int32 is a 32-bit integer, as in 4 bytes
int64 is a 64-bit integer, as in 8 bytes
The "root cause" is that when the sqlite3 (pysqlite2) wrapper asks the python variable what "type" it is, it responds that the closest type it is to a supported type is a "byte array", "buffer", or "blob", so the byte-array is bound to the statement as a blob.
pysqlite2 (and apsw) only recognize Python base types None, int (long), float, str (unicode) or types that conform to the buffer protocol.
np.int32 and np.int64 are not python types that correspond to any python type, but do support the buffer protocol, so they are sent to the database as blobs.
Registering the adapters tells the pysqlite2 wrapper layer how to convert that type into a supported python type.
The same thing happens if you used a ctypes type. pysqlite2 does not know that, for example, ctypes.c_short can be converted to a python integer. It does however support the buffer protocol, so the bytes are stored as a blob. If you want to store them as an integer, you have to convert them to an integer first, or register an adapter that can convert the arbitrary type into a recognizable and supported base type.
```
>python
Python 3.9.1 (tags/v3.9.1:1e5d33e, Dec 7 2020, 17:08:21) [MS
Type "help", "copyright", "credits" or "license" for more inf
>>> import ctypes
>>> a = ctypes.c_short(4)
>>> a
c_short(4)
>>> import sqlite3
>>> db = sqlite3.connect('')
>>> db.execute('create table x(x integer)')
<sqlite3.Cursor object at 0x00000210DBE75AB0>
>>> db.execute('insert into x values (?)', (a,))
<sqlite3.Cursor object at 0x00000210DBE759D0>
>>> for row in db.execute('select x from x'):
... print(row)
...
(b'\x04\x00',)
>>> a.value
4
>>> sqlite3.register_adapter(ctypes.c_short, lambda x: x.value)
>>> db.execute('insert into x values (?)', (a,))
<sqlite3.Cursor object at 0x00000210DBE75AB0>
>>> for row in db.execute('select x from x'):
... print(row)
...
(b'\x04\x00',)
(4,)
>>>
```