Hi, thanks for sharing this.
The first thing to check is whether there is an inconsistency in the way the tests are counted. When NCrunch reports the number of tests it's monitoring, this is the number of leaf-node tests only - it does not include virtual constructs such as fixtures, parameterised enclosures, assemblies, etc.
The next thing to check for is tests that may be bundled together by NCrunch. Depending upon your test framework and framework utilisation type, NCrunch may be representing groups of tests as singular tests to work around test framework limitations (particularly if you are using static analysis to discover tests).
Following this, I'd recommend trying some exports to see if you can spot any tests that are obviously missing from the results you are expecting. The difference is large enough that simply sifting through the tests in alphabetical order should yield a difference fairly quickly. Once you've found one or two, you've likely discovered the pattern and you'll know a great deal more :)
