In memetic colonies, multiple belief systems are pressured to exhibit internal coherence and ruthlessly undermine each other. Over time, the competing memeplexes exhibit drift. However, real-world memes are not constrained to an hermetically sealed bubble of causality, with messy psychological, sociological, economical, etc. influences constantly acting on them. Given this, it wouldn't be realistic to attempt to put together a perfect model of memetics with infinitesimal uncertainty.
That said, despite being unable to craft a perfect model, we might still be able to craft a useful model of memetics to help target our AGI. In the best case scenario, such a model would provide well-calibrated measures of uncertainty in its outputs (e.g. confidence intervals) and prove relatively on track with empirically observed dynamics (e.g. in historical data, in lab environments, etc.).
Additionally, we might agree on what makes a memetic phylogeny realistic (e.g. competitiveness), and go forward with a model which at least implements those features, despite exhibiting growing uncertainty with respect to the haphazard timeline we happen to find ourselves in. This reminds me of the contrast between existence and uniqueness proofs in maths, where the very existence of plausible memetic trajectories (e.g. ones where each tiny shift in worldview is formally plausible) being better than nothing despite not having a uniqueness proof pointing at one golden timeline in particular.
Additionally, memetic colonies relax the ambition of predicting exact memes by enabling us to study the overall compatibility of memeplexes with certain ideas (i.e. via Overton probes). There might be multiple memeplexes endorsing a belief for different reasons, lowering the bar for at least figuring out the general properties of actual memetic development.