Open
Description
Currently we only ever update the lower-bound on a channel's liquidity estimate if we manage to pay across the channel but then fail later on (or are probing). Because we assign a lower failure probability to a channel with a lower bound compared to one without, this causes us to prefer reusing path prefixes which we have failed across before, just replacing the end of the path prefix. I'm not sure but I fear this may cause us to strongly prefer to scan "depth-first" and not take paths that may be better via other prefixes. But maybe that's okay because we're super confident in their prefixes?
Unclear if a bug but just noting.