Description:
When you design a LV like Atlas, you probably do it using separate tanks and fuel priority. The problem is, when you do so, in-flight vacuum delta-V readings change over time. Moreover, just after the stage separation, your can see your second stage gaining more delta-V while the engines are still burning. It looks like mechjeb gets confused by the upper tank residuals. It considers it as a "free fuel", therefore increasing the remaining delta-V amount.

MechJeb version: DEV 2.14.3.0-1412

Steps to reproduce:
Use this craft as an example.
Atlas.zip
Launch from Baikonur.
Orbit parameters: 190x190km, current inclination, 40m/s pitch start, 0.85d/s pitch angle.

Expected result:
Mechjeb's prediction about if you can reach orbit 10 seconds before the orbital insertion should match the actual result. If it tells you you have some margin - you should reach orbit. If you don't - you should not.

Actual result:
Till the very end mechjeb believes it's got enough delta-V, but still fails to reach orbit just 100-150m/s short.