Four microseconds precision chronometer

I have tested precision with 249 and one interrupt by millisecond and this choice is guided by timer implementation.

Apparently TCNT1 is cleared at the next front clock when TCNT1 is equal to OCR1A register, so it is equivalent to count 250.

I had found this bug when I had set 250, the chronometer took delay, with 249 the chronometer appears to be ok.

In interpolating, 24999 should be the good value and justified by datasheet of ATmega168 but if someone can justify it ?