diff --git a/README.md b/README.md
index 8879e1716..68596f18c 100644
--- a/README.md
+++ b/README.md
@@ -165,7 +165,7 @@ Everything:
 - The concept itself is all about eliminating any unnecessary computation. Think `msbuild`, but for your method call results: what's computed and consistent is never recomputed.
 - Fusion caches call results in memory, so if it's a hit, they're instantly available. No round-trips to external caches, no serialization/deserialization, etc.
 - Moreover, there is also no cloning: what's cached is the .NET object or struct returned from a call, so any call result is "shared". It's way more CPU cache-friendly than e.g. deserializing a new copy on any hit.
-- Fusion uses its own `Stl.Interception` library to intercept method calls, and although there is no benchmark yet, these are the fastest call interceptors available on .NET - they're marginally faster than e.g. the ones provided by [Castle.DynamicProxy](http://www.castleproject.org/projects/dynamicproxy/). They don't box call arguments and require just 1 allocation per call in average.
+- Fusion uses its own `Stl.Interception` library to intercept method calls, and although there is no benchmark yet, these are the fastest call interceptors available on .NET - they're marginally faster than e.g. the ones provided by [Castle.DynamicProxy](http://www.castleproject.org/projects/dynamicproxy/). They don't box call arguments and require just 1 allocation per call.
 - The same is true about `Stl.Rpc` - a part of Fusion responsible for its RPC calls. Its [preliminary benchmark results](https://servicetitan.github.io/Stl.Fusion.Samples/rpc-benchmark) show it is ~ **1.5x faster than SignalR**, and ~ **3x faster than gRPC**.
 - `Stl.Rpc` uses the fastest serializers available on .NET – [MemoryPack](https://github.com/Cysharp/MemoryPack) by default (it doesn't require IL Emit in runtime), though you can also use [MessagePack](https://github.com/MessagePack-CSharp/MessagePack-CSharp) (it's slightly faster, but requires IL Emit) or anything else you prefer.
 - All critical execution paths in Fusion are heavily optimized. [Archived version of this page](https://web.archive.org/web/20201212144353/https://github.com/servicetitan/Stl.Fusion) shows the performance on above test currently 3x better than it was 2 years ago.