Hmm, Is drawdown expressed in percent? My hunch is the first backtest shows a drawdown of -300%. That implies that one not only lost all their money at one time but then 2x more than that.

In general a very good single number to identify a 'best' algo would be the Sharpe ratio. The bigger the better. Google 'Sharpe ratio' and you should get some good background, as well as tangents leading to other methods, to rate portfolio performance.

However, one needs to ensure any algo passes some minimum criteria. Drawdown is one. One should eliminate any with drawdowns over 100% (those lost all their money and unless you are Warren Buffet you probably can't borrow any more). Realistically, one should eliminate any with drawdowns over 35% or so. Those are just too scary of a ride for most mortals.

I look at volatility too. Much like drawdown it's a measure of 'scary'. Volatility over 30% is generally a bit too rough a ride for most realistic strategies and should be eliminated.

Also, it isn't stated what leverage is being used. One should generally eliminate any algos with leverage over 1. This can be raised later but 1 is a good starting point for comparison. Realistic retail investor strategies should never be over 2 (that's all brokers will ever lend). Institutional investors can go higher to 3 or 4.

So, if the minimum drawdown really is -300%, then none of the order sizes yield a practical algorithm.

Not sure where 300% came from, that is 3% drawndown. I have updated the original post to help make it more clear.

I do definitely have to learn more about volatility, because the "scary" factor is indeed important!

My mistake for assuming 3 was 300%. So, without any further info, I'd stick with the the highest Sharpe ratio. That is one of the runs with a 3.25 ratio. Not sure it's coincidence, but those all have a reasonable drawdown of 6% or less. If the drawdown got much above 10-15% I'd be wary.

With the new backtest screen, create a notebook and run a 'tear_sheet'. This provides a wealth of other info to compare the backtests.

Good luck!