fix(tools): avsync says "unmeasurable" instead of trusting a 0.08 correlation

On ty-20261112-1401-952-0000-2.mp4 the per-window estimates ranged from +400 to -370 ms and
only one window cleared 0.35, yet the tool printed "-370 ms, adjust by +370" from an overall
correlation of 0.08. A number that wrong is worse than no number: it invites another build.

Now: windows above 0.35 must also agree within 150 ms, and anything weaker reports
"unmeasurable" with the remedy — three claps with the hands in the webcam frame, two seconds
apart, which gives a sharp visual and audio transient instead of a mouth-motion proxy.

Also reverted an onset-based envelope tried on the same clip: it dropped the best window's
correlation from 0.47 to 0.08, so the plain level envelope stays.

What the run did establish: lengths match (15.555 s video against 15.601 s audio, 0.3%), and
the log confirms "mic delay 350 ms" was applied — the drift fix and the delay wiring both hold.

scripts/avsync.py

@@ -129,10 +129,16 @@ def offset_ms(audio: np.ndarray, motion: np.ndarray, fps: float, max_ms: float =
         lag, corr = best_for(m[start:start + seg], a[start:start + seg])
         windows.append((start / grid, lag, corr))
     strong = sorted(lag for _, lag, corr in windows if corr >= 0.35)
-    if strong:
+    if len(strong) >= 2:
         mid = len(strong) // 2
         robust = strong[mid] if len(strong) % 2 else (strong[mid - 1] + strong[mid]) / 2
+        spread = strong[-1] - strong[0]
+        if spread > 150:
+            # Windows that disagree this much cannot be averaged into an answer. Say so.
+            return float("nan"), max(corr for _, _, corr in windows), windows
         return robust, max(corr for _, _, corr in windows), windows
+    if strong:
+        return float("nan"), max(corr for _, _, corr in windows), windows
     return overall_lag, overall_corr, windows
 
 
@@ -168,13 +174,18 @@ def main() -> int:
     print(f"delta   {diff:+8.3f} s  ({pct:+.2f}% of the video length)")
     verdict = "lengths match" if abs(diff) < 0.05 else "LENGTH MISMATCH — a clock problem"
     print(f"        {verdict}")
-    quality = "strong" if corr >= 0.35 else "weak — treat the offset as a hint only"
     for start, wlag, wcorr in windows:
         mark = "*" if wcorr >= 0.35 else " "
         print(f"        {start:5.1f}s {wlag:+7.0f} ms  corr {wcorr:.2f} {mark}")
-    print(f"offset  {lag:+8.0f} ms  (correlation {corr:.2f}, {quality})")
-    direction = "sound LATER than the lips" if lag > 0 else "sound EARLIER than the lips"
+
+    if lag != lag or corr < 0.35:
+        print("offset  unmeasurable — too little agreement between mouth motion and speech")
+        print("        record 3 claps with your hands in the webcam frame, 2 s apart")
+        return 0
+
+    print(f"offset  {lag:+8.0f} ms  (correlation {corr:.2f})")
     if abs(lag) >= 20:
+        direction = "sound LATER than the lips" if lag > 0 else "sound EARLIER than the lips"
         print(f"        {direction} — adjust the mic delay by {-lag:+.0f} ms")
     else:
         print("        within a frame or two — leave it alone")