Calculating expected total populations based on evolutionary advantage How to set up a spreadsheet to run scenarios based on the following: 
If people A have genetics that give an $a\%$ advantage over other people $B$ of surviving to age $m$, mating age, and the average person has $p$ progeny, and the average age of death is $d$, after $g$ generations how what percentage of the population will have people $A$'s genetics?  
The idea being, let's say a certain mutation gives a person a $5\%$ better chance of surviving to mating age. If the average mating age is $15$, couples have an average of $4$ children, and everyone dies at age $50$, after $1000$ generations how much better will people with that mutation do as based on their percentage of total population.
 A: Real populations are, unfortunately, discrete, but let's slice some people in halves to make things easier. Let's just pretend those are some nuisance factors - (un-)lucky breaks, basically.
We're dealing with pairs of people multiplying, which suggests two things to me: one is that to maintain our, and our hardware's sanity, we'd better log-scale it, and that since it's pairs - let's use base-2 for that.
Now, we can model each generation separately. Let's just pretend out of the population $P$, each survivor $S$ of the attrition rate $A$ forms a single pair with the opposite-sex partner - they are conveniently always 1:1 - and together, they produce their statistical number of kids $K$. 
Bearing in mind those are all $log_2$, this can be expressed as:
$$P_{n+1} = ((1-A)*P_n) + K - 1$$
The $(-1)$ corresponds to dividing by 2 to get pairs.
This is nice classically Markovian AR(1) process if the attrition rate is deterministic, so it's almost trivial to set up in Excel.
A nice feature of logs is that if you're just interested in ratios, you don't even need to exponentiate. You can just grab some groups logarithmic population and divide by the sum of the rest to get them.
This is just the baseline though. Uncertain attrition can be handled with a Beta distribution. You could get more accurate simulations at low populations by not using logarithms and doing floor division for pairs, but it quickly explodes in size at larger ones.
Pooling the generations' populations is doable if you pretend the reproduction and death ages are fixed, it's just incredibly tedious - you'd need to calculate them separately, then histogram the overlapping periods, and in the long run it's approximated by the raw generations anyway.
