diff --git a/dfextract_3.py b/dfextract_3.py
new file mode 100644
index 0000000..0fdffdd
--- /dev/null
+++ b/dfextract_3.py
@@ -0,0 +1,68 @@
+import bz2, struct, os, stat
+from io import BytesIO
+
+sfai = open('dfdata.sfai', 'rb')
+
+h = struct.unpack('128L', sfai.read(0x200))
+assert h[0] == 0x49414653 # "SFAI"
+start = h[6]
+size = h[7]
+print("INDEX: %08X => %08X" % (start, start + size))
+sfai.seek(start)
+buff = sfai.read(size)
+
+buff = b'BZh9' + buff[4:]
+idx = BytesIO(bz2.decompress(buff))
+
+csizes = {}
+files = {}
+fsizes = {}
+fcount = {}
+for x in range(0, 32):
+    try:
+        files[x] = open('dfdata_%04d.sfad' % x, 'rb')
+        csizes[x] = 0
+        fcount[x] = 0
+        fsizes[x] = os.path.getsize('dfdata_%04d.sfad' % x)
+    except FileNotFoundError:
+        break
+
+num_tomes = len(files)
+
+if not os.path.isdir('out'):
+    os.mkdir('out')
+log = open('out/file.lst', 'w')
+
+while True:
+    x = idx.read(0xBD)
+    if not x:
+        break
+    n = x[0:x.find(0)].decode('utf-8')
+    id, t, s1, fn, p, s2, u3, u4 = struct.unpack('7Lb', x[160:189])
+    csizes[fn] += s2
+    fcount[fn] += 1
+    pp = 'out/'
+    for y in n.split('/')[:-1]:
+        pp += y + '/'
+        if not os.path.isdir(pp):
+            os.mkdir(pp)
+    print("%(n)r ID:%(id)d @ %(p)08X (T%(fn)02d) Size:%(s1)d/%(s2)d %(u3)08X %(u4)02X" % locals(), file=log)
+    files[fn].seek(p)
+    buff = files[fn].read(s2)
+    if s1 != s2:
+        if buff[:4] == b'SFZM':
+            buff = b'BZh9' + buff[4:]
+            buff = bz2.decompress(buff)
+        else:
+            print('Warning! %r unknown compression.' % n)
+    f = open('out/' + n, 'wb')
+    f.write(buff)
+    f.close()
+
+log.close()
+for x in range(num_tomes):
+    files[x].close()
+
+print("================ DONE ==================")
+for x in range(num_tomes):
+    print("Tome %02d - %d files, %.3f%% lost" % (x, fcount[x], 100.0 * (fsizes[x] - csizes[x]) / fsizes[x]))