fix(idf_waveforms): _find_waveform_body_offset() — scans every 00 02 00 magic past offset 0x0E00, runs decode_waveform_v2 on each candidate, picks the one that returns the most samples. Validated on 483 prod IDFW files: 0 preamble-only events (was ~50%), 355/483 fully decode, 126/483 partial (BW codec walker-stops-early on loud events — known issue).

IDFH now synthesises a 1-sample-per-interval array from the binary intervals and writes an .h5 so the existing renderer works unchanged. Each "sample" is the per-interval peak ADC count → h5_value = count × geo_fs/32768 yields the right bar height.

sfm/waveform_store.py

@@ -600,10 +600,28 @@ class WaveformStore:
             n_samples = max((len(idf_samples.get(ch, [])) for ch in ("Tran", "Vert", "Long", "MicL")), default=0)
             ev.total_samples = ev.total_samples or n_samples
 
-        # 7. Write the .h5 clean-waveform file when we actually have samples.
-        # Histograms (IDFH) don't have waveform samples — skip h5 for those.
+        # For IDFH histograms there are no per-sample waveform arrays — the
+        # device stores one peak ADC count per interval per channel.  Synthesise
+        # a 1-sample-per-interval array so the existing h5+renderer pipeline
+        # (which groups samples down to ``n_intervals`` bars via max-per-group)
+        # produces a non-blank histogram chart.  Each "sample" is the peak ADC
+        # count for that interval, so the h5 writer's ``count × geo_fs/32768``
+        # conversion yields the right physical value for the bar height.
+        if is_histogram and idf_intervals:
+            hist_samples = {
+                "Tran": [iv.peak_count("Tran") for iv in idf_intervals],
+                "Vert": [iv.peak_count("Vert") for iv in idf_intervals],
+                "Long": [iv.peak_count("Long") for iv in idf_intervals],
+                "MicL": [iv.peak_count("MicL") for iv in idf_intervals],
+            }
+            ev.raw_samples = hist_samples
+            ev.total_samples = ev.total_samples or len(idf_intervals)
+
+        # 7. Write the .h5 clean-waveform file when we have samples to write
+        # (either the IDFW per-sample stream, or the IDFH synthesised per-
+        # interval peak array).  The renderer treats both shapes the same way.
         hdf5_filename: Optional[str] = None
-        if idf_samples is not None and not is_histogram:
+        if ev.raw_samples:
             hdf5_path = self.hdf5_path_for(serial, filename)
             try:
                 event_hdf5.write_event_hdf5(