diff --git a/docs/RESULTS.md b/docs/RESULTS.md
index 79cf78c..30b327d 100644
--- a/docs/RESULTS.md
+++ b/docs/RESULTS.md
@@ -144,11 +144,13 @@ reproducing its documented tier on this replication.
 
 Gate ≤ teacher + 0.5pp: the student misses by two orders of magnitude.
 
-**RETRACTION (2026-08-24):** this verdict is CONFOUNDED — every Arabic
-label generated before the byt5 `decode_joined` fix was mojibake
-(double-encoded targets); both Arabic students trained on corrupted
-labels, and their identical DER scores are the bare-text constant, not
-a capacity result. The numbers stand as measured but the capacity
-conclusion for Arabic is UNPROVEN pending a clean-label re-run. The
-Thai tiny verdict is unaffected (umt5/sentencepiece labels were
-byte-exact); the pretrained-backbone law rests on Thai evidence.
+**RETRACTION (2026-08-24) and RESTORATION (same day):** the original
+verdict was retracted when the labels proved mojibake (the byt5
+decode_joined bug); the clean-label re-run restores it — 33M student,
+11,792 byte-exact r6 labels, train CE 1.55, windowed DER-CE **82.87%**
+vs the teacher's 1.32% on the same 300-paragraph harness. The failure
+mode is total: the trained student emits EOS immediately on free
+running (empty output; it fits the training set under teacher forcing
+but cannot sustain generation). The capacity conclusion for Arabic is
+now UNCONFOUNDED and matches Thai: sub-100M from-scratch byte students
+do not generalize; a pretrained backbone is non-negotiable.
diff --git a/src/gpu/modal_distill.py b/src/gpu/modal_distill.py
index 76a11d4..7899372 100644
--- a/src/gpu/modal_distill.py
+++ b/src/gpu/modal_distill.py
@@ -995,7 +995,20 @@ def __getitem__(self, i):
 
     save_every = 500
     step = 0
-    ckpts = sorted(out_root.glob("step-*"), key=lambda p: int(p.name.split("-")[1]))
+    import hashlib
+
+    labels_digest = hashlib.sha256(
+        teacher_labels_path.read_bytes()
+    ).hexdigest()[:12] if teacher_labels_path.exists() else "none"
+
+    def _usable(ck: Path) -> bool:
+        marker = ck / "labels.sha"
+        return marker.exists() and marker.read_text().strip() == labels_digest
+
+    ckpts = sorted(
+        (c for c in out_root.glob("step-*") if _usable(c)),
+        key=lambda p: int(p.name.split("-")[1]),
+    )
     if ckpts:
         student.load_state_dict(
             torch.load(ckpts[-1] / "student.pt", map_location="cpu", weights_only=True)
@@ -1028,6 +1041,7 @@ def __getitem__(self, i):
             if step % save_every == 0:
                 ck = out_root / f"step-{step}"
                 ck.mkdir(exist_ok=True)
+                (ck / "labels.sha").write_text(labels_digest)
                 torch.save(student.state_dict(), ck / "student.pt")
                 torch.save(optimizer.state_dict(), ck / "optim.pt")
                 CHECKPOINTS.commit()