGINE.txt`.
-Scored against itself, this reference matches in all 31 (task, check) cells.
+Scored against itself (`tools/reference_self_check.py`), the repeats-based checks land at 1.00
+on average and the others at about 0.7, since the self-check reuses the reference volumes
+(details in the repository's STATUS.md).