fix: waveform decode improved for accuracy.

feat: adds 5a diagnostic script to parse raw binary

CLAUDE.md

@@ -266,6 +266,73 @@ silence). Only the initial variable-size chunks contain actual signal.
 Removed. Terminator detection is via `page_key == 0x0000` in `read_bulk_waveform_stream`,
 not frame index.
 
+### SUB 5A — re-probe at counter=0x1000 (DLE collision, FIXED 2026-04-15)
+
+**Root cause (confirmed from diagnostic output, desk-thump event key=01110000):**
+`bulk_waveform_params()` sets `p[3] = (counter >> 8) & 0xFF`.  For chunk 4, counter =
+`4 * 0x0400 = 0x1000`, so `p[3] = 0x10` — the DLE byte.  Because `build_5a_frame` writes
+params RAW (no DLE stuffing), the on-wire byte sequence contains a bare `0x10`.  The
+device DLE-decodes its own receive buffer: `10 00` (the p[3]/p[4] pair) is collapsed to
+`00`, so the counter field reads 0 — a probe request.  The device re-sends the initial
+probe response (containing the STRT record and first waveform bytes).
+
+**Effect:** In a 36-frame stream for key=01110000, fi=4 was a byte-for-byte duplicate of
+fi=0 (same db=1101B, same w[0:32] bytes, STRT present at w[10]).  The old code treated it
+as a regular chunk, decoded the STRT bytes as int16 samples (producing T=21587="ST",
+V=21586="RT"), and shifted the running byte alignment for all subsequent frames.
+
+**Fix (`_decode_a5_waveform`):** Check for STRT in every frame, not just fi==0.  Any
+non-fi=0 frame containing STRT is a re-probe — log it and skip (do NOT add to
+`all_chunks`).  Regular chunk path is reached only when `w.find(b"STRT") < 0`.
+
+**Note:** This DLE collision is key-specific.  For key=01110000 (`key4[2:4]=0x0000`),
+counter = `chunk_num * 0x0400`, so counter=0x1000 occurs at chunk 4 for every event with
+this key.  For other keys (e.g. key=0111245a, `key4[2:4]=0x245A`), the chunk counter
+formula `key4[2:4] + n*0x0400` produces different values; the collision only occurs when
+the high byte of any counter is 0x10.
+
+### SUB 5A — metadata false-positive (FIXED 2026-04-15)
+
+**Root cause (confirmed from diagnostic output):**
+The old metadata-frame test was `b"Project:" in w` (single anchor).  For the 36-frame
+desk-thump stream, fi=15 had `b"Project:"` at w[93] inside live ADC data — a coincidental
+4-byte pattern in the waveform.  The frame (134 live sample-sets) was incorrectly skipped.
+
+**Fix:** Require BOTH `b"Project:" in w` AND `b"Client:" in w` to classify a frame as
+metadata.  The real metadata frame (fi=6 in the desk-thump stream) contains both strings
+as part of the compliance-setup ASCII block; random ADC data is statistically unlikely to
+contain both 8-byte sequences.
+
+**Updated check in `_decode_a5_waveform`:**
+```python
+elif b"Project:" in w and b"Client:" in w:
+    log.info("_decode_a5_waveform: fi=%d skipped (metadata frame)", fi)
+    continue
+```
+
+### SUB 5A — 0xFF tail frames beyond record window (confirmed 2026-04-15)
+
+The device bulk-streams flash pages beyond the configured record window.  Un-written flash
+pages read as 0xFF.  Decoded as int16 LE, `0xFF 0xFF = -1`, which maps to ~0 in/s after
+the geo scale factor is applied — producing a visible flat-line at the end of the waveform.
+
+**Confirmed from diagnostic output (desk-thump event, 1024 sps, record_time=3.0 s):**
+- total_samples = 256 (pretrig) + 3072 (post) = 3328
+- samples_decoded = 4417 (36 frames)
+- Excess tail: 4417 − 3328 = 1089 samples (8.7 frames of 0xFF data)
+- Flat-line onset: sample 1960, t=1664ms post-trigger (within the active signal window
+  — earlier than expected because fi=4 re-probe and fi=15 false-positive removed ~270
+  samples; once those bugs are fixed the real onset should be at or beyond total_samples)
+
+**Fix (two parts):**
+1. `_decode_a5_waveform` (Python): already returns `total_samples` alongside
+   `samples_decoded`; the field is populated from compliance config:
+   `total_samples = pretrig_samples + round(record_time * sample_rate)`.
+2. `sfm_webapp.html` (`_buildWaveformCharts`): `display = Math.min(decoded, total_samples)`;
+   `times` array and per-channel `samples` are both sliced to `display` length before
+   plotting — `(channels[ch] || []).slice(0, display)`.  Without this, the chart rendered
+   all `decoded` samples including the 0xFF tail.
+
 ### SUB 1E / 1F — event iteration null sentinel and token position (FIXED, do not re-introduce)
 
 **token_params bug (FIXED):** The token byte was at `params[6]` (wrong). Both 3-31-26 and
@@ -1021,3 +1088,4 @@ call-home.
 - Modem manager — push RV50/RV55 configs via Sierra Wireless API
 - RV55 DCD/DTR issue — newer RV55 firmware doesn't assert DCD by default; units don't
   resume monitoring after call-home disconnect (`--restart-monitoring` flag deferred)
+

diagnose_5a_frames.py

@@ -0,0 +1,328 @@
+#!/usr/bin/env python3
+"""
+diagnose_5a_frames.py -- Frame-by-frame diagnostic for SUB 5A waveform streams.
+
+Usage:
+    python diagnose_5a_frames.py [--host HOST] [--port PORT] [--event INDEX]
+
+Connects to the device, downloads the waveform for the specified event (default 0 =
+most recently stored), and prints detailed per-frame info for every A5 response frame:
+
+  fi=N  | db=NNN B  w=NNN B  | "Project:" in db=[offsets]  in w=[offsets]  <-- METADATA if detected
+        w[0:32]  = <hex>
+        w[-8:]   = <hex>
+        [waveform bytes or ASCII snippet]
+
+Then shows:
+  - total non-metadata frames, total waveform bytes, total sample-sets decoded
+  - compliance-config expected vs decoded counts
+  - sample values at the flat-line onset region (~1700-1820)
+  - first near-zero run location (|T| < 20 for 10+ consecutive samples)
+
+Run with: python diagnose_5a_frames.py 2>&1 | tee /tmp/diag_output.txt
+"""
+
+from __future__ import annotations
+
+import argparse
+import logging
+import struct
+import sys
+
+# -- Setup logging -------------------------------------------------------------
+logging.basicConfig(
+    level=logging.WARNING,   # suppress library noise; we print our own output
+    format="%(asctime)s %(levelname)s %(name)s: %(message)s",
+    stream=sys.stderr,
+)
+
+from minimateplus import MiniMateClient
+from minimateplus.transport import TcpTransport
+
+log = logging.getLogger("diagnose")
+log.setLevel(logging.INFO)
+
+
+def decode_int16_sets(wave: bytes, n: int = 8) -> list[tuple[int, int, int, int]]:
+    """Decode up to n sample-sets from wave bytes as [T, V, L, M] int16 LE."""
+    sets = []
+    for i in range(min(n, len(wave) // 8)):
+        off = i * 8
+        t = struct.unpack_from("<h", wave, off)[0]
+        v = struct.unpack_from("<h", wave, off + 2)[0]
+        l = struct.unpack_from("<h", wave, off + 4)[0]
+        m = struct.unpack_from("<h", wave, off + 6)[0]
+        sets.append((t, v, l, m))
+    return sets
+
+
+def find_all(data: bytes, needle: bytes) -> list[int]:
+    """Return all offsets where needle appears in data."""
+    positions = []
+    start = 0
+    while True:
+        pos = data.find(needle, start)
+        if pos < 0:
+            break
+        positions.append(pos)
+        start = pos + 1
+    return positions
+
+
+def sep(label: str = "") -> None:
+    width = 80
+    if label:
+        pad = max(0, (width - len(label) - 2) // 2)
+        print(f"\n{'-' * pad} {label} {'-' * max(0, width - pad - len(label) - 2)}")
+    else:
+        print("-" * width)
+
+
+def diagnose(frames_data: list[bytes], compliance_config=None) -> None:
+    """Analyse all A5 frames and print diagnostic info."""
+
+    sep("PER-FRAME ANALYSIS")
+    print(f"Total A5 frames received: {len(frames_data)}")
+    print()
+
+    all_chunks: list[tuple[int, bytes]] = []   # (fi, wave_bytes)
+    cumulative_wave_bytes = 0
+
+    for fi, db in enumerate(frames_data):
+        w = db[7:]  # what _decode_a5_waveform sees (db[7:])
+
+        # Find "Project:" in both the full frame data and the w=db[7:] slice
+        proj_in_db = find_all(db, b"Project:")
+        proj_in_w  = find_all(w,  b"Project:")
+
+        # The live detector in client.py uses: b"Project:" in w
+        detected_as_metadata = bool(proj_in_w)
+
+        flag = " <-- METADATA (skipped)" if detected_as_metadata else ""
+        print(f"fi={fi:3d}  db={len(db):5d}B  w={len(w):5d}B  "
+              f"Project: in db={proj_in_db}  in w(db[7:])={proj_in_w}{flag}")
+
+        hex_head = w[:32].hex(' ')
+        hex_tail = w[-8:].hex(' ') if len(w) >= 8 else w.hex(' ')
+        print(f"       w[0:32]  = {hex_head}")
+        print(f"       w[-8:]   = {hex_tail}")
+
+        if fi == 0:
+            sp = w.find(b"STRT")
+            if sp >= 0:
+                strt = w[sp:sp + 21]
+                print(f"       STRT at w[{sp}]: {strt.hex(' ')}")
+                wave = w[sp + 21:]
+                if wave:
+                    sets = decode_int16_sets(wave, 4)
+                    print(f"       wave[sp+21:] first 4 sets (T,V,L,M): {sets}")
+                    all_chunks.append((fi, wave))
+                    cumulative_wave_bytes += len(wave)
+                    print(f"       cum_bytes={cumulative_wave_bytes}  cum_sets={cumulative_wave_bytes // 8}")
+            else:
+                print(f"       *** STRT NOT FOUND ***")
+
+        elif detected_as_metadata:
+            # Print the ASCII content to confirm this is the real metadata frame
+            try:
+                snippet = w.decode("ascii", errors="replace")
+                # Find the first 200 printable characters
+                printable = snippet[:200].replace("\x00", ".").replace("\r", "\n").replace("\n", "\n")
+                print(f"       ASCII: {repr(printable[:140])}")
+            except Exception as e:
+                print(f"       (decode error: {e})")
+
+        else:
+            # Regular chunk: strip 8-byte header
+            if len(w) >= 8:
+                wave = w[8:]
+                all_chunks.append((fi, wave))
+                cumulative_wave_bytes += len(wave)
+                sets = decode_int16_sets(wave, 4)
+                # Count near-zero Tran values
+                all_sets = decode_int16_sets(wave, len(wave) // 8)
+                nz = sum(1 for s in all_sets if abs(s[0]) < 20)
+                print(f"       wave[8:] first 4 sets: {sets}")
+                print(f"       cum_bytes={cumulative_wave_bytes}  cum_sets={cumulative_wave_bytes // 8}  "
+                      f"near-zero(|T|<20): {nz}/{len(all_sets)}")
+
+        print()
+
+    # -- Waveform value analysis ------------------------------------------------
+    sep("WAVEFORM DECODE")
+
+    cc_sr = 1024
+    cc_rt = None
+    pretrig = 256
+    total_expected = 0
+
+    if compliance_config:
+        cc_sr = compliance_config.sample_rate or 1024
+        cc_rt = compliance_config.record_time
+        pretrig = int(round(0.25 * cc_sr))
+        if cc_rt:
+            total_expected = pretrig + int(round(cc_rt * cc_sr))
+        print(f"Compliance: sr={cc_sr} sps  record_time={cc_rt} s  "
+              f"pretrig={pretrig}  total_expected={total_expected}")
+    else:
+        print("No compliance config -- using defaults: sr=1024, pretrig=256")
+
+    total_wave_bytes = sum(len(w) for _, w in all_chunks)
+    total_sets_raw = total_wave_bytes // 8
+    print(f"Non-metadata frames: {len(all_chunks)}  "
+          f"Total wave bytes: {total_wave_bytes}  "
+          f"Raw sample-sets: {total_sets_raw}")
+
+    # Alignment-corrected decode (matches _decode_a5_waveform exactly)
+    tran: list[int] = []
+    running_offset = 0
+    for fi, wave in all_chunks:
+        align = running_offset % 8
+        skip  = (8 - align) % 8
+        if skip > 0 and skip < len(wave):
+            usable = wave[skip:]
+        elif align == 0:
+            usable = wave
+        else:
+            running_offset += len(wave)
+            continue
+        n_usable = len(usable) // 8
+        for i in range(n_usable):
+            tran.append(struct.unpack_from("<h", usable, i * 8)[0])
+        running_offset += len(wave)
+
+    n_decoded = len(tran)
+    print(f"Alignment-corrected decoded Tran samples: {n_decoded}")
+    if compliance_config and cc_rt:
+        print(f"Expected: {total_expected}  Decoded: {n_decoded}  "
+              f"Excess (tail): {max(0, n_decoded - total_expected)}")
+
+    print()
+    print(f"First 16 Tran: {tran[:16]}")
+    if n_decoded >= 32:
+        print(f"Last  16 Tran: {tran[-16:]}")
+
+    # -- Flat-line onset search -------------------------------------------------
+    sep("FLAT-LINE ONSET (first run of 10+ consecutive |Tran| < 20)")
+
+    run_start = None
+    run_len = 0
+    onset_found = False
+    for i, v in enumerate(tran):
+        if abs(v) < 20:
+            if run_start is None:
+                run_start = i
+            run_len += 1
+        else:
+            if run_len >= 10:
+                t_ms = (run_start - pretrig) * 1000.0 / cc_sr
+                print(f"  First near-zero run: sample {run_start}-{run_start + run_len - 1}  "
+                      f"(t={t_ms:.1f}ms post-trigger)  length={run_len}")
+                onset_found = True
+                break
+            run_start = None
+            run_len = 0
+    else:
+        if run_len >= 10 and run_start is not None:
+            t_ms = (run_start - pretrig) * 1000.0 / cc_sr
+            print(f"  Near-zero run at end: sample {run_start}-{n_decoded - 1}  "
+                  f"(t={t_ms:.1f}ms post-trigger)  length={run_len}")
+            onset_found = True
+
+    if not onset_found:
+        print("  No near-zero run of 10+ samples found (waveform looks active throughout)")
+
+    # Print samples around the expected flat-line onset (~1700-1820)
+    if n_decoded >= 1700:
+        print()
+        print("Tran samples [1700:1820] (10 per line):")
+        for row_start in range(1700, min(1820, n_decoded), 10):
+            row = tran[row_start:row_start + 10]
+            t_ms_row = (row_start - pretrig) * 1000.0 / cc_sr
+            print(f"  [{row_start:4d}] (t={t_ms_row:6.1f}ms): {row}")
+    else:
+        print(f"  Only {n_decoded} samples decoded -- range 1700-1820 not available")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Diagnose A5 5A waveform frames")
+    parser.add_argument("--host",  default="63.43.212.232", help="Device IP")
+    parser.add_argument("--port",  type=int, default=9034,  help="TCP port")
+    parser.add_argument("--event", type=int, default=0,     help="Event index (0=first stored)")
+    args = parser.parse_args()
+
+    print(f"Connecting to {args.host}:{args.port} ...")
+    print(f"Target event index: {args.event}")
+    print()
+
+    transport = TcpTransport(args.host, port=args.port)
+    with MiniMateClient(transport=transport) as client:
+        info = client.connect()
+        print(f"Device: serial={info.serial}  firmware={info.firmware_version}")
+        compliance_config = info.compliance_config
+        if compliance_config:
+            print(f"Compliance: sample_rate={compliance_config.sample_rate}  "
+                  f"record_time={compliance_config.record_time}")
+        print()
+
+        proto = client._proto
+        assert proto is not None
+
+        # -- Walk to the target event ------------------------------------------
+        log.info("Reading first event key (SUB 1E) ...")
+        first_key4, first_data8 = proto.read_event_first(token=0)
+        print(f"First event key: {first_key4.hex()}")
+
+        cur_key4   = first_key4
+        cur_data8  = first_data8
+        event_idx  = 0
+
+        while event_idx < args.event:
+            # 0A required before each 1F to establish device context
+            proto.read_waveform_header(cur_key4)
+            next_key4, next_data8 = proto.advance_event(browse=True)
+            if next_data8[4:8] == b"\x00\x00\x00\x00":
+                print(f"Only {event_idx + 1} events available; cannot reach index {args.event}")
+                return
+            cur_key4  = next_key4
+            cur_data8 = next_data8
+            event_idx += 1
+            print(f"  advanced to event {event_idx}: key={cur_key4.hex()}")
+
+        print(f"\nDownloading event {args.event}: key={cur_key4.hex()}")
+
+        # -- Full download sequence (matches get_events download-mode) ---------
+        log.info("0A: read_waveform_header ...")
+        proto.read_waveform_header(cur_key4)
+
+        log.info("1E(0xFE): arm device for 5A ...")
+        proto.read_event_first(token=0xFE)
+
+        log.info("0C: read_waveform_record ...")
+        wfm_raw = proto.read_waveform_record(cur_key4)
+        print(f"0C waveform record: {len(wfm_raw)} bytes")
+
+        log.info("1F(0xFE): arm 5A state machine ...")
+        arm_key4, _ = proto.advance_event(browse=False)
+        print(f"1F(arm) returned key: {arm_key4.hex()}")
+
+        log.info("POLLx3 ...")
+        for i in range(3):
+            proto.poll()
+            print(f"  POLL {i+1}/3 OK")
+
+        print(f"\nStarting 5A bulk stream for key={cur_key4.hex()} ...")
+        frames_data = proto.read_bulk_waveform_stream(
+            cur_key4,
+            stop_after_metadata=False,
+            max_chunks=2048,
+        )
+        print(f"5A complete: {len(frames_data)} A5 frames")
+        print()
+
+        # -- Run the diagnostic ------------------------------------------------
+        diagnose(frames_data, compliance_config)
+
+
+if __name__ == "__main__":
+    main()

minimateplus/client.py

@@ -1440,29 +1440,50 @@ def _decode_a5_waveform(
     for fi, db in enumerate(frames_data):
         w = db[7:]
 
-        # A5[0]: waveform begins immediately after the 21-byte STRT record.
-        # Confirmed 2026-04-14: there is NO preamble after STRT — bytes 21+
-        # are raw ADC sample data.  The earlier sp+27 skip was eating 6 bytes
-        # of real waveform, misaligning the channel decode for all subsequent
-        # frames.
-        if fi == 0:
+        # ── Probe frames (fi==0 AND any re-probe the device sends mid-stream) ────
+        # A5[0] always contains the STRT record.  For event key 0x01110000,
+        # chunk 4 (counter=0x1000) has 0x10 in the counter high byte; the device
+        # DLE-decodes the params and sees counter=0x0000 (probe), so it responds
+        # with a duplicate probe frame containing the same STRT.  The diagnostic
+        # from 2026-04-15 confirmed this: fi=4 was byte-for-byte identical to fi=0
+        # (same db length 1101B, same STRT at w[10], same first 32 bytes).
+        #
+        # Handling: any frame — not just fi==0 — that contains the STRT magic is
+        # treated as a probe frame.  Waveform starts at strt_pos + 21 (no preamble).
+        # Re-probe frames are complete duplicates of fi=0 (device re-sends the
+        # beginning of the event), so their post-STRT waveform bytes are DROPPED
+        # to avoid injecting duplicate data into the stream.
         sp = w.find(b"STRT")
-            if sp < 0:
-                continue
+        if sp >= 0:
+            if fi == 0:
                 wave = w[sp + 21 :]
                 log.info(
-                "_decode_a5_waveform: A5[0] waveform starts at sp+21; "
-                "first 24 wave bytes: %s",
-                wave[:24].hex(' '),
+                    "_decode_a5_waveform: A5[0] probe — STRT at w[%d], "
+                    "waveform starts at sp+21; first 24 wave bytes: %s",
+                    sp, wave[:24].hex(' '),
                 )
+            else:
+                # Re-probe frame: device re-sent probe in response to a chunk
+                # request whose counter byte happened to be 0x10 (DLE).
+                # The post-STRT bytes are a duplicate of the initial waveform
+                # — drop this frame entirely to avoid double-counting data.
+                log.info(
+                    "_decode_a5_waveform: fi=%d re-probe (STRT at w[%d]) — "
+                    "skipped (duplicate probe response from device)",
+                    fi, sp,
+                )
+                continue
 
-        # Metadata frame: contains "Project:", "Client:", etc. strings.
-        # Originally assumed to be always fi==7 (A5[7] in 4-2-26 blast capture),
-        # but confirmed variable position — it appears at whatever chunk index the
-        # device places it (observed at fi=6 for desk-thump events 2026-04-14).
-        # Skip ANY frame whose raw bytes contain b"Project:" — this is the same
-        # anchor used by stop_after_metadata in read_bulk_waveform_stream.
-        elif b"Project:" in w:
+        # Metadata frame: contains BOTH "Project:" and "Client:" strings.
+        # Requiring two compliance anchors prevents false positives where ADC
+        # bytes accidentally spell "Project:" (confirmed false positive at fi=15
+        # in the 2026-04-15 desk-thump download — only "Project:" appeared there,
+        # not "Client:").  The real metadata frame always contains both.
+        # This is the same anchor used by stop_after_metadata in
+        # read_bulk_waveform_stream (which only checks "Project:" — see note
+        # there about the asymmetry: stopping early is fine with one anchor,
+        # but skipping a waveform frame requires higher confidence).
+        elif b"Project:" in w and b"Client:" in w:
             log.info("_decode_a5_waveform: fi=%d skipped (metadata frame)", fi)
             continue
 
@@ -2248,20 +2269,4 @@ def _decode_monitor_status(data: bytes) -> MonitorStatus:
     # Payload length varies (46–49 bytes) but the battery/memory block is always
     # the last 10 bytes. No checksum byte — it was stripped by S3FrameParser.
     #
-    #   section[-10:-8]   battery  × 100   uint16 BE   0x02A8 = 6.80 V
-    #   section[-8 :-4]   memory_total     uint32 BE   ≈ 960 KB on BE11529
-    #   section[-4:]      memory_free      uint32 BE   decreases as events fill
-    #
-    # Confirmed stable across IDLE (46b), MONITORING (48-49b) variants.
-    if len(section) >= 10:
-        batt_raw     = struct.unpack(">H", section[-10:-8])[0]
-        battery_v    = batt_raw / 100.0
-        memory_total = struct.unpack(">I", section[-8:-4])[0]
-        memory_free  = struct.unpack(">I", section[-4:])[0]
-
-    return MonitorStatus(
-        is_monitoring=is_monitoring,
-        battery_v=battery_v,
-        memory_total=memory_total,
-        memory_free=memory_free,
-    )
+    #   section[-1

sfm/sfm_webapp.html

@@ -1555,6 +1555,14 @@ function _buildWaveformCharts(data, chartsEl, emptyEl, chartsStore) {
   const sr      = data.sample_rate || 1024;
   const pretrig = data.pretrig_samples || 0;
   const decoded = data.samples_decoded || 0;
+  // Clip display to total_samples (pretrig + post_trig from compliance config).
+  // The device bulk-streams zero-padded (0xFF = -1) frames beyond the configured
+  // record window; without clipping these appear as a flat line at ~0 in/s past
+  // the end of the actual recording.  Confirmed 2026-04-15: a 36-frame 5A stream
+  // for a 3.25s event (total_samples=3328) contained 19 trailing all-0xFF frames
+  // (2457 extra samples) that caused a visible flat-line in the waveform display.
+  const total   = (data.total_samples && data.total_samples > 0) ? data.total_samples : decoded;
+  const display = Math.min(decoded, total);
   const channels = data.channels || {};
 
   // Destroy old chart instances
@@ -1574,7 +1582,7 @@ function _buildWaveformCharts(data, chartsEl, emptyEl, chartsStore) {
     return;
   }
 
-  const times = Array.from({length: decoded}, (_, i) => ((i - pretrig) / sr * 1000).toFixed(2));
+  const times = Array.from({length: display}, (_, i) => ((i - pretrig) / sr * 1000).toFixed(2));
   if (emptyEl) emptyEl.style.display = 'none';
   chartsEl.style.display = 'flex';
   chartsEl.style.flexDirection = 'column';
@@ -1584,8 +1592,8 @@ function _buildWaveformCharts(data, chartsEl, emptyEl, chartsStore) {
   const micPeakPsi = data.peak_values?.micl_psi ?? null;
 
   for (const [ch, color] of Object.entries(CHANNEL_COLORS)) {
-    const samples = channels[ch];
-    if (!samples || samples.length === 0) continue;
+    const samples = (channels[ch] || []).slice(0, display);
+    if (samples.length === 0) continue;
 
     const isGeo = ch !== 'Mic';
     let plotData, peakLabel, yUnit, ttFmt, tickFmt;
@@ -2118,39 +2126,4 @@ document.getElementById('api-base').value = window.location.origin;
       </div>
       <button onclick="closeWfModal()"
               style="background:none; border:none; color:var(--text-dim); cursor:pointer;
-                     font-size:20px; line-height:1; padding:0 2px; flex-shrink:0;"
-              title="Close (Esc)">×</button>
-    </div>
-
-    <!-- Peaks bar — reuses .peaks-bar styles from live Events tab -->
-    <div class="peaks-bar" id="wf-modal-peaks">
-      <div class="pk"><div class="pk-label">Tran</div><div class="pk-value pk-tran" id="wf-mpk-tran">—</div></div>
-      <div class="pk"><div class="pk-label">Vert</div><div class="pk-value pk-vert" id="wf-mpk-vert">—</div></div>
-      <div class="pk"><div class="pk-label">Long</div><div class="pk-value pk-long" id="wf-mpk-long">—</div></div>
-      <div class="pk"><div class="pk-label">MicL</div><div class="pk-value pk-mic"  id="wf-mpk-mic">—</div></div>
-      <div class="pk"><div class="pk-label">PVS</div><div class="pk-value pk-pvs"  id="wf-mpk-pvs">—</div></div>
-    </div>
-
-    <!-- Debug panel (same as live debug panel, hidden by default) -->
-    <div id="wf-modal-debug"
-         style="display:none; background:#0d1117; border-bottom:1px solid #21262d;
-                padding:5px 16px; font-family:monospace; font-size:11px; color:#6e7681; line-height:1.8">
-      <span style="float:right; cursor:pointer; color:#484f58; text-decoration:underline"
-            onclick="document.getElementById('wf-modal-debug').style.display='none'">hide</span>
-      <div id="wf-modal-debug-content"></div>
-    </div>
-
-    <!-- Waveform area -->
-    <div style="flex:1; overflow-y:auto; min-height:200px;">
-      <div id="wf-modal-empty"
-           style="display:flex; flex-direction:column; align-items:center;
-                  justify-content:center; padding:60px 20px; color:var(--text-dim); gap:12px;">
-        <p>Loading…</p>
-      </div>
-      <div id="wf-modal-charts" style="display:none;"></div>
-    </div>
-
-  </div>
-</div>
-</body>
-</html>
+                     font-si