How much does the AI Change between Easy and Legendary?

[RedKnight]

Good tips, fbe. It takes a while to figure out all the factors affecting population happiness Nightwindking, and occasionally the displays are hard to read even for a veteran, but there is a lot of info there. If you didn't know about the spread of religion, now you know to keep a much better eye on it. Christianity is a really volatile thing in this game, and mastering it (whether playing Buddhist or Christian) is a lot of fun.

I'm sure there are many guides to agents to be found if you search. Some tips are:

1. Monks and Priests are fairly easily level-able by siccing them on armies (demoralize). The smaller the stack, the easier to demoralize and less chance of failure (which often equals death for a beginner priest).
2. Likewise, ninja sabotaging armies is an easy choice. Again, smaller armies are better (cheaper and more chance of success). Sabotaging buildings is also a choice; the lower the level of the building (less improved), the less the cost and higher the chance of success. Just click on all the choices you have in range and choose the easiest and/or cheapest, when training your ninja.
3. I find metsuke the hardest to train because they mainly need to confront other agents... and unlike armies and buildings, enemy agents can be hard to predict. Often enough, but hardly always, AIs generate them at their capitols.

For the record, your agents get 15 points for a successful action and 3 for a failed one. Also, 1 point/turn for a "maintenance" action (like a metsuke being embedded in a town to increase its taxes).

Also for the record, I don't believe the percentages shown in game are always right, particularly at the lower ends (below 50% chance). Above 80%, they seem to be on target. Also doing something like trying to assassinate a daimyo seems to be really overstated; you probably only have a third of the chance shown.

Something that may not be obvious is that any failed action against an AI also incurs a diplomacy hit, which you can see if you mouse over the other AI in the diplomacy screen to see the reasons for your current score against them. If you want to worry about this, you can choose to harrass rebels and/or only war enemies... but the point where AIs decide to war with you seems to be complex, so I'm not yet sure how much it matters. For one thing, they seem to declare war first if someone is adjacent to them, and only secondly based on warscore if they have no one to declare war on, next to them.

Sometimes early in the game I will have a ton of young monks and ninjas around one hapless one-unit stack, which can't move and is incredibly demoralized. Poor bastards, ha ha.

[frogbeastegg]

Also for the record, I don't believe the percentages shown in game are always right, particularly at the lower ends (below 50% chance). Above 80%, they seem to be on target. Also doing something like trying to assassinate a daimyo seems to be really overstated; you probably only have a third of the chance shown.

:twitches: The success chances were one of my biggest banes when researching the guide. :twitches: I'm permanently scarred. :twitches: This nervous tick won't go away, every time someone mentions success chances :twitches: it starts. What should have taken ten hours took tens of hours because fail, fail, fail, fail, fail, fail, fail - er, sorry. As I said, scars. :twitches:

I found that there are a pair of notable, repeatable exceptions to the '80% or higher equals mostly successful' rule: missionaries causing rebellion and metsuke bribing. The missionaries' rebellion mission in particular seems to work out as 80+% reported success equals 0.5/10 succeed, 1.5/10 to be executed, and 8/10 to fail. Must have taken me in the region of 11 hours alone to get sufficient successes out of them to report back on the effects of that single mission, and I still loathe the mere mention of the agent type some 3 months later as a result! Metsuke's bribery at 80+% is more like 2/10 succeed, 7.5/10 to fail, 0.5/10 to be executed. For those two missions it does not matter what level the agent is, what skills you give them, or what target you send them after. 80% success = FAIL!1! became my frustrated mantra :twitches:

With other all missions (and agents), I found the reported chances were quite accurate. On a very rough level, 80% means 2 out of every 10 attempts will fail and the rest succeed. This is why the language in my agents section is passively biased towards ninja (e.g using their mission names as examples in the text). They work as advertised and I love them for it.

The most baffling thing is that there are people out there who do not experience the 80% = FAIL effect with causing rebellion and bribing, and instead have it succeed reliably time after time. I copied their methods, their skill setups, their difficulty levels, and I couldn't replicate it.

With the lower percentages, like 15%, now the lower difficulties I think there might be an added element of luck added on top of the displayed calculation. It's as if the AI rolls a second dice to see if it can override the result in your favour. On the highest difficulties I very rarely succeeded with any agent on a low success chance, whereas on normal and easy success with a low chance happened more often. This supports a luck theory, as the game applies more slight of hand to make you successful on the lowest difficulties. Of course, 15% is still sufficiently meaningful percentage that success should happen naturally now and then, even on very hard.

[RedKnight]

:twitches: The success chances were one of my biggest banes when researching the guide. :twitches: I'm permanently scarred. :twitches: This nervous tick won't go away, every time someone mentions success chances :twitches: it starts. What should have taken ten hours took tens of hours because fail, fail, fail, fail, fail, fail, fail - er, sorry. As I said, scars. :twitches:

I think the game holds on to the random number; I've worked out a way to work around it if you think it's too ... conservative. Reloading the game does not work.

[frogbeastegg]

I think the game holds on to the random number; I've worked out a way to work around it if you think it's too ... conservative. Reloading the game does not work.

Reloading does work; I've seen it happen many times during my testing. Sometimes you get a different result on your first reload, others it takes many goes. So many attempts that you might be convinced it won't work. You don't need to give any other orders, move armies about etc to get the game to spit out a different result; I found it didn't make any appreciable difference. The process I used was quick save, send agent on mission, quick load, send agent on mission, repeat until success or 10 failures, in which case advance to the next agent/turn.

Reloading also changes failure/critical failure. When testing via reloads, sometimes agents would be executed where they'd repeatedly failed and escaped in the previous reloads.

I wouldn't recommend reloading for normal play, though. It's cheesy, and tedious. Those determined to reload should note that if you don't get a different result within the first 5 reloads you will probably be there a very long time before the game spits out a success. I wondered about including this information in the guide, and decided against.

I also did agent tests without reloading at all. It didn't help with the 80%=FAIL thing.