View Issue Details
| ID | Project | Category | Date Submitted | Last Update | |
|---|---|---|---|---|---|
| 0023857 | AI War 2 | Balance Issue | Oct 2, 2020 12:15 am | Dec 30, 2020 12:48 pm | |
| Reporter | Suzera | Assigned To | Chris_McElligottPark | ||
| Status | resolved | Resolution | fixed | ||
| Product Version | 2.602 Transports Fixed In Multiplayer | ||||
| Fixed in Version | Beta 2.710 The AIs Visit An Accountant | ||||
| Summary | 0023857: Game bugged: Supercat AI died. | ||||
| Description | Vanilla, no expansions. Settings default with 10/10/10 counterattacker other than these to help reduce rng issues: -GCAs 0x - I planned to use no turrets, so GCAs are just an rng aip bump or hacking point waste compared to their usual value for this run. Other AIP bumping defendables left in. -Remove unlock variance in Capturables -I allowed myself to use explore to see what was on the map to get a good (but not perfect) seed of ships I was going to use. I was aiming for about 6 Raid/Light ship strikecraft lines on the map. I did not use explore-all for gameplay, so I still needed to capture or hack to unlock travel to planets. General strategy: 1) Start in middle of map with 2 links that basically splits the map in half, minimizing the maximimum distance to AI homeworld. 2) No turrets. I don't have enough power to run them and keep enough fleet to actually hit the ai homeworld. 3) Don't blow up AI owned AIP destructibles like raid engines, magnifiers, troop accelerators, etc. Just gotta deal with it unless they're parked on a planet with ships in a transport I need. If they're in the middle of the defense snake, taking the planet is skipped unless it's a raid engine. 4) Logistics station line so I could move a defense fleet where it needed rapidly, as I am reliant on fleet defenses and they have the most forcefields to delay death from attacks. 5) Initially, 1 fleet that does everything. Stay in mk1 until every guard post visible is dead. 1 data center visible before mk2. 6) On ready for mk2, take a couple more stations to expand reach and snipe some more data centers, staying under threat wave AIP (100). Clear all guard stations visible again. 7) Split fleet into attack fleet and defense fleet since I probably have enough spare ships I don't need all on attack. 8) Claim a spire frigate or armored/hive golem (I'd make sure one of these was on the map before playing). This will form a significant part of the defense backbone. 9) Use offense fleet and raid frigates to continue clearing everything (I didn't quite clear all the stuff on side opposite AI HW). 10) Pick apart the AI Homeworld guard posts with suicide runs 11) Possibly optional: drop the overlord phase 1 to 1% hp, then repair to 100% fleet strength before final run. 12) If overlord phase 2 is not defeated in the first run, retreat to your planets and wait for it to arrive on one. 13) Put everything on the planet it attacks, and apply fleetball directly to overlordhead to win. Ultimately, I didn't do anything super cheesy. In the end, my attrition strategy did not seem work because the economy bonuses are too high to really run the AI out of budget to defend the homeworld, and I probably didn't need to spend quite as much time destroying non-hw guard posts as I did. I could probably complete the game in half the game hours with that and some more smaller optimizations in execution. Since I was playing really low AIP (sub-100 for about 12 game hours, and then sub-150 for another 12), there wasn't enough science to spend on power-increasing like CC upgrades, which did not allow enough power for turrets to be effective without rebuilding them before attacks. However, since there were no defenses, attacks were also too frequent to spend the metal rebuilding turrets for each attack. Another turret confounding factor is there wasn't enough capture wiggle room to make defense chokepoints. Everything had to be dedicated to reaching fleet increase targets, picking up transports/golem, and the ai core. Which means to use them EVERY planet would need them, and I just didn't have the power for that. Add to that the extra science expenditure to make them comparable to fleet strength and I just don't see a good time to use them unless your fleet weapon(s) happens to overlap a turret type. For this run, I focused on raid/light. Raid ships are very good general attackers that are hard to stop and can often catch retreating units, though having no turret tech overlap or specialization damage multipliers is a significant drawback. Since I didn't have the power for turrets, the no turret drawback was not a significant factor in my decision. Raid starting fleet also gets agile transport, which helps in the midgame when shuffling space planes around somewhat before they're split off into defense fleet. Station watching frigates were built, but usually left under the forcefields in non-pursuit mode just to increase apparent strength and make the threat attacks in slightly larger chunks. The station watching frigates cost a lot to make and don't really do any damage. It was much better to just leave them covered and repair forcefields until the main fleet could arrive. Station assault frigate autobuild was turned off due to being a waste of metal/power and at mk1 they didn't really increase apparently strength any. For science I had these things by the end: Raid 4 (lucked and had a tech vault, but not a major factor. made early game harder and later game slightly easier) Light 2 Logistics 1 Forcefields 1 Engineers 2 Metal 1 7 logistics stations with 1 direct upgrade each Home command 2 direct upgrades Armored golem 3 direct upgrades The early game is mostly focused on getting raid 4 and light 2. I spent one upgrade in engineers before raid 4 since I had a raid tech vault on the map. The second priority was mk 3 force fields and engineers. After that was having enough for a few mks on the golem and then everything else poured into metal and a bit of safety power against brownouts. While balance seems ok, the strength formula and balance could probably use a few tweaks. Things that overperform: Anything that is a double damage/half health form, like daggers or gunbots. Generally, killing faster is more valuable since more dangerous things are not automatically prioritize so they can be buffered by their normal damage/health counterparts. Add to that that killing things faster means you can move on to other things faster, letting you defend more things or attack more things. While they are rare rolls, I think multiplying damage by 1.5x and dividing defenses by 1.5x might be a more appropriate number. And they probably don't need extra side bonuses like extra speed (daggers) or actually 2.5x damage (gunbots). AI paralysis at higher difficulties. The more concentrated these are, the proportionally more deadly each one gets. Paralyzing 1% vs 2% of your fleet is not much of a difference at a little more than 1% dps loss, but paralyzing 90% vs 91% means you've lost 10% of your remaining dps. The main problems were large stacks of mk7 warbird frigates (some CCs only picked those) and paralysis guardians (had a whole bunch pop out of a hack a few times). While I don't think they get to potentally game ending concentrations in diff 8-, the player can't field 100 mk 7 warbird frigates into a 200 strength fight, so this is a fairly one-sided problem. These did cause some reloading and having to bunker under forcefields to clear out the initial numbers or else every single strike craft would get paralyzed. Things that underperform: Most turrets don't seem as good as making more fleet. While some turrets look good on paper (beam, plasma, spider turrets looking at you), they also take a whole lot of power you probably won't have at highest difficulties AND make the incoming threat attacks stronger when they do arrive so your fleet has a harder time dealing with them or they drill through force fields much faster. To get a good static defense strength size to get the ai to attack in chunks that your fleet can usually readily come in to help clean up doesn't take a lot. On higher difficulties, the overconfidence factor being 1 or less means they player should expect to lose unless they move a fleet to stop it, so if you have to micromange fleet to every battle anyway, that reduces the effective power of turrets and makes them more of a convenience thing than a need thing. The other potential strategy is chokepointing, which lets say we had the aip to do while reaching everything you need to, you'd still have to use fleet to clear threat at highest difficulties, but you'd need to do so pre-empitively before it accumulates to be stronger than your fleet can deal with in a reasonable timeframe if you really turtle up to where the threat attack could overpower your fleet. GCAs. To add onto all of the above problems with turrets, GCAs add 10 aip if destroyed and lost permanently. Which is either aip you can't afford at low aip, or could be game ending if your defenses failed since now your defenses are even weaker unless you had a junk GCA that didn't have compatible techs. Battlestations and citadels. In addition to turret issues like above, the raw stat increase for dps/etc seems abysmal. If chokepointing were more viable at highest difficulty, maybe at least they'd be a decent turret count. Or maybe if the battlestation turrets were cheaper to build so you could more easily take it to where waves come in. As is, they don't seem worth the resource expenditure or seeking to grab from the map so it's needless clutter as far as I'm concerned right now. Having so many weapon techs. I feel like there's too many, and getting the science to get more doesn't scale in comparison to AI attacks very well at higher difficulties (and probably even moderate ones like 6-7) vs stacking into one and rerolling everything to that one tech or getting a good seed rng. It's a big rng factor regarding valuing what is in transports, GCAs and ARS, making them vary between extremely valuable to potentially worse than useless (or at least huge hack sink for rerolling). I think if the weapon categories were halved, tech vault count were halved, and science costs remained the same as a starting point the game would be better for it and reduce the amount of whole-campaign-difficulty affecting rng. Also if so, the ship armor upgrades could also stand to have 2k knocked off their prices to keep them about in line with weapon upgrades. Lower mark strikecraft. It's always more worth it to stack tech into one than spread it. By the time you're doing mk6->7, you're getting 20% more on something that is 43x stronger than a mk1 unit between multiply all the cap increases by the damage ratio by the health ratio. The mk1->mk2 increase is 1.4 damage * 1.25 health * 1.25 count = 2.1875 at cap compared to mk1. Mk7 is 64.683 multiplying everything together. The tech cost for mk1 using weapons is 1500, and for mk7 it is 1500 + 2750 + 5000 + 7500 + 7000 + 10000 = 26250. The power per science for mk1->2 using weapon is 2.1875 / 1500 = .001456 and for mk7 is 64.683 / 26250 = .002462, or about 70% more power per science to go max mark on one weapon tech EVEN CONSIDERING just using the armor for one weapon/armor combo and picking the first weapon tech for upgrading your mk1. What I think needs to happen here is that strikecraft unit count increases need to scale like frigates (1.1 per mark), so instead of about tripling in count they only increase by about 60%. Doing this would also make the supercat run dramatically harder to where I'm not sure I could have done it. This may even be a core scaling issue. Picking ships and turrets to counter the AI picks. In large part due to the point above about deep investing into marks, specifically picking correct counters to the AI's ship picks is not particularly a good idea. It doesn't matter if the ship you have is half as strong against what the AI has as another ship that might theoretically be better suited if your mk 7 deathball ships are 8-20x as strong at base as the counter would be because of mark differences. This is a problem for the game since "pick the ship category you want at the very start and just stack it regardless" is not as strategically interesting as "observe what the AI has and adapt over the course of the game". Another problem on this front is that ARS lock you into ship choices, so you can't go back and redo your ship picks to be better counters. GCAs have this issue at a smaller scale since rerolling costs hacking points and it can take a lot of hacking points to get what you what. Player frigates. This is likely due to the high mark problem and that they only get 1.1 per tier instead of 1.25/1.2/1.33/1.25/1.2. Strikecraft vastly outperform frigates in standard fights once your tech is rolling higher. If strike craft are brought down a bit like suggested above, frigates might be comparably more useful other than specialized uses. Higher mks (5+ but also kind of 4) of golems, spire frigate, arks, battlestations and factories. They don't increase in number, but their power stats increase kind of like strike craft and frigates. While they start off science efficient in low-mid AIP games because of high base stats it becomes less so in later marks. If strike craft count increases are brought to frigate increase numbers this also might be more useful and make these things have more impact for their AIP cost. Going to upload some saves in first note so all this typing doesn't get lost. | ||||
| Tags | No tags attached. | ||||
|
|
mkchoice was when I was deciding to hack some explore on data centers to stay mk 1 longer or not. I chose not, and I don't think this had a huge impact. golemornot was when I was deciding whether to take the armored golem. I chose to do so, and think this helped a lot. bleed it out was when I started focusing on clearing the AI worlds near the core. LAST TECH was when I was getting the last science points I would for the game and deciding what to spend them on final approach was when I started sniping out the guard posts from the ai core world. |
|
|
A few things I noticed after loading the save file (though I didn't see a "victory" save or "about to kill Phase 2" which is normally expected for those kind of AAR). -Turrets reinforcement is crazy -Sub 100 AIP games are the most proeminent balance problem with those victories. True you can't take the absurd scaling of diff10, but staying below 100-140 AIP is how all of those games are won -Monotech is unsurprisingly king. -SuperTerminal is as expected not that useful -Fleets are small (unsurprising but it's also disappointing to play with such small fleets) -Not surprising that Warbird caused trouble, since their best counter are turrets -I do notice this is an "easier" AI rather than full ensemble, this probably helped a bit -I also noticed the Strikecraft scaling up better than frigate issue too. With a calculator to compare DPS gain between strikecraft and frigates, it's really obvious. I did a much more difficult difficulty 10 myself (posted earlier in the tracker) that used turrets a lot and was wrapped up in 3h30, so you can definitively make turrets work. I do agree most of them aren't worthwhile and that's why we're currently trying to work out a rebalance for them. I am amazed you managed to do this without turrets, Suzera. I suspect you relied on the innately high speed of raid units in order to keep moving around. Very impressive. |
|
|
Full ensemble is also in the easier AI category. Overall, I do think the occasional massive counterattacks (and especially trying to break into the core world) were more of an issue than the AI having more unit variety would have been. I didn't upload a victory save because it didn't seem like it was really a decision junction, which was the idea behind the saves I did upload. I wanted to go for the victory asap once starting to take down the posts, or if it failed for some reason I was going to reload the pre-post destruction save and power up some more. If I do another supercat run of note I'll add one next time. If absolutely need be for this one I can just load my pre-overlord phase 1 kill save (which I don't think I uploaded) and do it again though. The whole posts to victory segment took about 20-30 minutes if I recall correctly. Even without raid units, transports are still pretty fast so it should be workable with any units. Raid units do make it a lot more convenient though. Since the patch a couple weeks ago the warden horde seems a lot thinner, which massive warden fleets getting in the way constantly was basically what took half the time. Between killing far less guard posts and the warden fleet seeming to be much smaller for some reason, another diff10 run right now would probably be a lot faster and easier. |
|
|
Balance is shifted around a lot in the most recent beta! Will be interesting to see how it fares. There are more changes that are coming that are buffs for a variety of player styles, but at any rate we're headed for a bit of a new era of the game. Also Badger has made the AI smarter a couple of times since this ticket was opened. Thanks! |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| Oct 2, 2020 12:15 am | Suzera | New Issue | |
| Oct 2, 2020 12:18 am | Suzera | File Added: mkchoice.save | |
| Oct 2, 2020 12:18 am | Suzera | File Added: mkchoice.savemet | |
| Oct 2, 2020 12:18 am | Suzera | File Added: golemornot.save | |
| Oct 2, 2020 12:18 am | Suzera | File Added: golemornot.savemet | |
| Oct 2, 2020 12:18 am | Suzera | File Added: bleed it out.save | |
| Oct 2, 2020 12:18 am | Suzera | File Added: bleed it out.savemet | |
| Oct 2, 2020 12:18 am | Suzera | File Added: LAST TECH.save | |
| Oct 2, 2020 12:18 am | Suzera | File Added: LAST TECH.savemet | |
| Oct 2, 2020 12:18 am | Suzera | File Added: final approach.savemet | |
| Oct 2, 2020 12:18 am | Suzera | File Added: final approach.bak | |
| Oct 2, 2020 12:18 am | Suzera | Note Added: 0058891 | |
| Oct 12, 2020 2:05 pm | ArnaudB | Note Added: 0059170 | |
| Oct 23, 2020 1:43 am | Suzera | Note Added: 0059361 | |
| Dec 30, 2020 12:48 pm | Chris_McElligottPark | Assigned To | => Chris_McElligottPark |
| Dec 30, 2020 12:48 pm | Chris_McElligottPark | Status | new => resolved |
| Dec 30, 2020 12:48 pm | Chris_McElligottPark | Resolution | open => fixed |
| Dec 30, 2020 12:48 pm | Chris_McElligottPark | Fixed in Version | => Beta 2.710 The AIs Visit An Accountant |
| Dec 30, 2020 12:48 pm | Chris_McElligottPark | Note Added: 0060181 |
