kjf no good in 1942 online

@aardvarkpepper said in kjf no good in 1942 online:

Keep informed with the action log and war diary

I re-read my last post (usually do to clean up a bit), and this bit from 1942 Online’s Steam store page description popped out at me. Keep informed, sounds nice, right?

But actually, 1942 Online’s war diary doesn’t give you proper information. Last I saw, you check the record, you see what went in, you see what came out, but you don’t see what the attacker hit allocation decisions were, you don’t see what happened in each round of combat, it’s just boop, win or lose. I don’t think the war diary even accurately reflects the order of combats.

TripleA gives all the details right down to the dice rolls. You REALLY know what happened, you get the hit allocations, the attack/retreat decisions, you see it all. It’s not fair to compare TripleA, which has been in development for years, with 1942 Online. But still, things are what they are.

When is having that information important? This is one of those things - if a player doesn’t already understand then even when it’s pointed out to them exactly why it makes a huge difference they still may not understand. It’s just words on a page. But I’ll try to explain.

When an opponent initiates any attack, you can look at the attackers, the defenders, the probability distribution of outcomes, the strategic and tactical situation, and what your opponent could have done differently (could have done other attacks entirely, could have allocated different units to different combats, etc.) This gives you insight as to your opponent’s read on the board position and risk preferences.

What does it say if a USSR player tries a triple attack, a Baltic Sea attack, or Baltic States/West Russia, or Ukraine/West Russia? Each of those attacks says something different about the player. A triple attack player is not afraid of risk. They are going to hit out, if you have a close stack battle in the offing, you’d better watch out because your opponent might be coming at you (and with the two-peak model, things could turn out real ugly). A Baltic Sea attack says your opponent is very confident in their own ability, others say Baltic attack is bad but your opponent doesn’t care and your opponent might be good or bad, but they’re going to generally do things that aren’t “meta” plays. Baltic States/West Russia is another unconventional play though less so, and Ukraine/West Russia is pretty meta. You can’t assemble a complete opponent profile based off one move - maybe an opponent decided to do something “different” this game, or maybe they use conventional openings but then don’t know the “meta” for following turns so they’ll change things up - but you have some idea.

Even things like composition of attack makes a difference. A R1 West Russia / Ukraine open is different depending on whether you send two tanks or three tanks.

Then sometimes opponents make moves that might seem irrational but aren’t. Like if an opponent hits West Russia only but holds some units back, that might be thought a mistake if you’re going with the “conventional” reasoning that West Russia is simply a territory that USSR conquers first turn. But if you know about the West Russia strafe / retreat into Karelia line, well, there you go. You hold back units because you want at least one German unit to survive so you can retreat the masses to Karelia. Not that I’m saying I think that’s a solid line, but I’ll leave off discussing that.

Returning to getting information off your opponent. If you can tell all that information from an opponent just from what they attack with, then how much more information can you get from an opponent by knowing what casualties they assign to what units and when, when they decide to attack or retreat, and so on? Again, it’s not that each decision is isolated. Each decision is made in context of the board position as a whole, even knowing the order an opponent resolved combats is important.

Before continuing, to address that last. Suppose in the LHTR setup, Germany attacks UK’s destroyer/transport off East Canada with one submarine and attacks UK’s battleship/destroyer (and likely USSR sub) with two submarines, cruiser, and two fighters. The battle off East Canada should be conducted first. Why? If the UK destroyer is destroyed, the Allies can’t hunt any German submarine survivors of the UK battleship battle. But if the UK destroyer survives, the Allies can hunt any German submarines, and submarines are poor defenders. If a player initiates both combats but doesn’t do the East Canada battle first - well maybe it’s 1942 Online’s UI (I think you have to right click or something to determine order of battles and that isn’t in the documentation, but whatever). But maybe also an opponent doesn’t know what they’re doing. Watching out for these sorts of things clues you in as to your opponent’s abilities.

Returning to information-gathering. You get a lot of information off an opponent’s combat movement and order of resolution during the combat phase, but within each combat your opponent also gives information. Information you should have includes opponent hit allocation which can give you insight into your opponent’s plans, opponent decisions to press or retreat in the face of changing probability distributions, and that in turn gives you a read on whether your opponent can properly reassess combats from sub-phase to sub-phase and act appropriately. All such information is lost in 1942 Online.

All this doesn’t mean that you can’t enjoy 1942 Online. There are players that can and do. But if you’re a sharp player that pays attention to detail, you will feel the differences between 1942 Online and actual board game play.

@kakarrot1138 said in kjf no good in 1942 online:

Your UK1 east indies navy battle math is wrong. That site’s calc is assuming the casualty order for UK is sub, ftr, cruiser, ac, in order of ascending ipc cost. You need to use https://www.aatoolkit.com/conflict and toggle sub, cruiser, and ftr to be taken last, so that the AC gets taken first. % goes from a coin flip to over 90%.

Thanks for the correction; I edited the original post. But a few things came up.

For the edit, I used AACalc again but changed the order of loss.

http://calc.axisandallies.org/?mustland=0&abortratio=0&saveunits=0&strafeunits=0&aInf=&aArt=&aArm=&aFig=2&aBom=&aTra=&aSub=1&aDes=&aCru=2&aCar=1&aBat=&adBat=&dInf=&dArt=&dArm=&dFig=2&dBom=&dTra=&dSub=&dDes=&dCru=&dCar=1&dBat=1&ddBat=&ool_att=Car-Bat-Inf-Art-AArt-Arm-Sub-SSub-Des-Fig-JFig-Cru-Bom-HBom-dBat-Tra&ool_def=Bat-Inf-Art-AArt-Arm-Bom-HBom-Sub-SSub-Des-Car-Cru-Fig-JFig-dBat-Tra&battle=Run&rounds=&reps=10000&luck=pure&ruleset=AA1942&territory=&round=1&pbem=

It gives 64.1% for attacker, using 10,000 runs.

I do use aatoolkit sometimes, though it uses 1,000 runs. Setting the OOLs as I thought proper, a few runs did consistently come in at over 90%, as you wrote. (Note - I used “1942” for aatoolkit for all runs in this post.)

I know AACalc has aggregate rounding errors. But a 26% spread seemed weird to me, so I looked around a bit.

In aatoolkit I kept “take unit last” on attacking sub, cruisers, and fighters, but turned “take unit last” off defenders completely. Attacker odds dropped from 90% plus to around 85%.

If the defending fighters die before the defending carrier, the attacker odds drop? That doesn’t sound right does it? I changed order of loss on AACalc; there taking defending fighters before defending carrier saw attacker odds rise by about 6%.

I know the attacking submarine changes things. But it does look to me like aatoolkit’s implementation might be incorrect, even though your point about order of loss was correct.

==

Besides that is another point that I didn’t mention in the first few posts. One of the big things about the East Indies attack and the bid is the projected UK survivors. There’s some complications, but it’s possible for UK to end up with a defensive fleet that includes a carrier and fighters (and possibly cruiser(s) against Japan’s counter of destroyer, two fighters, and bomber. It’s not the worst battle for Japan, but even trying it will reduce Japan’s options against other targets first turn, may well make further inroads into Japan’s airpower, and even the best-case scenario for Japan probably sees the UK fleet (including that expensive carrier) being able to slip out to the east next turn to safety, after which it can later shadow the US fleet.

Projected UK survivors is also an issue for any projected attack on East Indies. I took the liberty of running 4 attacking infantry against 2 defending infantry on aatoolkit and aacalc; aatoolkit returned 92%+, aacalc returned around 77%. Anyways, if you consider UK1-controlled East Indies, a UK carrier pressuring Japan into a more defensive position, order-of-loss for defending UK carrier being able to take fighters last (though if a cruiser survives I’d say even fighters before carriers would be all right in view of risks to Japan’s resulting position), again, the bid is important.

==

. . . so is aatoolkit functionally incorrect? Well, let’s look at a simple calculation; 2 attacking infantry versus 1 defending infantry. aatoolkit reports 91%+ attacker win with 6%ish defender win, aacalc reports 67.5% attacker with 27.5% defender win.

1 infantry vs 1 infantry, both hit 1/18; attacker only hits 2/18, defender only hits 5/18, both miss 10/18. We change the numbers as “both miss” results in another round of combat; result is 1/8 tie (all destroyed) 2/8 attacker win, 5/8 defender win.

2 infantry vs 1 infantry, both hit 11/108, attacker only hits 22/108, defender only hits 25/108, both miss 50/108. Again, we change the numbers as “both miss” results in another round of combat; result is 11/58 attacker win (with 1 infantry), 22/58 attacker win (with 2 infantry), 25/58 changes to 1 infantry vs 1 infantry.

We multiply 25/58 by the aforementioned results to get 25/464 tie (all destroyed), 50/464 attacker win (1 infantry), 125/464 defender win.

Attacker win with 2 infantry: 22/58, about 37.93%.
Attacker win with 1 infantry: 11/58 + 50/464, about 18.97% + 10.77% = 29.75% (Aggregate for 2 or 1 infantry winning = 67.68%)
Tie: 25/464 = 5.39%
Defender win: 125/464 = 26.94%.

So we should be looking for 67.68% attacker win, 5.39% tie, 26.94% defender win. Again, aatoolkit reports 91%+ attacker win with 6%ish defender win, aacalc reports 67.5% attacker with 27.5% defender win.

So is aatoolkit functionally incorrect? Well, if it just so happens that aatoolkit’s PRNG implementation is identical to 1942 Online’s implementation, then aatoolkit would be functionally correct. But if PRNG implementation in 1942 Online puts out functionally random numbers then it does look like aacalc is functionally correct.

@kakarrot1138

You may find that with the corrected order of loss that “coinflip” no longer applies. But I left the term in deliberately.

With the original quoted statistics, the odds were near 54.3% for attacker, after the correction 63.9% (though the points I made about UK survivors and the subsequent position, versus simply trying to “win” the immediate battle apply). You could say, with fairness, that 63.9% isn’t really a coinflip.

But when I write “coinflip” I mean “an unnecessary attack with no good contingencies in case of bad dice.”. In the case of a UK1 attack on Japan’s East Indies fleet, I can see how that attack could be considered “necessary” in terms of desired objectives, and you could even say that if the attack went somewhat bad that there would be some gain. But is it “necessary” in terms of winning or losing? Rather than playing conservatively and looking for an opening from dice results or opponent action, might not the UK1 attack absent bid itself possibly create an opening for an opponent off bad dice results?

That is how I think about the UK1 East Indies attack without a bid. If it goes fantastically, obviously great. If it goes off all right but not spectacularly, then Germany still keeps its Mediterranean fleet and that can be a problem if the Axis player is competent. If the UK1 attack goes badly you just shot yourself in the foot.

So what is not a coinflip attack? Suppose Germany has a stack on Ukraine and plans to capture Karelia this turn. Suppose USSR has a stack on West Russia that threatens Karelia, and that Japan isn’t in position to reinforce German-controlled Karelia with fighters. Now suppose Germany has an attack of 54% or so on West Russia, and that Germany does not expect to be able to defend Karelia well. Say also that Germany can move infantry up to reinforce on following turns. Now, Germany can’t use its fighters to defend Karelia, but Germany can use its fighters to *attack West Russia. It’s possible that the projections will be that even with bad dice that an attack into West Russia will weaken USSR’s stack enough that Germany will be projected to hold Karelia, which will mean two additional German units on the front. That is, in case of good dice Germany can break the game open, in case of neutral dice Germany expects to hold Karelia and after retreating moving infantry reinforcements up means Germany’s precious tanks won’t be at risk, and even in case of bad dice, Germany still expects to hold Karelia and protect its tanks after infantry reinforcements are moved up. Even with the German West Russia attack nominally having coinflip odds, the board position means Germany is likely to gain regardless, so the attack is not what I call “coinflippy”.

Contrast to UK1 East Indies attack. If it fails, it really is a bad position, right? For battles that don’t have some sort of contingency against bad dice, I prefer at least 85% to not think of an attack as “coinflippy” and even then if I’m playing accurately I still won’t do 85% if I think I can increase the odds to above that on later turns.

With all of this, I don’t mean to say the UK1 attack on Japan’s East Indies fleet without bid is bad. Different players have different risk preferences, and even absent that, if one thinks that an opponent outclasses them, they could try for a line that their opponent might not be ready to handle. But returning to the original post, I’d still certainly say that I thought KJF in 1942 Online was rather an uncertain prospect, certainly worse than at GenCon in any event.

@kakarrot1138 said in kjf no good in 1942 online:

I have no idea what you could’ve done to make aatoolkit show 2 inf v 1 inf as over 90% odds. It shows around 68% for me every time

I opened up aatoolkit again and ran 2 inf vs 1 inf and got 68.3%. But I originally got the 90%+ figure off aatoolkit that I mentioned after running the naval projections. Maybe that impacted it? You can see below that you do get consistent 90%+ for whatever reason sometimes; I spammed the “Calculate” button to get a few outputs so you’d see it’s not just a one-time thing.

https://imgur.com/a/TnBJY9f

(I turned charts off so there’s less clutter, but the charts showed the same thing as summed in those screenshots). Could it be aatoolkit has some bit of code somewhere that doesn’t get cleaned up properly that means improper results get returned?

@kakarrot1138 said in kjf no good in 1942 online:

Also, thanks for pointing out that Japan is actually better off taking off fighters before AC in that battle. I’m assuming it’s because if the AC survives the 1st round, a 2nd round sub hit can be assigned to it instead of the battleship, preventing it from dying before it can fire in the 2nd round.

That would explain it in theory, but in practice (and this is what aacalc indicates as well), I expect it not to work that way. I’ve run a load of projections on combats involving submarines; generally the odds of a submarine hitting are low, and though there especially with low unit count battles that counts for something, I don’t expect the odds of the sub hit in this case to increase the odds, when you’re looking at the low chance a carrier hits, preventing the low chance of a sub hit, balanced against the high chance a fighter hits. (Yes, the fighter won’t hit the sub, but the overall chances of battle drop if the fighter is selected as a casualty* before* the carrier? Weird? Anyways aacalc below:)

Attacking carrier first, defender carrier-fighter-battleship, attacker overall 64%

http://calc.axisandallies.org/?mustland=0&abortratio=0&saveunits=0&strafeunits=0&aInf=&aArt=&aArm=&aFig=2&aBom=&aTra=&aSub=1&aDes=&aCru=2&aCar=1&aBat=&adBat=&dInf=&dArt=&dArm=&dFig=2&dBom=&dTra=&dSub=&dDes=&dCru=&dCar=1&dBat=1&ddBat=&ool_att=Car-Bat-Inf-Art-AArt-Arm-Sub-SSub-Des-Fig-JFig-Cru-Bom-HBom-dBat-Tra&ool_def=Bat-Inf-Art-AArt-Arm-Bom-HBom-Sub-SSub-Des-Car-Cru-Fig-JFig-dBat-Tra&battle=Run&rounds=&reps=10000&luck=pure&ruleset=AA1942&territory=&round=1&pbem=

Attacking carrier first, defender fighter-carrier-battleship, attacker overall 68%

http://calc.axisandallies.org/?mustland=0&abortratio=0&saveunits=0&strafeunits=0&aInf=&aArt=&aArm=&aFig=2&aBom=&aTra=&aSub=1&aDes=&aCru=2&aCar=1&aBat=&adBat=&dInf=&dArt=&dArm=&dFig=2&dBom=&dTra=&dSub=&dDes=&dCru=&dCar=1&dBat=1&ddBat=&ool_att=Car-Bat-Inf-Art-AArt-Arm-Sub-SSub-Des-Fig-JFig-Cru-Bom-HBom-dBat-Tra&ool_def=Bat-Inf-Art-AArt-Arm-Bom-HBom-Sub-SSub-Des-Cru-Fig-Car-JFig-dBat-Tra&battle=Run&rounds=&reps=10000&luck=pure&ruleset=AA1942&territory=&round=1&pbem=

That is, if defender drops fighter before carrier, then attacker chances go up, not down. That’s what I expect - not from actually calculating it out, mind! - but just informally my expectation off “eyeballing” it.

In science, it’s always important to remember the dangers of researchers interpreting data to fit a projected result.

Of course, I haven’t actually calculated out this case (it’s a bit more involved), and even if the calculations did indicate something, there is the question that I mentioned earlier of whether PRNG outputs functionally approximate actual random numbers. So I’m certainly interpreting data to fit a projected result myself, and if it’s a matter of selecting a dataset that fits a projection, I’m doing that too.

But informally speaking, don’t you think it a little weird? If you’re preserving the carrier and losing a fighter instead, you’re dropping a 4 to preserve a 2. For that 2 to make a difference, the 2 has to connect, which in turn is preventing an opponent’s 2 from taking a shot. But the probability of all that, contrasted to a good solid 4? Attacker composition 1 carrier 1 sub 2 cruiser 2 fighter, defender composition 1 carrier 2 fighters 1 battleship? Doesn’t it even look like maybe a defending 4 that couldn’t connect on an attacking 2 and would instead start chopping away at attacking 3’s, would offset a defending 2, even if that defending 2 could possibly negate a submarine surprise attack? Especially if both fighters are set to expire before the carrier. If the attacker had a composition that mostly couldn’t be hit by fighters, then I could see more a case for it, but it’s just that one submarine.

I don’t expect a UK1 attack against East Indies (with no bid and with 2 UK fighters) has over 90% to win. I think aatoolkit may be wrong. Well, maybe not. Be nice if someone calculated it out.