mirror of
https://github.com/lightningnetwork/lnd.git
synced 2026-08-18 13:07:58 +02:00
83f99803ae
In this commit, we close the last hole in the exp-002 claim. That writeup concluded parameter tuning cannot beat lnd's defaults, but it never evaluated the configuration that would make lnd's own machinery match this environment: its bimodal estimator with scale_msat near five percent of a typical channel, which is what our liquidity generator draws. The sweep tool learned to pass a params file, and seven scales bracketing the matched value ran against lnd's shipping apriori default and the champion. The claim survives on both sealed tiers, and the way bimodal fails is the useful part. It raises success while more than doubling attempts, because a better liquidity prior makes lnd more willing to keep trying without changing what it retries: findPath takes the amount as a fixed argument, so any estimator still retries the same amount over a different route, where the champions read their upper-fail bound and retry a different amount. The estimator is worth two hundredths of objective and the paradigm is worth two tenths. |
||
|---|---|---|
| .. | ||
| experiments | ||
| DECISIONS.md | ||
| IDEAS.md | ||
| NOTEBOOK.md | ||
| WHY.md | ||