It seems fairly similar to LL's autopilot as far as what it takes into account. I think my inherent dislike of these is the notion an algo can solve it. Allocating to channels, atleast initially, is more of a fuzzy logic. Attempts to automate will result in centralization and suboptimal routing, and the lack of traction any of these tools have is a good indication they don't work at scale.

Swaps to rebalance are an obvious and simple enough solution for some conditions, but over-reliance on them presents its own problems and costs. If you're going to the chain often you're undermining the efficiency of the second layer.

This comes back to my thesis that Lightning ideally should never have had its own gossip network, and if Nostr had existed when Lightning was created it'd have been a better medium for peer discovery based on actual personalized economic activity and webs of trust.