diff --git a/README.md b/README.md
index 6b198775f..a1086312d 100644
--- a/README.md
+++ b/README.md
@@ -34,14 +34,87 @@ type Y = Exp<N2, P3>;
 assert_eq!(<Y as Integer>::to_i32(), -8);
 ```
 
-For a non-trivial example of its use, see one of the crates that depends on
-it. The full list is
-[here](https://crates.io/crates/typenum/reverse_dependencies). Of note are
+
+And how to use constraints:
+
+```rust
+use typenum::{self, IsGreater, True, U0};
+trait NonZero {
+    type NonZero: IsGreater<U0, Output = True>;
+}
+```
+<details>
+
+  <summary>Unfold here to see a (somwhat contrived) exploitation of type-level integer arithmetic:</summary>
+  
+  ```rust
+  // Demo-time. A simple "flatten a 2d bool-array to a 1d byte-array"
+  // Let's encapsulate a 2d array, similar to `[[bool; WIDTH]; HEIGHT]`
+  pub struct FlattenDemo<
+      Width: ArrayLength,
+      Height: ArrayLength,
+  > {
+      unflattened: GenericArray<GenericArray<bool, Width >, Height>,
+  }
+  
+  // For fun and profit, let's wrap-up the behavior in a trait.
+  trait Flatten {
+      // NOTE: the `generic-array` crate is pretty cool.
+      type FlattenedLen: ArrayLength;
+      fn flattened(self) -> GenericArray<bool, Self::FlattenedLen>;
+  }
+  
+  
+  // So here is the fun part: Flattening using compile-time maths, but using the type-system.
+  // Flattening into a byte-array, you must ensure a multiple of 8, to Illustrate:
+  // A: 2 x 3 = 6. so need a 1-byte array
+  // B: 2 x 4 = 8: 8 is the nearest round-up, again; 1-byte array
+  // C: 3 x 7 = 21: need to round up to 24: 3-byte array
+  //
+  // This is done by taking the needed bits, adding 7, then integer-division to 8
+  // A: (6 + 7) / 8 = 13 / 8 = 1
+  // B: (8 + 7) / 8 = 15 / 8 = 1
+  // A: (21 + 7) / 8 = 28 / 8 = 3
+  impl<Width: ArrayLength, Height: ArrayLength> Flatten for FlattenDemo<Width, Height> 
+  where
+      // Types are not values: Unlike integer values, all of which implement the behavior of
+      // integers (multiplication, addition, division, etc) as an inherent part of the language,
+      // the compiler has no way of knowing if a given type implements a given operation unless 
+      // you explicitly specify...
+      // 
+      // 1. We must specify that the `Width` type must implement behavior in which it `Mul`tiplies
+      //    the `Height` type:
+      Width: Mul<Height>,
+      // 2. We also must specify that this `Mul`tiply behaviors `Output` implements the 
+      //    "`Add`ition on `U7`" behavior:
+      op!(Width * Height): Add<U7>,
+      // 3. And so on: This is the constraint "The result of (width * height) + 7 must implement
+      //    division by 8", but without the `op!` convenience macro:
+      <<Width as Mul<Height>>::Output as Add<U7>>::Output: Div<U8>,
+      // With the convenience macro:
+      // op!((Width * Height) + U7): Div<U8>,
+      op!((Width * Height + U7) / U8): ArrayLength,
+  
+  {
+      // Et, voila! Through the power of the type system, we have lifted arithmetic to compile 
+      // time, without the use of nightly, the need to think about machine-representation of 
+      // values (usize/u64/etc).
+      type FlattenedLen = op!((Width * Height + U7) / U8);
+      fn flattened(self) -> GenericArray<bool, Self::FlattenedLen> {
+          todo!()
+      }
+  }
+  ```
+</details>
+
+For more non-trivial examples, with real-world use, check the full list of
+reverse dependencies [here](https://crates.io/crates/typenum/reverse_dependencies). Of note are
 [dimensioned](https://crates.io/crates/dimensioned/) which does compile-time
 type checking for arbitrary unit systems and
 [generic-array](https://crates.io/crates/generic-array/) which provides arrays
 whose length you can generically refer to.
 
+
 ### Error messages