diff --git a/training/amygdala_training/train_steering_vectors.py b/training/amygdala_training/train_steering_vectors.py
index 54603ab..353ebb0 100644
--- a/training/amygdala_training/train_steering_vectors.py
+++ b/training/amygdala_training/train_steering_vectors.py
@@ -850,12 +850,13 @@ def main() -> None:
     ap.add_argument(
         "--subspace-k",
         type=int,
-        default=512,
+        default=99999,
         help="Max top-k right singular vectors per story for subspace method "
-             "(clamped to n_tokens per story). Default 512 is enough to span "
-             "each story's full natural subspace including per-attention-head "
-             "contributions on a hidden_dim=5120 residual stream. Smaller "
-             "values (e.g. 20) discard per-head discriminability.",
+             "(clamped to min(n_tokens, hidden_dim) per story). Default is "
+             "effectively 'keep full per-story subspace' — each story's V_i "
+             "spans its entire natural row space. On a hidden_dim=5120 "
+             "residual and ~500-token stories, that's ~500 vectors per story. "
+             "Memory is fine: 112 × 5120 × 500 × 4 bytes ≈ 1.1 GB.",
     )
     ap.add_argument(
         "--quality-report",