Are you contact printing the test negs or eyeballing them ?
If a contact print is giving you the feedback, good.
You have a classic problem. There is a classic solution.
Try a second bath of 10% borax or kodalk, effectively making D76 a two bath process.
Develop in D76, using 75% of your best 'shadow' time.
(straight, or 1+1, test and see)
Drain the film, don't rinse, and develop 3' in the Borax solution.
This is just a variation of Ansel's 2 bath D23 (which is functionally D76, anyhow).
Barry Thornton talked about this technique,
and there is a nicely coherent explanation of what to do:
Dead Link Removed