A proposition to make f# docs a lot easier and testable, like in @elixirlang : http://elixir-lang.org/docs/stable/ex_unit/ExUnit.DocTest.html

current.fs

(** 
### Evaluation demo 
The following is a simple calculation: *)
let test = 40 + 2

(** We can print it as follows: *)
(*** define-output:test ***)
printf "Result is: %d" test

(** The result of the previous snippet is: *)
(*** include-output:test ***)

(** And the variable `test` has the following value: *)
(*** include-value: test ***)

new.fs

(** Sums a list of integers.

## Remarks

- Produces an `ArgumentException` for an empty list. 
- It would be better if we'd use `List.fold` with a `0`, so it doesn't throw upon an empty list.

## Examples

```F#
> let somelist = [1;2;3;4];;
> listsum somelist;;
val it : int = 10	
> listsum [];;
System.ArgumentException...
> listsum [4;5];;
val it : int = 9
```
In a codeblock:
- If the source line starts with `>`:
  - execute it and store the output lines
- If a source line doesn't start with `>`:
    - If the source line ends with `...` compare everything in front of it with the current output line
    - If it doesn't, compare the source line with the current output line
- go to the next output line & source line

*)
let listsum = List.reduce (+)

new_generated.markdown

listsum:: int list -> int

Sums a list of integers.

Remarks

• Produces an ArgumentException for an empty list.
• It would be better if we'd use List.fold with a 0, so it doesn't throw upon an empty list.

Examples

> let somelist = [1;2;3;4];;
> listsum somelist;;
val it : int = 10
> listsum [];;
System.ArgumentException...
> listsum [4;5];;
val it : int = 9

In a codeblock:

• If the source line starts with >:
  • execute it and store the output lines
• If a source line doesn't start with >:
  • If the source line ends with ... compare everything in front of it with the current output line
  • If it doesn't, compare the source line with the current output line
• go to the next output line & source line

The fsharp code tags could be easily tested like this, and not a lot of extra parsing is needed (as opposed to your code).