Not exactly, due to being able to input halfway or right on a poll. It's not exactly cut and dry raw frames because there's also software latency to handle the "reload" button being multiple purpose, which codepath would be re-used for melee vs assassination. For instance, if you started a player reload right when the button is pressed (shortest latency), you would reload before you activate a button or get in a vehicle or picked up an objective, which is what players usually do not want, so you have to wait a moment to see if they held. For assassinations, this means detecting the melee input, then waiting a moment to see if they wanted to just smack or assassinate.
let's look at a crude representation of controller button samples, assuming a straight doubling due to a 30->60fps change. The only polls that detect a button press are where the player held it when the poll started.
So Reach/4 has to make a decision - a long press in Reach/4 can't be detected as
unpressed until later on, so you try to read the player intent and assume after a minimum amount of time the player intends to keep holding and have the client send "HOST, I WANT TO DO THE FANCY KILL". 5 knows it'll be unpressed soon, so the client goes "I already see it unpressed, so just do a normal back whack". Also, latency and the host vetoing an assassination input and turning it into a normal backwhack (Reach/4 do it when you get into certain situations, I assume 5 has similar logic) also come into play.
tl;dr
halo 5 polls faster, resulting in what the engine in 4/Reach recognized as a long press as being one frame too short to count in 5, because reusing a button requires software prediction and assumptions.
tl;dr;pt2
I reduced the frames polled just to make it easier to parse/read for this example. I don't know actually how many frames they sample,
ps.
this also assumes they didn't tweak the give time for halo 5, which I assume they did, exactly because people felt it was too easy to do in Reach/4 by accident.
pss
just to cover all my bases, the controller also likely ignores a too-short input (player brushing the button by accident, ala Anti-Clench that's in Reach/4/5) meaning the extra poll it sees at the start doesn't really get 'read'.