To expand on a thought that I've just posted in the GRoDT hub thread:
Originally posted by tonypmm
I tell in advance which expensive games I exclude from the accounting, they're only a small part of my activity anyway. The idea is similar to the concept of a solid defence line in Tankroll BRM: I have challenge games that I'm testing for reliable beatability for a living in the long run, which allows me to look into the future with optimism and venture into some non-challenge games from time to time that I probably don't beat.
While I dismiss the Tankroll itself because almost no one (except those crushers who would be playing higher than NL100 anyway) can beat NL100 at 10+ bb/100 post-rakeback (otherwise moving up with 10 BIs is suboptimal), I suspect that the general construction of a solid 'defence line' (the stake that one can surely beat) plus shots at higher stakes might work.
However, I think the shots should be limited in time regardless of how well things go in monetary terms. The only purpose that they should serve is measurement of one's winrate when it's not very predictable (e.g. in the case of switching sites and thus fields, like from PKR's $12s to Stars' $30s).
Here, the fact that jackpot-type SnGs have a measure of skill (the chip EV) that converges way faster than the actual winnings comes in handy because it means that it suffices to risk as smaller portion of the BR for the purpose of winrate measurement than it's necessary in games where all-in races make up for only a minority of the variance, like 100 bb NLHE cash.
So, though I'm not done yet even with the non-Bayesian part of the BRM research (i.e. for the case when the winrate at each stake is known or can be predicted), I might want to incorporate Bayesian inference (calculating the winrate prediction basing on both an a priori assumption and the actual results, the latter mattering more as the sample size grows). This model is known as 'optimal betting under uncertainty', and it was applied in 'The Mathematics of Poker', which showed that the incomplete info on the winrate makes the BRM guidelines much tighter (I'm not sure by how much in the jackpot case).
In particular, the potential EV loss from overestimation of the ROI at the 50s-60s can be way bigger than the missed gain from underestimating it and not playing the 50s-60s instead of the 10s-30s of various sites.