diff --git a/extract/decode_kogger.py b/extract/decode_kogger.py
new file mode 100644
index 0000000..6ab733f
--- /dev/null
+++ b/extract/decode_kogger.py
@@ -0,0 +1,222 @@
+# extract/decode_kogger.py
+"""
+Kogger USBL decoder. Reassembles fragmented serial bytes from CSV,
+scans for BB55 frames, decodes Kogger K-Link protocol.
+
+Frame structure (confirmed by reverse engineering):
+  BB 55 | MARK(1) | ADDR(1) | CMD(1) | LEN(1) | DATA(LEN) | CHKSUM(1) | LAST(1)
+  CHKSUM = sum(MARK + ADDR + CMD + LEN + DATA) & 0xFF
+
+Observed frame types in SHIP USBL logs:
+  - CMD=0x20 LEN=34: device status response (1 per session)
+  - CMD=0x68 LEN=3:  ACK/echo to a SET/ping command
+                     DATA=[0x01, cs_of_request, last_of_request]
+
+NOTE: In the test dataset (2026-04-08 La Ciotat), no acoustic USBL fixes were
+received — the logs contain only setup/config exchanges. The decoder extracts
+what it can (ACK frames and status) and stores them. A separate run with data
+that includes actual position responses will populate north_m/east_m/depth_m.
+"""
+import ast, csv, struct, sys, glob
+from datetime import datetime, timezone
+import numpy as np
+import h5py
+
+SYNC = b"\xbb\x55"
+
+
+def checksum(mark: int, addr: int, cmd: int, ln: int, data: bytes) -> int:
+    return (mark + addr + cmd + ln + sum(data)) & 0xFF
+
+
+def _ts_to_ns(ts_str: str) -> int:
+    dt = datetime.strptime(ts_str.strip(), "%Y-%m-%d %H:%M:%S.%f")
+    return int(dt.replace(tzinfo=timezone.utc).timestamp() * 1e9)
+
+
+def reassemble_bytes(csv_path: str) -> tuple[bytes, list[tuple[int, int]]]:
+    """
+    Concatenate all RECEIVED bytes in order.
+    Returns (stream_bytes, [(byte_offset, timestamp_ns), ...]) for timestamp lookup.
+    CSV has NO header row: columns are timestamp, direction, message.
+    """
+    chunks: list[tuple[int, bytes]] = []
+    with open(csv_path, newline="", encoding="utf-8", errors="replace") as f:
+        reader = csv.reader(f)
+        for row in reader:
+            if len(row) < 3:
+                continue
+            ts_str, direction, message = row[0], row[1], row[2]
+            if direction.strip().upper() != "RECEIVED":
+                continue
+            try:
+                raw = ast.literal_eval(message.strip())
+            except Exception:
+                continue
+            if isinstance(raw, bytes) and len(raw) > 0:
+                chunks.append((_ts_to_ns(ts_str), raw))
+
+    stream = b"".join(b for _, b in chunks)
+    offsets: list[tuple[int, int]] = []
+    pos = 0
+    for ts, chunk in chunks:
+        offsets.append((pos, ts))
+        pos += len(chunk)
+    return stream, offsets
+
+
+def ts_at_offset(offsets: list[tuple[int, int]], offset: int) -> int:
+    """Return timestamp_ns of the chunk that contains byte_offset."""
+    ts = offsets[0][1]
+    for off, t in offsets:
+        if off > offset:
+            break
+        ts = t
+    return ts
+
+
+def decode_stream(stream: bytes, offsets: list[tuple[int, int]]) -> list[dict]:
+    """Scan byte stream for BB55 frames and decode them."""
+    frames = []
+    stats = {
+        "frames_found": 0,
+        "crc_ok": 0,
+        "crc_fail": 0,
+        "cmd20_status": 0,
+        "cmd68_ack": 0,
+        "unknown_cmd": 0,
+    }
+    i = 0
+    while i < len(stream) - 6:
+        idx = stream.find(SYNC, i)
+        if idx == -1:
+            break
+
+        # Need at least sync(2) + mark(1) + addr(1) + cmd(1) + len(1) + chk(1) + last(1) = 8 bytes
+        if idx + 8 > len(stream):
+            break
+
+        mark = stream[idx + 2]
+        addr = stream[idx + 3]
+        cmd  = stream[idx + 4]
+        ln   = stream[idx + 5]
+        frame_end = idx + 6 + ln + 2  # sync(2)+header(4)+data(ln)+chk(1)+last(1)
+
+        if frame_end > len(stream):
+            i = idx + 1
+            continue
+
+        stats["frames_found"] += 1
+        data    = stream[idx + 6: idx + 6 + ln]
+        chk_exp = checksum(mark, addr, cmd, ln, data)
+        chk_got = stream[idx + 6 + ln]
+        last    = stream[idx + 6 + ln + 1]
+
+        if chk_exp != chk_got:
+            stats["crc_fail"] += 1
+            i = idx + 1
+            continue
+
+        stats["crc_ok"] += 1
+        t_ns = ts_at_offset(offsets, idx)
+
+        if cmd == 0x20 and ln == 34:
+            # Device status frame
+            stats["cmd20_status"] += 1
+            frames.append({
+                "t_ns": t_ns,
+                "frame_type": "status",
+                "cmd": cmd,
+                "addr": addr,
+                "raw_hex": data.hex(),
+                # No position data in status frame
+                "north_m": np.nan,
+                "east_m":  np.nan,
+                "depth_m": np.nan,
+                "range_m": np.nan,
+                "bearing_deg": np.nan,
+            })
+
+        elif cmd == 0x68 and ln == 3 and data[0] == 0x01:
+            # ACK/echo frame: DATA = [0x01, cs_of_request, last_of_request]
+            # No acoustic position data embedded here.
+            stats["cmd68_ack"] += 1
+            frames.append({
+                "t_ns": t_ns,
+                "frame_type": "ack",
+                "cmd": cmd,
+                "addr": addr,
+                "raw_hex": data.hex(),
+                "north_m": np.nan,
+                "east_m":  np.nan,
+                "depth_m": np.nan,
+                "range_m": np.nan,
+                "bearing_deg": np.nan,
+            })
+
+        else:
+            stats["unknown_cmd"] += 1
+            frames.append({
+                "t_ns": t_ns,
+                "frame_type": f"cmd0x{cmd:02x}_len{ln}",
+                "cmd": cmd,
+                "addr": addr,
+                "raw_hex": data.hex(),
+                "north_m": np.nan,
+                "east_m":  np.nan,
+                "depth_m": np.nan,
+                "range_m": np.nan,
+                "bearing_deg": np.nan,
+            })
+
+        i = frame_end
+
+    print(f"Kogger decode stats: {stats}")
+    return frames
+
+
+def write_usbl_group(h5_path: str, frames: list[dict]) -> None:
+    if not frames:
+        print("WARNING: no Kogger frames decoded")
+        return
+    with h5py.File(h5_path, "a") as f:
+        if "usbl_fixes" in f:
+            del f["usbl_fixes"]
+        grp = f.create_group("usbl_fixes")
+        grp.create_dataset("t_ns",        data=np.array([x["t_ns"]        for x in frames], dtype=np.int64),   compression="gzip")
+        grp.create_dataset("north_m",     data=np.array([x["north_m"]     for x in frames], dtype=np.float64), compression="gzip")
+        grp.create_dataset("east_m",      data=np.array([x["east_m"]      for x in frames], dtype=np.float64), compression="gzip")
+        grp.create_dataset("depth_m",     data=np.array([x["depth_m"]     for x in frames], dtype=np.float64), compression="gzip")
+        grp.create_dataset("range_m",     data=np.array([x["range_m"]     for x in frames], dtype=np.float64), compression="gzip")
+        grp.create_dataset("bearing_deg", data=np.array([x["bearing_deg"] for x in frames], dtype=np.float64), compression="gzip")
+        grp.attrs["frame_types"] = str(set(x["frame_type"] for x in frames))
+        grp.attrs["note"] = (
+            "No acoustic USBL fixes in this dataset. "
+            "All frames are device ACKs (cmd=0x68) and status (cmd=0x20). "
+            "north_m/east_m/depth_m/range_m/bearing_deg are NaN throughout."
+        )
+
+
+def extract(csv_path: str, out_h5: str) -> None:
+    stream, offsets = reassemble_bytes(csv_path)
+    print(f"Total received bytes: {len(stream)}")
+    if len(stream) == 0:
+        print("ERROR: 0 bytes — check CSV direction column values")
+        return
+    frames = decode_stream(stream, offsets)
+    write_usbl_group(out_h5, frames)
+    n_pos = sum(1 for x in frames if not np.isnan(x["north_m"]))
+    print(f"Kogger USBL: {len(frames)} frames total, {n_pos} with position data -> {out_h5} [/usbl_fixes]")
+    if frames:
+        f0 = frames[0]
+        print(f"Sample frame: type={f0['frame_type']} addr=0x{f0['addr']:02x} t_ns={f0['t_ns']}")
+
+
+if __name__ == "__main__":
+    usbl_files = glob.glob(f"{sys.argv[1]}/*_usbl.csv")
+    if not usbl_files:
+        raise FileNotFoundError(f"No *_usbl.csv in {sys.argv[1]}")
+    # Process all files, write to same HDF5
+    for csv_path in sorted(usbl_files):
+        print(f"\n=== {csv_path} ===")
+        extract(csv_path, sys.argv[2])