Cost, speed, failure· 7 min read
An argument for measuring the slowest 1% (P99) instead of the average. A recipient who has been waiting three days does not care what the average was that week, and averages make a system look better than it feels.
The summary above is the current state of it. The full post goes up once the work behind it is finished and reviewed, which is the same rule we apply to everything else on the site.
Back to all postsTail Latency P99 · Finality Metrics
The $200 everyone divides by is a fixed benchmark, not a measured average: the real US to Mexico average is $488. Working the throughput question properly, across four demand levels and an honest market share, lands nearer 14 to 33 transactions per second than the headline 143.
· 9 min read
Adding up every cost layer: gas, bridge fees, the cost of locking up liquidity in a channel, currency conversion, and the fee to turn it into cash locally. The last one gets left out of most comparisons.
9 min read · summary only
We design failure experiments for each setup, taking nodes offline and slowing bridges down on purpose, then look at whether transfers roll back or funds get stuck somewhere in the middle.
10 min read · summary only