Card: describe each directory as a trained probe

README.md

@@ -17,15 +17,15 @@ layers and returns one score per transcript. Load them with the `probe-inference
 ```python
 from probe_inference import load_probe_from_hub
 
-bank = load_probe_from_hub("qwen3.5-9b/efc")  # this repository at the package's pinned revision
-score = bank.score(acts, bank.read_mask(prompt_mask, completion_mask, followup_start_position))
+probe = load_probe_from_hub("qwen3.5-9b/efc")  # this repository at the package's pinned revision
+score = probe.score(acts, probe.read_mask(prompt_mask, completion_mask, followup_start_position))
 ```
 
 ## Layout
 
-`<model>/<arch>/`, with `<arch>` in `linear`, `mlp`, `efc`, `axial`. A linear or MLP cell holds one probe per
-layer (`layer_<L>/config.json`, `layer_<L>/model.pt`); an EFC or axial cell holds one cross-layer probe
-(`config.json`, `model.pt`). Every cell has `probe_metadata.json`: the model and revision, the layers, the
+`<model>/<arch>/`, with `<arch>` in `linear`, `mlp`, `efc`, `axial`. Each directory is one trained probe. A linear or MLP probe is one
+small probe per layer (`layer_<L>/config.json`, `layer_<L>/model.pt`); an EFC or axial probe is one module
+that reads all its layers at once (`config.json`, `model.pt`). Every probe has `probe_metadata.json`: the model and revision, the layers, the
 read window (`obfuscate_over`), the token aggregation and, for linear and MLP, the layers whose sigmoids
 are averaged (`layer_rule.used_layers`). `model.pt` files are plain float32 state dicts.