match_count change depending of the time that detects

[gonzafirewall]

I am using olaf some time and detects that the match_counts differs some times largely depending of when in time the stored audio is detected in the segment to query, let met explain a little more in detail with some examples.

[MDB database statistics]
=========================
> Size of database page:        4096
> Depth of the B-tree:          2
> Number of items in databases: 1134
> File size of the databases:   0MB
=========================

  key   duration(s)     Prints(#)       Prints(#/s)     path
  3771614152    37.488s   1136fps       30.303fps/s     '/home/gonza/audios/customer/AUDIO.mp3'
Number of songs (#):    1
Total duration (s):     37.488
Avg prints/s (fp/s):    30.303

This is the audio stored and for example i check where is in an audio segment of 1 hour

olaf query ./reaudit_files/segments/2024-12-03T00:00:00.000Z.mp3
1, 1, 2024-12-03T00:00:00.000Z.mp3, 0, match count (#), q start (s) , q stop (s), ref path, ref ID, ref start (s), ref stop (s)
1, 1, 2024-12-03T00:00:00.000Z.mp3, 0, 127, 1832.00, 1859.85, /home/gonza/audios/customer/AUDIO.mp3, 3771614152, 2.25, 30.10
Matched 147153 fp's from 3619.0s (41 fp/s) in 4.383s (826 times realtime)

This report 127 match counts and detect from 1832 to 1859 a total aprox of 28 seconds if round to upper

and only changing the original segment cutting 12 seconds of the begging with this command

ffmpeg -i ./reaudit_files/segments/2024-12-03T00:00:00.000Z.mp3 -ss 12 -acodec copy ./reaudit_files/segments/2024-12-03T00:00:00.000Z.mp3_2.mp3

olaf query ./reaudit_files/segments/2024-12-03T00:00:00.000Z_2.mp3
1, 1, 2024-12-03T00:00:00.000Z_2.mp3, 0, match count (#), q start (s) , q stop (s), ref path, ref ID, ref start (s), ref stop (s)
1, 1, 2024-12-03T00:00:00.000Z_2.mp3, 0, 131, 1818.29, 1849.64, /home/gonza/audios/customer/AUDIO.mp3, 3771614152, 0.55, 31.90
Matched 145160 fp's from 3607.0s (40 fp/s) in 4.379s (824 times realtime)

This ends in a detection of aprox 31.3 seconds i test a lot and change a lot depending of the seconds cutted trom the original segment is this ok? I need to handle this variations? or can tune some config on olaf source to stabilize this detections in a better way.
Thanks in advance your work is great.