For a fun deceleration we simply reverse the order of adding frames and evaluating.

So:

fun f x = x

is drawn as

step 1: [[]]

step 2: [[]]<--[f, ]

step 3: [[]]<--[f, ]

^

|

+--oo

p: x

b: f x

+------+

| |

step 4: [[]]<--[f, +] /

^ |

| V

+-----oo

p: x

b: f x

Note that we can't great this closed loop with the purely functional subset

of ML. At this point we need to either be dealing with refs, or simply

discussing the environment model as a formal model. I suggest the latter.