Queller + Quell AI - General Release, Version 10 by Quitch(y)
-------------------------------------------------------------

This file lets you know what I've added to this version, what problems I already know about, and generally what you should be telling me about (do my additions work for example).

Oh, and as ever I recommend you don't look through the weights and limits. Firstly, this may cloud your judgment, secondly, you'll have less fun if you know what is going to hit you ;).

==============
  Version 10
   10/11/99
==============

ADDITIONS
---------

* Knocked down limits on Shooters and Annihilators to more practical numbers. Also turned down their weights, as they clog up factories. I have been considering adding AI Shooters, ones which will cloak, as well as AI cloakable Fusion.
* Fiddled slightly with the resource to plant ratios.
* Stopped the AI building air straight away (or at least, decreased the chance of it happening).
* Weighted Slashers slightly higher to insure that the AI had sufficient numbers of them, and wasn't split too much between them and Instigators. Also decreased the limit on the number of Instigators allowed.
* Turned shell units down slightly, back in line more with their V8 weights.
* Limited defenses slightly lower, especially on Metal maps as the AI's base is concentrated in one area.
* Ships have been turned off, although occasionally useful in the current profiles, they were usually built in a tiny patch of water, and were just a resource drain. This also means there is no longer any true need for a WindAcid profile. Water defence is still used however to provide some resistance to human ships, or any nearby ground troops.
* The usual general changes have been made, 0.5 of an increase here, 0.5 of a decrease there, the odd unit allowed, and the odd unit banned. The usual.
* Oh, and a HUGE pat on the back for Core, Commander, who noticed that some of the CORE units had CRO at the beginning of their code, rather than COR. Luckily this error was only attached to some unimportant items, but none the less, that could have been embarressing. Thanks again!
* Turned off hovers.  Not only were they a drain, but they contributed to Queller hitting the limit very quickly, and they are really just level 1 units.  Queller doesn't want to hit the limit with loads of level 1 and no level 2.
* The above pretty much concludes all work on the following profiles: Default, Wind, Metal and WindMetal. I will make a few general changes, but these will not be specifically for these profiles. There may be a couple of errors where I have forgotten to transfer values from one profile to the rest of them, but I hope these are minimal.
* Quell comes into being, all hacked units are removed from it's profiles, but no other changes are made.
* Queller and Quell are released for the first time!
* CC now totally assumes you have the 1.2 bugfix enabled for CC mode. Although you do not need it to play CC, Queller CC will not work as well without it. However, Queller TA assumes that you don't have the fix, and you should disable it for any standard TA games you play.

PROBLEMS
--------

* The AI has slight problems with wind maps. For some reason, the AI handles solar well, but even though wind usually gives more (otherwise I wouldn't use it), the AI doesn't get enough energy! The AI doesn't seem to treat wind the same way as solar (it doesn't have the same built in codes for it). I am working on a resource system to counter this.
* I haven't tracked down the rouge fake silo file yet, but since this is a general release, I'm sure someone will find it.
* There are bound to be some problems with Quell. Deleting and renaming is not very exciting, and the mind wanders. Some useful thing may have been deleted, or typing mistakes may have been made. Let me know if you encounter any.

AI on AI Games
--------------

Just to see how Queller has come along I put 2 Queller's v10 BETA against 4 TA AI's (with bugfix activated).  Queller won every game on Painted Desert within 25 minutes, and on Metal Heck and Metal Heck 2 within 15-20. Version 5 took an hour, so this is quite an improvement. I guess we can say Queller is now at least twice as good as 3.1.

**********************
* HISTORY OF QUELLER *
**********************

Queller has been through a lot since it was first released for testing. Here is a history of the changes Queller went through.

=============
  Version 9
   02/9/99
=============

ADDITIONS
---------

* Corrected all weighting mistakes I could find, there should be no more of them left.
* Changed the two sides defensive weights slightly. HLT's are used more frequently, Vipers less so.
* Added Switeck's AI aircraft, to allow the AI to use it's planes far more effectively than it currently does. Thanks to Switeck for allowing me to use them.
* Added more missing weightings and removed unnecessary lines to improve speed. Corrected various mistakes in the files.
* Upped weightings for some of the bigger guns and power sources. Nukes will be used a lot more frequently by the AI.
* Made EMP and Neutrons less important than nukes.
* Increased the number of metal extractors the AI can build, I thought it wasn't expanding far enough on maps such as Yerrot Mountains where each spot is not worth a lot due to a low metal weighting, but I found it had just hit it's limit. This has been corrected.
* Weighted solar slightly higher to deal with early problems.
* Made a lot of changes to level 1 Kbot weightings for both sides, so that it will produce Kbots which get built faster earlier.
* Aircraft plants should get built eventually, but not first, and move quickly onto level 2.
* Removed some pictures from the EMP and Neutron that had files missing, I think they may be needed to use the weapon if you capture it, but until I know for sure, I wiped them.
* Changed layout of AI files after my continued e-mail talks with Switeck.
* I have deleted the Medium and Easy difficulty levels. For now.
* Greater use of rocket troops by both sides.
* Added TAMEC skirmish maps to the maps list, along with Metal Heck 2.
* Created Metal, WindMetal and WindAcid profiles. See the readme for which maps they work with.
* Updated the OTA files.
* Lowered unit limits to more reasonable values. This also means the AI gets construction units earlier, but not so early as to stop it's first assault.
* Turned artillery down in weight so that the AI made less use of it.
* Revised the ThePass profile.
* The ARM use their Annihilator more than the CORE use the Doomsday Machine, and the CORE use their Neutron more than the ARM use their EMP.
* Updated the map list after downloading virtually every TA map from TAMEC that got an overall rating of 10. I also got a lot of their 9 rating maps too.

PROBLEMS
--------

* You *still* cannot use captured AI plants. This should be fixed next version. I know how to do it, it's just a matter of implementing that knowledge.
* There are grey boxes where Switeck's Fake Silo's should be. I am working on removing them.
* No UH AI yet. Because I have no experience with it, it may be while before I add it, though it should (crossed fingers) be in by Version 10. Don't expect miracles though, I will need a lot of feedback from everyone with UH experience.
* For some reason my AI will not use Levelers on The Pass. Please let me know if you're having the same problems.
* I am playing with the idea of AI Crawling Bombs, but I haven't had time to test or implement it yet. This is probably more of "just for fun" rather than any practical AI uses.
* I haven't checked really looked much at the maps I downloaded. They may be better off with alternate profiles. If you have one of them, please see if the profile it comes with works OK (if Queller supports it).

AI ON AI GAMES
--------------

The challenge set before Queller for this version was to better what I believe is seen as the most difficult to beat non-cheating AI out there, Mostly Harmless.

Seeing as the new profile I'd just developed was WindMetal, I decided the first game should be on Metal Heck 2.

I used my standard recommended settings, with Queller playing as ARM and Mostly Harmless CORE. It was a 2 on 2 to help eliminate any random mistakes the AI's may make.

* GAME 1

Mostly Harmless kicked off with it's standard production, while Queller went in for an immediate kill. Both Quellers went for the closest MH. The first attack was barely being held by this MH, when the second hit from the other direction. This was too much for MH who started to fall back. The second MH came down from on high and hit one of the Queller bases from the side. The HLT's held the attack, and eventually forced MH to fall back and regroup. The battle between the 2 Quellers and 1 MH wasn't too long, eventually they broke through, and the resulting Commander explosion took out both attack forces and the MH base. The other MH had not wasted time, and had expanded to such a size that it was larger than both Quellers put together.

The 2 Quellers gathered new attack forces and headed towards the one being launched by MH. This conflict was very slow, level 2 MH troops battled with level 1 Queller troops. MH forced Queller back, but very, very slowly. However, eventually, MH hit it's limits, and having failed to break Queller, was slowly driven back by weight of numbers. I quit at this point because the game speed had hit -3.

* Game 2

Disturbed by the fact that Queller had not won outright when having 2 to 1 odds, I tried a 1 on 1. MH simply out produced Queller, held it's early attacks, and drove it back. MH won.

* Game 3

I tried a Default map now, Painted Desert. Queller attacked quickly, it's first force hit a few troops, however these were enough to distract the AI until MH had built enough factories to force Queller back, and eventually win.

* Games 4 & 5

By now I had revised all my profiles, and so Queller came back fresh for the fight. It was not helped by the fact that it was playing as the weaker ARM (an AI always plays CORE better eventually). Both games were on Painted Desert, 1 on 1. Queller quickly got attack forces ready, it expanded swiftly, it got more factories and more attack units. MH was unprepared and fell back under merciless assault, nano-stalling as it lost it's outer resources. Even Goliaths were quickly overwhelmed by the shear numbers of the assault. Both games lasted less than 25 minutes, both were victories for Queller, which didn't lose a single building.

* Games 6 & 7

2 more games, both on Metal Heck 2. I didn't finish either, as I was skimming the Metal profiles for bugs more than anything, but at the end of the first game, Queller was driving MH back, when I quit the second, MH was pushing Queller back.

Conclusions
-----------

These games were more for friendly competition (and to show how great Queller is :P), than for any true testing. However I did learn some important lessons, from the mistakes of both sides. I wouldn't say the games were particularly fair, MH uses a lot of missile troops, and an AI has a hard time beating missile troops with anything other than more missile troops. MH also used CORE, which I find the AI handles better. However MH did show me the advantages of getting some construction units out early (I don't go in for right at the beginning, however I do think "earlier" would be better), and more importantly, expanding to multiple factories early.

=============
  Version 8
=============

ADDITIONS
---------

* More slight changes, allowing CORE and ARM to play to their level 1 advantages.
* Fiddled with the layout of the file, to take advantage of the fact that TA uses a top down reading to try to get it to run as fast as possible, and to even out the speed discrepancies between the ARM and CORE in decision making.
* The AI_Nuker file has been replaced by Switeck's Swi_Nuke which contains his AI-only Nukes, EMP and Neutron.
* Added missing CC codes.
* The AI now has a Wind and ThePass profile.
* Updated the map list in the readme to include all CC and Cavedog download maps.
* AI should get it's first plant up slightly faster than in Version 7 and more reliably.
* Hovercraft weightings improved.
* Accepted an offer by BSR to create a non-cheating AI for his multiplayer mod UH. I'm just too good :). I should have added UH compatibility by Version 9, this will become Queller UH.
* Turned off the original Silo's, EMP and Neutron.
* Added Creepermans Jammers, they have weapons of strength 0, and this allows them to accompany AI attack forces.
* Added some special AI Vehicle plants of my own design (thanks to BSR for his guidance) to replace the ARM Vehicle Plant, the CORE Vehicle Plant and the CORE Advanced Vehicle Plant. These plants have an exit out their back as well as front, this is to make it less likely that the AI will block off all the exits from a plant. However, as in Kbot labs, the AI may sometimes just sit a unit *in* the factory, stopping production. I don't want to add side exits though, as this would make the plants too hard to hit when open.
* Added OTA files to the ZIP to allow normal TA and UH users the benefits of Switeck's map profile changes.
* Turned the Warrior weighting right down, as it was taking so long to build that the ARM AI's reinforcements were drying up.
* Removed all irrelevant weights to help speed up the AI decisions.

PROBLEMS
--------

* Sometimes the AI tries to get a second plant up too quickly.
* You cannot use captured AI Vehicle Plants.

=============
  Version 7
=============

ADDITIONS
---------

* Defence and resource weightings have been completely redone, plant weightings have been changed accordingly. The ratios are not totally different, but I wished to move to lower figures, rather than the continual increase into large numbers that I had started.
* Corrected some small mistakes in the various profiles.
* Changed ship weightings after tester and TA forum suggestions.
* Prevented the lower difficulties from building some of the bigger guns.
* Level 1 Kbot and Vehicle weightings have been altered slightly for both sides.
* Added a CC default profile.
* CC weightings assume TA has been modified with the 1.1 bugfix by Switeck, the ARM especially rely on this as they now use the Warrior again, but without the fix the Warrior is useless.

PROBLEMS
--------

* Sometimes takes too long to create a plant on some maps.
* CC weightings are in early days.

=============
  Version 6
=============

ADDITIONS
---------

* The AI has been brought more in line with it's Version 4 state, since Version 5 created more problems than it solved.
* The AI attack forces at level 2 should now be larger. Probably.
* The AI is more likely to build radar Jammers now to stop cowardly long range attacks from behind large defenses :).
* Level 1 Kbot forces have been revised.
* Aircraft Plants have been made slightly less important.
* Expensive level 1 defence has been slightly less of a priority again.
* Corrected the Nuke weighting codes.
* Limited the number of plants the AI can use to allow it to have a steadier swarm of troops, rather than small groups of them.
* Easy and Medium profiles have been added. Easy lives up to it's name. Medium needs more testing.
* The AI uses more close range troops to defend it's missile troops.

PROBLEMS (always waiting for additions :))
--------

* Ship weightings are pretty much untested.
* The AI tries to get too much metal before it's first plant. This causes problems on maps where the metal is spread out.
* The AI doesn't get it's first plant up quick enough.
* The resource and defence weightings need an overhaul.
* It isn't building Navel Yards, I'm not sure how much I can do to correct this. It's the same as the Geothermal problem.

AI on AI Games during V6 development
-----------------------------------

All games were on Painted Desert with 1000 metal and energy and random positioning.

Queller in version 5 managed to win a 2 on 4 game against the TA AI.  Queller used the forces of ARM and TA was CORE (in my opinion TA was always stronger as CORE). The game was a rout, the first enemy TA AI fell quickly, with another almost crippled.  The game dragged on slightly as the TA AI fought back with counter attacks, but eventually fell to the better weightings and defence of Queller. TA never made it past the fringes of either Queller base. The game lasted one hour.

2 ARM Quellers took on 2 BSRAI CORE's. BSRAI started off better, it quickly grabbed resources, and built vehicle and Kbot plants. Queller made a mess of the beginning, it took a while to build a plant and when it did, both ARM's built an Aircraft one first. Queller soon came back however as BSRAI settled down to it's usual conservative game. Queller built a force, and then built construction vehicles. While they decimated the first CORE base the construction vehicles expanded the base, and increased the defence. The last CORE could not penetrate this defence, and with his ally gone, the two ARM's found him easy pickings. The game lasted 27 minutes.

The next game used true LOS in unmapped mode.

The ARM AI beat a Boneyards player who made the mistake of treating it like the TA AI, and allowing himself a construction period. The AI, after messing around for 10 minutes going forwards, regrouping, going forwards again without actually getting near the enemy, finally attacked en masse and easily overwhelmed the line rocket turrets, and then most of the base by destroying the enemy Commander. After that it was a mopping up operation. It took around 20 minutes.

Version 5 took on Version 3 using the standard settings. Version 5 attacked first, and it almost broke through very quickly, except it's construction units decided to build some of the heavier defence items, and this dried up its reinforcements. Version 3, with it's lighter defence model, quickly counter attacked, broke through the uncompleted defence, and won the game. Needless to say these problems (which were the crippling errors Version 5 always faced) have been corrected.

=============
  Version 5
=============

HINDSIGHT: This version was a disaster, creating so many problems that Version 6 was more in line with version 4 than version 5. We (myself and the testers) only speak about this version in hushed tones.

ADDITIONS
---------

* Further improvements of the AI resources. The AI is no longer as Moho happy.
* The AI should use a more intelligent defence mix, with some of the more powerful weapons made slightly more important.
* Changes made to the number of plants made available.
* The ARM can now hold their own at level 2 against the CORE. Usually.
* Aircraft plants are built faster, while only rarely being built as the AI's first plant.
* The AI weightings are almost complete for offensive units, and this now includes variation of units. It should be able to deal with most situations, without compromising it's ability to fight.
* The AI limits on silos actually work now.

PROBLEMS
--------

* Geothermal energy is still not being taken advantage of by the AI enough.
* The AI does not go for Big Berthas and Intimidators quick enough (though, I don't know how quickly you lot get them either but even a losing TA AI beat my AI to them).
* I think the AI's defence is a bit light, although I'm not entirely sure this is the case. I am going to look into this a bit more. Mind you, if you let them start to win, it's not.
* The ARM doesn't use big enough groups of Level 2 units.

=============
  Version 4
=============

ADDITIONS
---------

* The AI will now build sea units in it's default profile (thanks to noo_bee for his report).
* The AI will build less level 2 factories.  This is an attempt to combat it's constant nano stalling at level 2.
* I have introduced a wider range of units, especially ones that have the ability to "lob" shells.

PROBLEMS
--------

* The AI does not counter Nukes (I cannot correct this, but if anyone knows of a patch that allows the AI to build Anti-Nukes, please let me know).
* The Ship weightings are terrible due to my complete lack of experience in this area.
* The AI is not getting to level 2 planes quickly enough.

=============
  Version 3
=============

ADDITIONS
---------

* The AI will use cheaper defenses, but in greater numbers. This extra use of level 1 defence also means that metal extractors generally have a turret somewhere nearby.
* I have been working on improving the level 2 mix of ARM troops.
* Moho Mines and Fusion have been given much higher weightings. Long range artillery, such as the Big Bertha, has also been made more important to help break stalemates where the player may dig himself in.
* General refinement of what the level 2 construction units build.
* Stopped the AI building units and buildings that it cannot use.
* The AI should upgrade to level 2 slightly later than before, because it's construction units build more defence.
* The AI should have less energy problems now.

PROBLEMS
--------

* The CORE still beats the ARM. The level 2 mix is still not right.
* I'm not sure if half of the metal drain comes from construction units building Moho Mines. I may limit the number of construction units further.
* The ARM leaves aircraft too late.
* The ARM never breaks through the CORE lines.

=============
  Version 2
=============

ADDITIONS
---------

* The AI will now expand further across the map. This has been achieved by increasing the importance of metal extractors greatly. When it builds at a metal point, it will create other buildings around that point, thus expanding its base. It will also go further afield to gain metal points.
* No longer will the AI avoid building planes like the plague. The ARM in particular will now value level 2 planes, as a way of fighting back against the Goliaths and Sumos. Many thanks to Aklein for suggesting the greater use of planes.
* The ARM AI now uses more long range troops, such as Mobile Artillery and Bulldogs, and relies less on Panther tanks to carry the day.
* Both sides now use similar troops during the initial level 1 battles. Due to the time it takes for the computer to update it's target, Flash tanks were being blown to pieces by missiles. The CORE have a couple of close range tanks thrown in, to distract opponents while the missiles rain in from afar.
* Metal has been made a much higher priority of the AI, as it always seems to run out. Level 2 construction units are brought out faster, and give Moho mines a high priority. If this problem continues, I may consider the use of Metal Makers.
* To save resources, and protect from the new aircraft raids, Rocket Turrets have been given higher priority. This also helps the AI save it's resources. Plasma Batteries have had their weightings reduced, since the AI was crippling itself by building them.

PROBLEMS
--------

* The AI doesn't have enough energy to begin with.
* The AI doesn't have enough metal to support construction.
* The CORE AI still beats the ARM AI at level 2.
* Planes are still under used.
* The AI still does not expand far enough.

=============
  Version 1
=============

The original version, it only included a default profile. It attacked quickly, but lacked the ability to play a long game. It was tested until it could beat the BSRAI on Mounds of Mars easily.