@@ -37,7 +37,7 @@ First off, Urbit is about 10x faster for most users. With some careful debugging
38Second, we got our memory usage under control. Urbit still uses a 2GB memory arena. If the state of your Urbit grows to over 2GB things, uh, fall apart. This used to happen regularly, and now it’s pretty much unheard of. We implemented [tooling](https://github.com/urbit/urbit/pull/3235) to compact and deduplicate the memory image when it’s growing too large. We brought the memory usage of the compiler [down ~90%](https://github.com/urbit/urbit/pull/3041). There are areas where we can be more disciplined about how we use memory, but this is more or less a solved problem for the near term.
38Second, we got our memory usage under control. Urbit still uses a 2GB memory arena. If the state of your Urbit grows to over 2GB things, uh, fall apart. This used to happen regularly, and now it’s pretty much unheard of. We implemented [tooling](https://github.com/urbit/urbit/pull/3235) to compact and deduplicate the memory image when it’s growing too large. We brought the memory usage of the compiler [down ~90%](https://github.com/urbit/urbit/pull/3041). There are areas where we can be more disciplined about how we use memory, but this is more or less a solved problem for the near term.
40−Third, we implemented a completely new build system. Well, we actually got rid of the build system altogether and [made it a part of the filesystem](https://urbit.org/blog/ford-fusion/). This made it possible for us to fix our over the air (OTA) updates and make them both fast and reliable. In April, an OTA could bring down your node and would take hours or even days to apply. Today, we regularly deploy OTAs with no fear of them breaking any nodes — they often go from galaxy → planet in a minute or less.
40+Third, we implemented a completely new build system. Well, we actually [got rid of the build system](https://urbit.org/blog/ford-fusion/) altogether and [made it a part of the filesystem](https://github.com/urbit/urbit/pull/3060). This made it possible for us to fix our over the air (OTA) updates and make them both fast and reliable. In April, an OTA could bring down your node and would take hours or even days to apply. Today, we regularly deploy OTAs with no fear of them breaking any nodes — they often go from galaxy → planet in a minute or less.
42We also rewrote a large part of our runtime in Haskell and made overall progress on [network scaling](https://github.com/urbit/urbit/pull/3174). And: we implemented a [principled approach](https://github.com/urbit/urbit/pull/2366) to [error handling](https://github.com/urbit/urbit/pull/3064), making nodes much less likely to become unrecoverable.
42We also rewrote a large part of our runtime in Haskell and made overall progress on [network scaling](https://github.com/urbit/urbit/pull/3174). And: we implemented a [principled approach](https://github.com/urbit/urbit/pull/2366) to [error handling](https://github.com/urbit/urbit/pull/3064), making nodes much less likely to become unrecoverable.