Hey @JosephKnecht, thanks a lot for the stimulating conversation over the WE 
I agree the immunity to mean reversion (winner-loser effect) is an important point that needs to be clarified based on empirical, out-of-sample data which I have endeavoured to do since Sunday.
Unfortunately, while I was extracting the āaverageā and ālosersā portfolio performance from the backtest covering May to July '21, I discovered an issue with my price feed API whereby random bits of price history would be missing across the 800-ish tokens currently monitored.
The methodology behind iRobot relies on characterizing historical returns against downside volatility to capture early upside potential or exit of accumulation phases : complete datasets are required to be relevant / unbiased.
My priority before going into DG2 is therefore to understand the effect that this issue has got on iRobotās composition as well as on backtesting performance.
Concretely, this means comparing my actual datasets with the results coming from another API, which is a fair bit of work : I will continue updating you and the community on my findings.
As shared in our chat, progressing iRobot without ensuring 100% transparency and integrity is the last thing I want to do.
On the bright side, testing another API is something I intended to do before launch anyway since my current one is also starting to be a limiting factor in terms of available price feeds.
Iām happy to have caught this issue now, and Iām sure we will look at this step back positively once the 1st decentralized Robo-Advisor is finally out on the market 