Agent Action % chance is a lie

[Martyrofsand]

So out of curiousity I decided to do some testing with agent actions. I gathered 3 agents around 1 enemy lord saved the game. I then went around from agent to agent performing their actions on the lord and took note of what happened each time.

I had 57% success chance, 40% chance of failure and 3% chance of critical failure when performing this test, and this was pretty much the same for all 3 of my agents against the lord.

For nearly an hour I would perform the agent actions reload the game and do it all over again recording the result. I expected to get a variety of results all ranging from success, failure, etc etc. Instead what I got was...

Critical Failure
Failure
Failure

In that exact order, no matter what agent perform an action first, or what order the 3 went in, none of that mattered, the first one was always a critical failure followed by 2 failures. Every single time for over an hour reloading the game and attempting the agent actions over and over again what is presented in the game as random % chance to succeed or fail actually has a very formulatic pattern to it. This doesn't seem intended, I mean the game does present agent action as % based chance, and yet if that was the case there is no way I would be met with this result, especially with only a 3% chance for critical failure.

[Matayus22]

I think the odds aren't rerolled everytime you reload. The odds stand and that's why you get the same result. This is meant to prevent people from savescumming. Any other action prior to this will reroll he odds though. Hope this helps.

[Ruinae Retroque Rursus]

Every single time for over an hour reloading the game and attempting the agent actions over and over again what is presented in the game as random % chance to succeed or fail actually has a very formulatic pattern to it

Random numbers are never truly random. Search up 'pseudo-random numbers' if you would like to know a little more. (Wikipedia's great!)

The algorithm for random numbers in TW takes into account the state of the game. If you reload the same save and do exactly the same thing, by this number-crunching algorithm you will get exactly the same number. This is called the 'seed.'

However, if you do anything different; change an item on a character, build a building, order a commandment, etc etc, that 'seed' will change and you will get a different result.

[Cratarc]

just before you load the game all your agent actions are predetermined based of that % chance, once you load it you get the same results because it is now fixed to that "saved state" to break it you have to change that saved state by equipping items razing buildings etc

[Longtoke]

Lol, I love savescummers, especially when the do this sort of “testing” .

As others above have highlighted, there an anti-savescumming sort of feature built in due to the nature of generating random numbers on a pc.

[Jerroser]

Not to sound harsher then necessery, but after repeating this "experiment" for an hour and getting the exact same result each time. Did it seriouly not occur to you that the results had been pre-calculated by the game before you initiated the agent action?

[Ruinae Retroque Rursus]

Useful note: in Medieval 2, the 'seed' is changed every time you make a new save, so making new saves and re-loading them will result in different results.

This is not the case in WH2 though.

[Wh♂♂par]

*queue in Einstein's take on being crazy*

[Elitewrecker PT]

10/10 great testing.
PS: I remember reading doing any action like queuing a building, or moving an army just a bit, alters the seed so you might get random results if you do something before the actions.

[Martyrofsand]

Originally posted by Jerroser:

Not to sound harsher then necessery, but after repeating this "experiment" for an hour and getting the exact same result each time. Did it seriouly not occur to you that the results had been pre-calculated by the game before you initiated the agent action?

Yeah this is why I was pointing out the events where not in fact random. A random event wouldn't be "pre-determined" and would instead change in every attempt.

However, if you do anything different; change an item on a character, build a building, order a commandment, etc etc, that 'seed' will change and you will get a different result.

Actually tested this a couple of the times as well, IE went around my bases and built different buildings, trained new troops and changes weapons on different generals. Still ended up with the same result on those occasions to.

[Elitewrecker PT]

Oh well, maybe those actions don't change the hero action seed then. They can make the AI do different things though, once you end the turn.

[Jerroser]

Originally posted by Martyrofsand:

Originally posted by Jerroser:

Not to sound harsher then necessery, but after repeating this "experiment" for an hour and getting the exact same result each time. Did it seriouly not occur to you that the results had been pre-calculated by the game before you initiated the agent action?

Yeah this is why I was pointing out the events where not in fact random. A random event wouldn't be "pre-determined" and would instead change in every attempt.

The point that you seem to be missing is that the calculation is made before action takes place. Essentially the way to look at it is that the game constantly runs a "what if" calculation each turn for all agent actions that could be taken. Then will only reveal it if either the player or AI decided to actually use that agent for an actions in the game.

[nimrock]

Originally posted by Cratarc:

just before you load the game all your agent actions are predetermined based of that % chance, once you load it you get the same results because it is now fixed to that "saved state" to break it you have to change that saved state by equipping items razing buildings etc

I thought it was exactly like that.
However courple days ago I succesfully managed to slow the enemy army movement. Then due to very high loses at autoresolve I choose to reload the start of my turn and this time I failed with my agent (same action, sam % chance).
Still I never could get positive result retrying failed action.

[Mactalon]

Originally posted by Jerroser:

Originally posted by Martyrofsand:

Yeah this is why I was pointing out the events where not in fact random. A random event wouldn't be "pre-determined" and would instead change in every attempt.

The point that you seem to be missing is that the calculation is made before action takes place. Essentially the way to look at it is that the game constantly runs a "what if" calculation each turn for all agent actions that could be taken. Then will only reveal it if either the player or AI decided to actually use that agent for an actions in the game.

^ +1

That 40% or 60% or whatever chance still remains true, but it's just done that roll at a different time than what you expect. When you have the level 30+ heroes that have 90% chance rolls you'll still see them get successes with very rare failures.

The % chance isn't a lie, it just doesn't work the way you think it does.

[Wh♂♂par]

Originally posted by nimrock:

Originally posted by Cratarc:

just before you load the game all your agent actions are predetermined based of that % chance, once you load it you get the same results because it is now fixed to that "saved state" to break it you have to change that saved state by equipping items razing buildings etc

I thought it was exactly like that.
However courple days ago I succesfully managed to slow the enemy army movement. Then due to very high loses at autoresolve I choose to reload the start of my turn and this time I failed with my agent (same action, sam % chance).
Still I never could get positive result retrying failed action.

If you did just one action differently - other agent's action first, etc - it changes the seed calculation.

[HEXIIN]

It's really not that complex.