Compatibility
This library replicates Excel behavior. There're 199,252 tests verifying the results against Excel 2010 and their number can be raised significantly by adding new test values. Several tests check the function properties, e.g. that bond duration can't be greater than maturity.
The current version matches Excel 2010, which is slightly different from 2003, see Function Improvements in Excel 2010.
Note, that console tests require Excel, whereas the unit tests can be run even on mono  their parameters and expected results are stored in files.
However, there're still some differences comparing to Excel. Read more about OpenOffice vs Excel here.
COUPDAYS
The Excel algorithm doesn't respect equality coupDays = coupDaysBS + coupDaysNC
. The library result differs from Excel by +/ one or two days when the date spans a leap year. (office docs)
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 

VDB
In the Excel version of this algorithm the depreciation in the period (0,1) is not the same as the sum of the depreciations in periods (0,0.5) (0.5,1)
Notice that in Excel by using '1' (no_switch) instead of '0' as the last parameter everything works as expected.
In truth, the last parameter should have no influence in the calculation given that in the first period there is no switch to sln depreciation.
Overall, the algorithm is correct, even if it disagrees with Excel when startperiod is fractional. (office docs)
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23: 

AMORDEGRC
ExcelCompliant is used because Excel stores 13 digits. AmorDegrc algorithm rounds numbers
and returns different results unless the numbers get rounded to 13 digits before rounding them.
I.E. 22.49999999999999 is considered 22.5 by Excel, but 22.4 by the .NET framework. (office docs)
1: 2: 3: 4: 5: 6: 7: 8: 

DDB
Excel Ddb has two interesting characteristics:
1. It special cases ddb for fractional periods between 0 and 1 by considering them to be 1
2. It is inconsistent with VDB(..., True) for fractional periods, even if VDB(..., True) is defined to be the same as ddb. The algorithm for VDB is theoretically correct.
This function makes the same 1. adjustment.(office docs)
RATE and ODDFYIELD
Excel uses a different root finding algo. Sometimes the library results are better, sometimes Excel's. (office docs)
XIRR and XNPV
XIRR and XNPV functions are related: the net present value, given the internal rate of return, should be zero. However, XNPV works only for positive rates even though the XIRR results might be negative. The results can also be different because of the root finding functions. (office docs)
1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23: 
