AI War Beta 7.025-7.028 "Extermination Protocol MkI" Released!

[keith.lamothe]

(Crossposted from this blog post)

This one focuses on polishing recently-introduced rough edges (mainly in terms of balance) and in general is the first step in a new effort to fix The Bug (10/10 being winnable).  We don't want that setting to be an immediate "you die in the first 30 minutes, no questions asked" button, but things have clearly gotten out of hand.

Some other highlights:

- Hacking now has an extra source of suspense and challenge: once the hacking _stops_ there will be a rapid short-term post-hacking response from the AI proportional to how much you ticked it off during the hack.  If you're adequately prepared this won't be too big a problem (though I hope it will add to the excitement).  But if you've dug yourself too deep a hole... well, have fun with that ;)

- Carriers can no longer fly through your forcefields.  They've come a long way from the super-cheesy "literally invincible, ignores forcefields completely" incarnation of old, and are now basically just a hyper-concentrated ball of death that can attack and be attacked in a very similar fashion as... well, a big pile of ships.  Their attack is now more similar in potency to the ships inside, and is longer ranged (otherwise the larger forcefields and/or radar jamming really mess with them) but in general they won't just "cheat" their way through your defenses anymore.  They'll just hammer them flat, unless you're adequately prepared.

- The alt-progress for champs no longer asks you to walk into the lion's den of pain just to make all the toys available.  Previously you had to go to intensity 9, and if you followed the lobby's (bad, then) advice of setting a matching alt-nemesis intensity it was... well, varying from "interesting" to unwinnable.

- Salvage is now much less capable of making the Metal mechanic irrelevant for significant chunks of time (mainly this was a problem in superweapon games).

Update: 7.026 hotfix for several severe balance issues with the new hacking post-responses, and a few other fixes and changes that snuck in since 7.025.

Update: 7.027 just to do a quick balance iteration on the champion nemesis exos.

Update: 7.028 is another balance iteration on the champion nemesis exo stuff, along with a few other small changes from feedback during the week.

Enjoy!

This is a standard update that you can download through the  in-game updater itself, if you already have 4.000 or later. When you launch the game, you'll see the notice of the update having been found if you're connected to the Internet at the time.  If you don't have 4.000 or later, you can download that here.

[Toranth]

Conveniently, I had a save from a little while ago, where I was just about to start a SuperTerminal hack.  So, I upgraded, tossed in that save, and repeated the 60 AIP reduction ride I'd done before.

I noticed immediately that the spawns were significantly larger, due to the accounting for AI Difficulty.  Ok, good.  It really was too easy before.
Then, when I hit 60 AIP down (net 40 reduction) and had spent 95 HaP (35 net), leaving me with 90-ish left, I killed the SuperTerminal.
Instantly, over 3000 Mk II ships spawn - 68,000 base strength.  A surprise, all right, and they threatened my defenses a bit.
I was surprised.  All 60 ticks of the SuperTerminal previously had only accounted for maybe 2000 ships combined.  So I reloaded the save, and tried again. 

An 80 AIP reduction (60 net) cost 162 HaP (net 82), so I still had about 30 HaP left.  This resulted in 16,000 Mk II ships - 150,000 strength.  This overwhelmed my defenses.  My fleet plus 8 Battleship Champions plus Core turrets and Mod fort could only kill about 10,000 of the units before being wiped out.  The remaining 6000 also wiped out a few nearby outposts before eventually stalling at my Homeworld.

So I tried 100 AIP reduction.  This costs a net of 164 HaP (264 raw), which put me slightly negative.  About -50 HaP or so.  This produced an amazing 46,000 Mk IV ships - 1.3 million strength.  Obviously, I did not survive.


The problem I see is that all of the charged-up pulses expend at the final response level, as if the player were repeating the entire hack again, but this time at the new, much higher AI response.
This same thing happens when you do any other hack, such as a 15 minute (900 second!) Advanced Production hack.  Even worse, in those cases, the 'vigorous hacking response' happens even if you fail.  A successful hack's response produces 5-10x the number of units the actual hack did.

While the idea of an AI 'chaser' to kick off the end of your hack certainly has merit, I think this is too much.  The ST is now pretty much hard-capped at 80 or so, if you do a LOT of prep, or have a lot of warheads prepared.  Going negative on any of the facility hacks is nearly suicidal.  Encouraging players not to go negative is one thing, but if you consider it that much of a problem, I'd rather see a hard cap:  'You cannot start a hack unless you have enough HaP to finish it'.

I think there should be a few changes to hacking to increase the difficulty, but severely tone down this 'chaser'.  Here are my three suggestions (and one request).
First, currently the AI's Hacking Response = Max(Current_AIP - Current_HaP, 10).  Changing this to Max(HaP_Used, 10)  both makes more sense, produces a continually rising response, and is almost equivalent:  It's just Total_AIP - Current_HaP, after all.
Second, change the 'chaser' to be more like the Exo-waves that result when you kill a Core Guardpost.  Base the Exo budget off of a small, fixed number of response pulses: 5, say, or 10.
Third, when spawning Zombies as response, give them a leader to follow - or at least, orders to all attack a single target.  Part of the reason that Zombie spawns are not so threatening is that they throw themselves onto your defenses in ones and twos.  If they were throwing themselves onto your defenses in groups of a hundred, they'd be a lot more noticeable.

Finally, related, a side issue:  The AI can still spawn non-reclaimable ships as part of the hacking response, leading to huge groups of Maws or Zenith Medics joining the threatfleet as normal units while the Fighters all go off and get killed as zombies.


I like the new, reduced AI Champion Alt-Response so far.  I'll finish my current game to be sure, but it seems a lot more workable than before.

[keith.lamothe]

Thanks for the rapid and thorough feedback, as always.

So I tried 100 AIP reduction.  This costs a net of 164 HaP (264 raw), which put me slightly negative.  About -50 HaP or so.  This produced an amazing 46,000 Mk IV ships - 1.3 million strength.  Obviously, I did not survive.

Mission Accomplished. ... oh, all right, we'll change something

The problem I see is that all of the charged-up pulses expend at the final response level, as if the player were repeating the entire hack again, but this time at the new, much higher AI response.

That much I think is ok.  Partly people can learn to fear the consequences of such hacks, or at least understand what they can and cannot handle.  Though perhaps it doesn't need to do it at a 1:1 ratio of post-pulses to pre-pulses.

But for sanity's sake it shouldn't apply the negative-HaP penalty to the post-response spawns.  Otherwise, as you pointed out, it's basically a "die now" button.

This same thing happens when you do any other hack, such as a 15 minute (900 second!) Advanced Production hack.  Even worse, in those cases, the 'vigorous hacking response' happens even if you fail.  A successful hack's response produces 5-10x the number of units the actual hack did.

I think the main problem there is it's still (iirc) doing the wild-rolls on the post-response spawns.  That was intentional, but in retrospect not wise.

First, currently the AI's Hacking Response = Max(Current_AIP - Current_HaP, 10).  Changing this to Max(HaP_Used, 10)  both makes more sense, produces a continually rising response, and is almost equivalent:  It's just Total_AIP - Current_HaP, after all.

I agree that it makes more sense to have the response's base value be monotonically increasing.

Second, change the 'chaser' to be more like the Exo-waves that result when you kill a Core Guardpost.  Base the Exo budget off of a small, fixed number of response pulses: 5, say, or 10.

Hmm, for the superterminal I really do want the post-response to be proportional to how long of a ride it was.  It doesn't need to be 1 pulse per previous pulse but a 100 AIP ride should be much worse than 40, even aside from the difference in HaP spent at that point.

And on the other hacks I also want the response to be proportional to the severity of the hack, which is largely implied in its duration.  So perhaps the advanced constructor hacks shouldn't be 900 seconds, or perhaps the overall thing doesn't need to "charge" one pulse per 20 seconds, but that advanced constructor hack should have a much stronger post-response than a knowledge hack.

Third, when spawning Zombies as response, give them a leader to follow - or at least, orders to all attack a single target.  Part of the reason that Zombie spawns are not so threatening is that they throw themselves onto your defenses in ones and twos.  If they were throwing themselves onto your defenses in groups of a hundred, they'd be a lot more noticeable.

Hmm, the zombie-unit AI is... well, not much of an AI.  Zombie, that sort of thing.  It doesn't take orders from the normal AI and so on.

But really those responses were made zombies to avoid reclamation exploits, and ultimately I think they would be better as normal AI units so they could be more challenging as you point out.

Tricky, that.

Finally, related, a side issue:  The AI can still spawn non-reclaimable ships as part of the hacking response, leading to huge groups of Maws or Zenith Medics joining the threatfleet as normal units while the Fighters all go off and get killed as zombies.

Yea, if it couldn't spawn those the responses would be considerably less threatening (it still pays the strength costs, but the variety and the heavier units still help).

I like the new, reduced AI Champion Alt-Response so far.  I'll finish my current game to be sure, but it seems a lot more workable than before.

Glad that worked out


Anyway, on hacking, what I'm currently thinking is:

1) Make base hacking response value come from total HaP used.

2) Make the post-response spawns never apply the negative-HaP penalty, and not act as if HaP is lower than zero (which I guess is tricky in light of the first point, but basically deducting -current_hap if current_hap < 0).
- the latter part being to avoid situations where simul-hacking 10 fabs results in -6000 hap and a "you die immediately" post-response, even with the negative-hap penalty disabled for those spawns.

3) Have the non-superterminal hack post-response spawns not do wild rolls (the superterminal one already doesn't).  They might need further reduction but this would be a pretty major change I'd like to see the results of first.

4) Reduce the superterminal post-response spawn strength by maybe 20%.

5) Have hacking response spawns in general be normal units instead of zombie again, to see if the overall effect is better.  If related exploits are truly a problem then there are some other things I could try (pulling zombies into normal AI would be one thing though it would undermine the meaning of zombie, or maybe just giving the hacking-spawned units an independent no-reclamation flag, or something like that)


Further thoughts?

[Lancefighter]

Should conducting hacking attempts at -6k hap not cause a loss?

As for the superterminal, I would personally prefer that it caused events that were actually scary on each tick, instead of a major tick at the end.

[keith.lamothe]

Should conducting hacking attempts at -6k hap not cause a loss?

Starting one at that level, yes.  Or even continuing one at that level.  But if your current batch of hacking ends, and that ending results in your hap going from positive to -6k, I don't want that post-response to auto-kill you in a way that the hack itself didn't.

As for the superterminal, I would personally prefer that it caused events that were actually scary on each tick, instead of a major tick at the end.

The during-ticks are scarier now that the bug is fixed where they weren't factoring difficulty.  Whether or not they're yet scary enough is a matter I'm quite open to feedback on.

[Draco18s]

As for the superterminal, I would personally prefer that it caused events that were actually scary on each tick, instead of a major tick at the end.

Agreed.  Superterminals and K-hacking should be more dangerous during and not so much about the cutoff point, as both are supposed to be along the lines of "how much pain can you handle?  Oh god oh god, turn it off."

[Winge]

First, currently the AI's Hacking Response = Max(Current_AIP - Current_HaP, 10).  Changing this to Max(HaP_Used, 10)  both makes more sense, produces a continually rising response, and is almost equivalent:  It's just Total_AIP - Current_HaP, after all.

I'm confused by this...by HaP_Used, do you mean the total HaP that the player has spent, or do you mean the HaP for that particular hack?  I assume the former, as the latter would mean that you would get the same response from hacking a Superterminal whether or not you already performed hacking activities.

I may be misunderstanding you, but I don't think that is in line with what the developers want for hacking.  IIRC, hacking at a lower AIP should incur more of a response than hacking at a higher AIP.  From gameplay, this is to prevent early hacking cheesery.  From lore, if the AI is focusing it's efforts on killing you directly, it is less likely to notice a hack (indirect action).  This cannot be determined from HaP_Used alone.  But at that point, why not keep the formula based on Current_HaP?

Am I missing something here?

[keith.lamothe]

I may be misunderstanding you, but I don't think that is in line with what the developers want for hacking.  IIRC, hacking at a lower AIP should incur more of a response than hacking at a higher AIP.

I suppose I probably did intend that at some point, yes.  Or at least that hacking with a large reserve of HaP be less dangerous than hacking right on the borderline.

Ultimately, though, I wonder if "less dangerous" is ever actually the goal.  Especially if it means "less dangerous than the last hack I did".  It's a bit counter-intuitive, but I think the feature is more fun if it ensures that it never gets less dangerous.  Otherwise it trends towards "this isn't really a challenge at all" and the various tediums and such that implies.

I don't want it to get to the point where spending positive HaP in a way that isn't obviously cruising-for-a-bruising (like riding the superterminal for a long time, or doing a bunch of hacks all at once, etc) is so crushingly hard that players actively choose not to spend that HaP (because it's become a bad deal), but I don't want them to feel like the hacking is not meeting any kind of effective resistance.  And most of the recent feedback I recall points towards the previous responses being fairly weak.

[Winge]

I may be misunderstanding you, but I don't think that is in line with what the developers want for hacking.  IIRC, hacking at a lower AIP should incur more of a response than hacking at a higher AIP.

I suppose I probably did intend that at some point, yes.  Or at least that hacking with a large reserve of HaP be less dangerous than hacking right on the borderline.

Ultimately, though, I wonder if "less dangerous" is ever actually the goal.  Especially if it means "less dangerous than the last hack I did".  It's a bit counter-intuitive, but I think the feature is more fun if it ensures that it never gets less dangerous.  Otherwise it trends towards "this isn't really a challenge at all" and the various tediums and such that implies.

I don't want it to get to the point where spending positive HaP in a way that isn't obviously cruising-for-a-bruising (like riding the superterminal for a long time, or doing a bunch of hacks all at once, etc) is so crushingly hard that players actively choose not to spend that HaP (because it's become a bad deal), but I don't want them to feel like the hacking is not meeting any kind of effective resistance.  And most of the recent feedback I recall points towards the previous responses being fairly weak.

Hmm...OK, let me ask this:  are you looking at keeping the inverse-relationship between hacking and AIP?  I liked that, as, even if a later hack is easier on its own, the AI is just going to turn around and bash your face in with waves/CPAs/Exos, etc.  I do think that hacking should be harder overall, though.

Also, if you are going to change the mechanic, I need to capture a lot of planets and hack a ton of stuff in my current high-AIP game.

[Faulty Logic]

Hacking now has an extra source of suspense and challenge: once the hacking _stops_ there will be a rapid short-term post-hacking response from the AI proportional to how much you ticked it off during the hack. If you're adequately prepared this won't be too big a problem (though I hope it will add to the excitement). But if you've dug yourself too deep a hole... well, have fun with that

I disapprove. This guarantees that the normal pulses will be boring, because if the pulses are strong enough to be interesting, then the post-response will kill you. Especially with the superterminal.

[keith.lamothe]

Hacking now has an extra source of suspense and challenge: once the hacking _stops_ there will be a rapid short-term post-hacking response from the AI proportional to how much you ticked it off during the hack. If you're adequately prepared this won't be too big a problem (though I hope it will add to the excitement). But if you've dug yourself too deep a hole... well, have fun with that

I disapprove. This guarantees that the normal pulses will be boring, because if the pulses are strong enough to be interesting, then the post-response will kill you. Especially with the superterminal.

Just to clarify, my understanding is that the normal pulses were boring before this.  So we went from "boring in the middle, boring at the end" to "boring in the middle, potentially interesting at the end", right?

If so, the midway pulses can also be buffed up, and the post-response can be made a lower proportion of what happened in the middle so that it doesn't get absurd.

Or am I missing something?

[chemical_art]

If so, the midway pulses can also be buffed up, and the post-response can be made a lower proportion of what happened in the middle so that it doesn't get absurd.

Or am I missing something?

I think the issue is that making the end pulse the strongest makes the normal pulses meaningless in the big picture. If you can survive the big pulse, the others don't matter.

I'm not a particular fan of a huge pulse at the end of hacking. It does make things interesting, but it is very obtuse for the lay person on when the cut-off point is.

With most things, like AIP, the responses to increased threat is mostly a bunch of small steps, so it becomes more apparent whenever a threat is reaching a breaking point. Yes, there are things like CPA's, but even they give a concrete description of the pain they will level before it actually occurs.

With the old pulse model, the turnaround time from potential pain to actual pain is fairly small, and just as importantly, could quickly be cutoff. It was still possible to dig into a hole if you failed to eliminate the terminal, for example, but usually the player was able understand fairly immediately how much is too much response.

With the new model, if a player waits until the pressure is too much from the normal waves, it is  game over. This is itself not so bad (but it is hardly good either). What is bad though is to counter this is you need a very fuzzy method of thinking when it is time to stop. There are no "X units approaching Y in Z time" like with waves. Nor is there are a killswitch. So the response is to achieve a desired cutoff point / choice between risk and reward is repeated save scumming to figure out at the point of time how much is too much.

Save scumming because you miscalculated the numbers can be OK. Save scumming because you cannot find numbers is not.

[Toranth]

Second, change the 'chaser' to be more like the Exo-waves that result when you kill a Core Guardpost.  Base the Exo budget off of a small, fixed number of response pulses: 5, say, or 10.

Hmm, for the superterminal I really do want the post-response to be proportional to how long of a ride it was.  It doesn't need to be 1 pulse per previous pulse but a 100 AIP ride should be much worse than 40, even aside from the difference in HaP spent at that point.
And on the other hacks I also want the response to be proportional to the severity of the hack, which is largely implied in its duration.  So perhaps the advanced constructor hacks shouldn't be 900 seconds, or perhaps the overall thing doesn't need to "charge" one pulse per 20 seconds, but that advanced constructor hack should have a much stronger post-response than a knowledge hack.

For the SuperTerminal, I understand.  I think the current response ramps up too fast, and ends up being rather cheesable (Cloaker + Warheads sitting next to the ST, and response isn't noticed).
For the other hacks, well, do you really need a special response to a Sensor or Sabotage hack?  Even a K-raid is less of a single hack, and more of a series of hacks.  It's also 8+ minutes long, and the AI response will creep up the entire time, as opposed to all other hacks.
The 5/10/15 minute hacks could have three different fixed Level 1, Level 2, Level 3 responses.  Maybe like the Reprisal, where 2 is twice 1, and 3 is twice 2?  But still, roughly fixed and hence predictable.  I also agree with Faulty, that if the final response is too big, it ends up being the only part of the hack that actually matters.  Finally, note that all non-SuperTerminal hacks take place on AI worlds:  This means running away is always a good option - then the response is just threat, like any other.

(Another side note:  Design Download should probably switch times with the Research Redirector - At least with the R-R, you still need to capture the planet.)

Anyway, on hacking, what I'm currently thinking is:
1) Make base hacking response value come from total HaP used.

2) Make the post-response spawns never apply the negative-HaP penalty, and not act as if HaP is lower than zero (which I guess is tricky in light of the first point, but basically deducting -current_hap if current_hap < 0).
- the latter part being to avoid situations where simul-hacking 10 fabs results in -6000 hap and a "you die immediately" post-response, even with the negative-hap penalty disabled for those spawns.

3) Have the non-superterminal hack post-response spawns not do wild rolls (the superterminal one already doesn't).  They might need further reduction but this would be a pretty major change I'd like to see the results of first.

4) Reduce the superterminal post-response spawn strength by maybe 20%.

5) Have hacking response spawns in general be normal units instead of zombie again, to see if the overall effect is better.  If related exploits are truly a problem then there are some other things I could try (pulling zombies into normal AI would be one thing though it would undermine the meaning of zombie, or maybe just giving the hacking-spawned units an independent no-reclamation flag, or something like that)

Points 1 and 2 sound good.  For three, the SuperTerminal normal spawns were already doing wild-rolls, but the final response roll did not.  As for #4, well, I think cutting it down to 20% would be more appropriate, but we'll see.
For 5), I think anyone using limited HaP as an attempt to reclaim ships gets what they deserve.  I would have preferred to see groups of zombies, but having them convert to normal threat is better than what's there now.  Could they be leader-ized easily as normal units?  Or spawn as mini-exos, targeted already at something on spawn? 

Hmm...OK, let me ask this:  are you looking at keeping the inverse-relationship between hacking and AIP?  I liked that, as, even if a later hack is easier on its own, the AI is just going to turn around and bash your face in with waves/CPAs/Exos, etc.  I do think that hacking should be harder overall, though.

Both lore-wise and previous intention-wise, I think you are correct.  But gameplay-wise, this meant that as long as your AIP reduction was more than your Total HaP Used, you got the minimum response to any hacking.  Considering every big enough map has at least 8 Datacenters and 4 Co-Processors, giving between 280 and 170 (the CoProcs count the full -120/-90 here), you could do almost all your hacking in a game without the AI ever moving off "AI Response: 10 (Very Low)".
This led to a lot of very easy, rather boring, hacking until that margin ran out.  Then, suddenly, after allowing hundreds of points of Hacking without reacting, the AI suddenly begins rolling out the big guns.

This may result in a need to adjust the AI response levels a bit.  Previously, a response in the 400-500 range was getting difficult, which meant in the 600-800 HaP Used range.  But I think there's worth to making hacking more difficult right now.

[keith.lamothe]

With the new model, if a player waits until the pressure is too much from the normal waves, it is  game over. This is itself not so bad (but it is hardly good either). What is bad though is to counter this is you need a very fuzzy method of thinking when it is time to stop. There are no "X units approaching Y in Z time" like with waves. Nor is there are a killswitch.

Well, this may or may not have been a good idea, but part of the point of the post-response thing is to make it so you don't have an immediate kill switch.  So you can't just summon AI units at-will (for whatever nefarious purposes) in that kind of on/off fashion.  You hold the pace, but you don't control the AI.

That said, perhaps the post-response is a bit overeager

Save scumming because you miscalculated the numbers can be OK. Save scumming because you cannot find numbers is not.

FWIW both for superterminal and non-superterminal hacks the alert box does give you the numbers you need to figure out what the post-response will be. 

For the superterminal it tells you how many pulses will happen on-death, so you just have to think "What was the last pulse like?  Do I want to handle something worse than X times that?"

For the non-superterminal ones it tells you how many seconds the post-response will run, so you just have to think "What was the last pulse like?  Do I want to handle more than X of those all in a row?"

Perhaps that's not enough information, I dunno.

For the other hacks, well, do you really need a special response to a Sensor or Sabotage hack?

Fair enough. 

Even a K-raid is less of a single hack, and more of a series of hacks.  It's also 8+ minutes long, and the AI response will creep up the entire time, as opposed to all other hacks.
The 5/10/15 minute hacks could have three different fixed Level 1, Level 2, Level 3 responses.  Maybe like the Reprisal, where 2 is twice 1, and 3 is twice 2?  But still, roughly fixed and hence predictable.

Sure, something like that would probably be fine.

For 5), I think anyone using limited HaP as an attempt to reclaim ships gets what they deserve.

And that's exactly what I thought... then I remembered you can basically run a fab-hack or advanced-constructor hack however long you like that's short of the full duration, and then cancel it, for no change in HaP.  At least, that's how I remember it.

But yea, I'd rather find another solution to the reclamation-hacking exploits, if they need a solution at all.

I would have preferred to see groups of zombies, but having them convert to normal threat is better than what's there now.  Could they be leader-ized easily as normal units?  Or spawn as mini-exos, targeted already at something on spawn?

Oh, sure, they can be mini-exo'd with the galaxy-wide-chase logic that makes them super-focused.  Did you have particular target choice logic in mind?

[chemical_art]

For the superterminal it tells you how many pulses will happen on-death, so you just have to think "What was the last pulse like?  Do I want to handle something worse than X times that?"

For the non-superterminal ones it tells you how many seconds the post-response will run, so you just have to think "What was the last pulse like?  Do I want to handle more than X of those all in a row?"

Perhaps that's not enough information, I dunno.

More information would help immensely, it could be provided a number of ways, from strength to equivalency to a fighter, etc, but almost anything is better then "X times whatever this pulse is" or "this will continue to Y time". I want more concrete numbers, not multiples of fuzzy numbers. Doesn't mean concrete numbers down to the unit are always best, but I hope I convey the spirit of the idea. Like how on exos do not provide exact numbers, but do provide numbers (in the form of large objects) at least give reference points to compare other exos. But even then, exos provide threat to give more data.