There has been a lot of discussion about whether we should let you know how the AR works or not. It looks like the Marquis spilled the beans on the main issue (that of personal combat history). Trust me, the auto resolver never looks at your personal win/loss history.

This rumor came from the way that we designed the AR in Skyrates 1 before Thradius joined the project. In old Skyrates, the AR would look at your win/loss %age for a specific combat level and roll a die based on that. This was messy for a lot of reasons:

1. It was horrible for the server, which had to do a query on every combat a Skyrate had ever fought and then filtered by combat level. Trying to make this only the more recent combats still wouldn't make the query any better.

2. It didn't make sense a lot of the time. Because it was based on player history, it didn't take into account anything about recent upgrades, plane purchases, etc.

3. Every time you reached a new combat level, it would have no data on the level, so you'd still have a 50% chance of losing regardless of your previous record.

4. Because there was no knowledge of planes, a player could build up a great combat record and then buy a really big cargo plane (which should be terrible at combat) yet still win AR combat.

It was very exploitable and really did a bad job of actually resolving combats at all correctly once anything changed about your Skyrate or her plane.

Thrad's system is much, much better. It takes into account your Skyrate's plane, your Skyrate's plane upgrades, and your Skyrate's combat skills. Also, rather than just going off of a simple combat level number, it actually generates planes to fight against you and does several volleys back and forth. It's a much, much better system that is actually less of a hit on the server than the old one.

As an aside: I say above that all of these things are your Skyrate's because in our design, you and your Skyrate are not the same (which is why he/she calls you "boss" all the time). When you're in a manual combat, you're taking over the controls from your Skyrate and teaching her how to fight, which is why she earns CP from the fight which can be spent on combat skills.

One of the things that I feel could be improved is the decision of when to flee. Right now, the system waits until your Skyrate's armor is down to 1 point before fleeing regardless of which plane you have. The system includes the ability to modify this, which would change the outcome of the AR considerably.
As is mentioned above, once you lose one combat now, it's followed by a string of losses with a few flees (because each combat starts with only 1 armor). We're considering changing this from 1 armor to 50% of your armor. That way, once your Skyrate's plane got to 50% armor, your Skyrate would start trying to run from combat. I believe that this would result in a change from a combat log that looked like:

Win, Win, Win, Loss, Loss, Loss, Flee, Loss, Loss, Flee

to one that was more like:

Win, Win, Flee, Flee, Flee, Flee, Flee, Loss, Flee, Loss

Armor would still be depleted when fleeing, but the higher amount of armor to begin with would make fleeing more likely.
Another thing that we would really like to do given the time would be to allow players to set how chicken (or prudent, if you prefer) their Skyrate is by setting a %age of armor that will trigger fleeing in AR combat. This would take changes to the interface, client, DB, and server, so it may not happen any time soon, but it is something that was originally planned, and it's still an idea that we like a lot.

-- Archon

So, it sounds like spending CPs on Skill Tree items that you never intend to use in a manual combat (such as Sideslip, for example) can help make the AR a better fighter. Is that correct?

Correct.

Boosting sideslip is fascinating, but still a bit silly. Having the game check against what the devs think my mix of skills and upgrades ought to produce is both static where the player is dynamic, and quite unrealistic (to the extent that's valuable).

Listen, I'm rolling around in a PK/CK Hades with Wings A and Mad Props. My Maneuver, Eagle Eye, Precision Fire, Crackshot, Gumption, Immelman, Split S, and Barrel Roll are all maxed. Mid-30s combats aren't, to omit Kyra's earlier modesty, even the middle of the curve - they're the beginning. Why? Because with that setup, my plane can sit in the middle of a storm of mid-30s level pirates with the trigger depressed and come out with full armor. Or at least, it can if I'm doing it - if the AR is, it'll lose somewhere between 4 and 6 combats per hour.

This setup isn't a radical situation for a combat player; I did it in a month of not-particularly-focused playing.

Changing the flee roll to start at half armor or some user-defined number would be nice. But this wouldn't solve the underlying problem, which is that the current coding for the AR that takes me to any armor level below full does not take seriously the actual value of the upgrades and skills.

Let me reiterate: a rock placed on top of the space bar would do far better in any given combat than the AR. When the code is losing out to a terrigenous compound of clastic sedimentary elements, the code is borked.

!!!!!!!!!!!!!!!!!!

That's good to know.

/me applauds.

I suggest people make use of waypoints to break up their long legs into sections of safe and risky flight. If your AR can handle flights fine up to 90 minutes, then put around 90 minutes of risky flight into your route and make the remaining portion safe flight. This does wonders to the results you will obtain from the AR.

Creight Lersing

Hahaha, Zabrak! is my hero.

Two things: first, it's not the length of the flight but the combat level that's the problem here. My AR can do up to level 28s maybe just fine, even on a long flight to grotto. But it can't handle even a single level 35 fight. Which means if I want the increased earnings of the higher level combats, I must be there to fight them myself, or the game will punish me horribly. The alternative is to always fly at AR-safe levels, and give up entirely on the extra cash I could be getting whenever I'm around to fight manually. Combat earnings are paltry enough compared to trading already, this just makes it even worse.

Second thing: waypointing safe/risky works for cutting down the number of combats, but have you noticed how tedious it is? Queueing up 8 legs is bad enough without essentially doubling the number of clicks I'd have to make per leg. On the other hand, a "replicate queue" button could fix that issue...

If I can plan my flight, I'll just do the safe route combined with a little island hopping to cut the length down to under 4 hours per leg, and without piling up missions. The AR is reliable enough that it can handle three or four mid-30s combats. Sure, you lose a bit of income in the form of the AR not grabbing what should be easy 2,500G combats, but it works out for me to only be 15,000G per hour, which is not worth getting steamed about in the long run.

The problem comes when you can't plan, or more accurately, you plan to be there for most of the flight or the half hour before the landing, but those plans aft gang agly - situations may include:

- the server goes down
- a traffic jam on the way home
- your SO surprises you with something scandalous at the door
- you find yourself in a comfy chair being poked with soft cushions and you doze off
- a special on 100 tacos for $100 down at the Tacomat
- a Doctor Who marathon on TV
- Animal care specialist
- TV/VCR repair
- or get your degree, you can major in business management or accounting

A combat pilot is going to schedule some long legs that he or she thinks can be fought manually, and naturally will plot them as risky because that's how they've customized our planes and skills to make a living. The AR in this situation becomes your worst enemy. Once you hit the "start flight" button on a long risky leg, you live in existential dread of what could be delightful and worthwhile distractions of life.

The whole point of casual play is that interruptions from the game shouldn't detract from gameplay. Yet having a sub-par AR does precisely that.

P.S. I was a bit of a jerk in my last message. I don't want to seem ungrateful - this is a fantastic game and the devs should have a thousand parts of praise heaped on them for every part scorn.

I suspect that the existing code might be able to handle many of these concerns if it were only slightly tweaked, mostly in how it handles the possibility of death. For instance:

1) Objective toggling. During a fight, allow the AR to switch in-between volleys from "trying to kill everything" to "trying to flee". Presumably the code (even now) rolls different dice based on whether it's trying to flee or trying to win; if not, it oughta. But it pretty clearly doesn't change its goals mid-fight, whereas again it oughta.

The criteria would obviously be essentially whether it was now looking like long odds or not. This could be done on some kind of abstracted odds basis (a running averaged slope of the my-armour-lost-to-total-enemy-armour-remaining curve would do it), or it could be done in more detail; {static}, one way to do it would be to let the AR do a lookahead search by "fighting" the combat say five volleys ahead, as its best guess as to how the combat will go. Switch modes to attempting to flee if that lookahead shows it crashing and burning in that time.

Distill this behaviour down to a single scalar, call it courage. Lower courage equals larger chance of quitting with less payout; with the lookahead, that would mean check more often and look ahead farther, while with the slope check it might just move the mean of a curve (odds of quitting at slope X). Make this a config option (how courageous you want the AR to be), and/or once Gunners exist make it their equivalent of the Nav's safe/risky switch.

That would, I suspect, cause it to have behaviour much closer to what players themselves do during combats. If it looks hopeless, they bug out. We all assume, from the Flees in the logs, that this is currently done at some level... but it sure doesn't look like it's trying as soon, or as successfully, as it should.

2) Check the high end. Fix the odds of fleeing for very fast planes who can, when at the keyboard, reliably escape anything. And the odds of scathing for planes which can sit there without maneuvering and come away unscathed. Basically, run the numbers a little more carefully at the high end, since they don't seem to be reflecting the equivalent inputs.

Simultaneously, however, start letting the system generate enemy planes with the same kinds of upgraded shinyness that players use. I don't know for sure, but it doesn't look like we're seeing sensibly-upgraded NPC planes, in combats of such level that this would be appropriate of course. If it did, then that hyperfast (but not yet triple-kitted) Inger is still gonna occasionally meet up with something that can keep up, at least maybe.

3) Make Negotiation skill matter to the AR. Again this would require some kind of option, or toggle, or limit switch - perhaps based on the expected losses (not just lost cargo - calculate an estimated value of unhappy crew and inability to win further fights this trip), divided by the expected average number of combats remaining to be faced. Giving the AR "inside info" in this matter isn't really a problem since the players don't see it; all they see is that the AR chose to bribe X number of times, at a cost of X.

Reassure players: Make it switch-offable. Explain that it does so based on the odds that it thinks it would have lost you more money than that, in the form of lost cargo etc, if it had AR'd the fight normally. Explicitly say that the AR only gets odds of a bribe, not a guarantee that it will do so past a certain cutoff. But the odds of it doing so would definitely depend on how much the bribe was going to cut into your profits, as well as on the best-guess assessment (see #1) of whether victory, or flight, were possible.

In fact, in the lookahead model, one way to do that would be for the lookahead to first transition its goal from "win" to "flee" if it foresaw a death in its future, and then from "flee" to "bribe" next time it checks, if the apparent odds of dying (instead of getting away) times the cost of doing so appear to be higher than the bribe price.

----------------

If the devs want to contact me, I can probably come up with some algorithms to put numbers behind the above. A lot of it can probably be formed from preexisting calculations which don't change much, such that the at-the-time stuff is kept as efficient as possible. The biggest risk is that it might achieve an AR which resolved somewhat slower, but I don't know whether that's an issue (because the arriving-at-skyland lag is 80% AR math already) or not (because the lag is at most 5% AR math, mostly it's image loading or what have you).

I helped Ad Astra Games get the math to let players lead their targets with railguns, and that's a board game - the players have to do the math by hand; I'm sure I could contribute to these estimating processes if the help would be useful.

We discussed it, as I recall, and it sounded like everyone was more or less on board in spirit. But to my knowledge it was never coded.

-PL-

I think double-kitting should be possible, but not without penalties. The kits should interfere with each other in a way that, if you were to get all three kits, that having three kits would revert your plane to 'stock' basically.

edit: edited for clarification

yeah i certainly agree there should be restrictions on it of one sort or another, though it remains to be seen what will work the best in terms of balance and complexity.

-PL-

Just another example where sucky AR can hurt - I had a combat leg planned, with high missions loaded. I was there to fight it, and I was swatting down the mid-40s combats with ease, when I got hit with a "not logged in" snowcone coming out of one of the fights. Lo and behold, the AR loses it, and now I've lost cargo, crew morale is down and my (previously full) armor is hosed. Not fun.


The 50% thing sounds like a really good idea, specially since with a halfway decent mechanic your plane should be able to recover some armor between fights as well. Ideally this could make the string of flees completely loss-free. Having a user-settable percentage would be amazing, but if it's too coding-intensive even just switching up to 50% would help a lot in the interim.

Okay, go set your autoresolver cowardice now.

what happens if I set it to 100?

Then we buy you a pretty dress and start calling you Mary.

On my trader, Rarkre, since I set it to 100%, all AR have been flees. Nothing dangerous enough to force a loss yet.

Awesome, thanks! I just saw that I can disable the trade protection also. Sweet.

(Everyone watch as I immediately screw myself over with terrible trades and then come whining for reparations.)