This answer comes very late - sorry for not seeing it before - anyway it might be helpful to someone dealing with a similar problem.

This is the kind of issues that multi-event models handle very easy and well.

You may define 3 states (Fem_Alive, Mal_Alive, Dead), 4 events (0= not capt, 1= capt as female, 2= capt as male, 3= capt as unknown sex) and, if you have some individuals whose sex is known from the very first time they are captured, 3 groups (females at 1st capt, males at 1st capt, unknown at 1st capt). Groups may help to get better precision because they provide a kind of ancillary information (more on this below).

Then, in GEPAT, that is where you define the probability matrices, you set something like

thisIf you have been able to define the groups you may use a sentence like this in GEMACO for the Initial State: "g.to" which is telling ESURGE to estimate an Initial State probability (in this case p of being male when first captured) for each group. Then, in IVFV, that's where you are allowed to fix parameters to a certain value, you fix g(1) (i.e. females) to 0, g(2) (i.e. males) to 1 and let it calculating the initial state probability of being male for the g(3).

You may have several different situations similar to this that is where ESURGE gives you a plus to modeling CMR data.

A good and illustrative paper to be read for that is:

Genovart, M., Pradel, R., & Oro, D. (2012). Exploiting uncertain ecological fieldwork data with multi‐event capture–recapture modelling: an example with bird sex assignment. Journal of animal ecology, 81(5), 970-977.