BabyYarros: pre-register the pair-vs-rawtext decision rule before the arms are read

scripts/r49-corpus/gen_beats_chat_yarros.py

@@ -39,6 +39,11 @@ ap.add_argument("--seeds", type=int, nargs="+", default=[1234, 5678])
 ap.add_argument("--max-new-tokens", type=int, default=320)
 ap.add_argument("--temperature", type=float, default=0.9)
 ap.add_argument("--top-p", type=float, default=0.95)
+# The pair-SFT arm was trained with the user turn rendered as "Beat: <x>" while this
+# harness has always sent "BEAT: <x>". Default is unchanged so every previously recorded
+# arm stays byte-reproducible; the flag exists so the casing can be MEASURED as its own
+# variable instead of being confounded with the adapter it is being used to judge.
+ap.add_argument("--user-prefix", default="BEAT: ")
 a = ap.parse_args()
 
 beats = json.loads(Path(a.beats).read_text())
@@ -71,7 +76,7 @@ with out.open("w", encoding="utf-8") as fh:
             torch.manual_seed(seed)
             text = tok.apply_chat_template(
                 [{"role": "system", "content": SYS},
-                 {"role": "user", "content": "BEAT: " + b["beat"]}],
+                 {"role": "user", "content": a.user_prefix + b["beat"]}],
                 tokenize=False, add_generation_prompt=True,
                 **({"enable_thinking": False} if "enable_thinking" in (tok.chat_template or "") else {}))
             ids = tok(text, return_tensors="pt", add_special_tokens=False).to("cuda")
@@ -87,7 +92,7 @@ with out.open("w", encoding="utf-8") as fh:
             hit = sum(1 for k in kws if k[:5] in para.lower())
             w = len(para.split())
             fh.write(json.dumps({"format": a.arm, "id": b["id"], "beat": b["beat"], "seed": seed,
-                                 "prompt": "BEAT: " + b["beat"], "paragraph": para,
+                                 "prompt": a.user_prefix + b["beat"], "paragraph": para,
                                  "ran_on": m is None, "words": w,
                                  "in_band": 90 <= w <= 140,
                                  "beat_keywords": kws, "keyword_hits": hit}) + "\n")