diff --git a/scripts/make_random_vhack.py b/scripts/make_random_vhack.py
index c4fc636..f74f8b2 100644
--- a/scripts/make_random_vhack.py
+++ b/scripts/make_random_vhack.py
@@ -19,9 +19,9 @@ from loguru import logger
 from safetensors import safe_open
 from safetensors.torch import save_file
 
-SRC = Path("out/vhack/v_hack_pairset_prog_wide.safetensors")  # run 31's v_hack
+SRC = Path("out/vhack/v_hack_pairset_prog_wide.safetensors")  # shape source: match its per-module rank/norm
 DST = Path("out/vhack/v_hack_pairset_prog_wide_randomV.safetensors")
-SEED = 157  # the task number; fixed so the control is reproducible
+SEED = 157  # fixed so the random-V control is reproducible
 
 
 def haar_orthonormal_rows(k: int, r: int, generator: torch.Generator) -> torch.Tensor:
diff --git a/scripts/probe_distill.py b/scripts/probe_distill.py
index f0e8e21..1339261 100644
--- a/scripts/probe_distill.py
+++ b/scripts/probe_distill.py
@@ -81,11 +81,9 @@ class Config:
     # samples. Used to populate the "no_hack" bucket for cosine comparison.
     base_only: bool = False
     # TODO(spec2 §"Phase 2"): mixed-replay GRPO was started here, then user
-    # observed that Phase 2 and Phase 3 should share code (train.py) with
-    # different --steps args, not build separate replay machinery. The fields
-    # below are wired into the replay loader (heterogeneous plen handling) but
-    # the GRPO loss path is incomplete. Either finish or remove; for now train.py
-    # at small scale is the canonical Phase 2 mechanism.
+    # FIXME: the replay fields below are wired into the loader (heterogeneous
+    # plen handling) but the GRPO loss path is incomplete -- finish or remove.
+    # train.py at small scale is the canonical Phase 2 mechanism.
     replay_dirs: str | None = None
     # Sandwich schedule: [0, pre) student-gen -> [pre, pre+replay) replay-distill
     # -> [pre+replay, steps) student-gen. With pre_warmup_steps=0 reduces to the