Summary
An engineering post-mortem dressed as an anniversary piece: three years of SimpleSwap routing KAS, and what that taught about the layer nobody puts on a landing page.
KAS was never the biggest asset by volume — what it surfaced was a decade-old assumption that no longer held.
The framing is aimed at anyone embedding a swap in their own product.
A user taps the button in your app; if the estimate is off or the funds arrive late, the complaint lands on your support team, not the provider's.
That asymmetry, the piece argues, defines the category: the question is not which provider lists the most assets, but what job you are hiring one to do.
Four demands follow from it. Keep the promise the interface made on the provider's behalf. Do not leave the integrator to explain a stalled swap.
Be there on the worst day, because any provider looks competent on a calm one.
And do not let integration become a permanent project — if every new asset becomes a roadmap ticket, you bought an obligation rather than infrastructure.
The KAS case is structural. An estimate is assembled from live quotes across venues that can fill the trade, weighted by the depth behind each at that moment, and it has a shelf life.
When simpleswap.io began routing it in 2023, Kaspa produced about one block per second; the Crescendo hard fork in May 2025 raised that to ten.
The failure was not the network's.
What broke was the inherited assumption that the chain would always be the slow part — as the article puts it, fast chains did not remove latency, they pushed it into a layer nobody had measured because there had never been a reason to.
The portable lesson closes it: an estimate is not a prediction about the market but a commitment with an expiry, and a system that does not treat it that way will keep making promises it cannot keep.
“Fast chains did not get rid of latency. They just pushed it into a layer no one had ever measured, because there had never been a reason to measure it.”
Key Takeaways
- A late or mispriced swap becomes the integrator's support problem, not the provider's.
- Four demands of an execution layer: keep the promise, own the explanation, hold up on the worst day, stay out of the roadmap.
- An estimate is built from live quotes weighted by depth, and it expires.
- Kaspa went from about one block per second in 2023 to ten after the Crescendo hard fork.
- Faster chains relocated latency into an unmeasured layer rather than removing it.
- The closing principle: an estimate is a commitment with an expiry, not a forecast.






