Sorry, but at least for me (and probably will be for others) completely unclear what "hook" really means, what it really does, especially given that it is "self-designed", how/when instrumentation and analysis are actually performed, etc, etc.
Also, as you might have noticed from past discussions, note that small JaCoCo Team (primary responders in this ML) works on project in spare time, has very basic knowledge of Android specifics,
and does not provide integration with Android directly (Java bytecode is not Dalvik bytecode, Java agents are not available on Android) - third-party integrations use our APIs.
Finally if for you it has "high-cost to verify", then, summarizing all above, for us even attempts to guess will have much higher cost.
Please also note that unlike "self-designed hook", all code and documentation about JaCoCo are available publicly.
The best way to get attention in investigation of your "interesting phenomenon" - is to provide complete, runnable, debuggable and verifiable by others example of this phenomenon.
However speaking about "total method num but with no coverage" :
if information about lines is there and no information about coverage at all, then
or there is no information about executed classes - see "sessions" page in html report, "execinfo" command in command line interface (
https://www.jacoco.org/jacoco/trunk/doc/cli.html ) and check third-party code that retrieves this information from Android and performs instrumentation.
Hope this helps.
Regards,
Evgeny