There was a group in Session B who noticed that the degrees of freedom in the binary model was not adding up. Or rather adding up to being too high.
This has been tracked down (thank you Mike Hunter) to an issue in C++ that is counting thresholds as observed statistics and this is being followed up.
The df should be equal to the number of observed statistics - the number of estimated parameters. And in the binary model the constraint added an observed statistic, so the df ought to be 3995 for the twins-only model (4001-6=3995) and it should be 5995 for the twins&sibs model (6001-6=5995).
Thanks for noticing this and asking about it