Well, I don't see any reason to think that you have to use the 999 trick if you never try to screw the AI.
I mean, all these deals are ways to exploit the AI - if all you ever do is make trades and not try to screw the AI, you never have to check the 999.
That might be fine if the "screwing" wasn't so powerful in the first place. If these kinds of agreements are allowed, there's no we we could keep from doing it if we want to stay competitive. So if it's allowed, no one will keep from screwing the AI.
Conceptually, you could run into a problem like this for a turn or two with the AI - say you buy steam power for 600 gpt then get electricity for free but screw up and don't sell it back to the AI until the next turn, for 580 gpt. Then there will be a turn where the AI might be running a big deficit, but I don't think anyone would see that as an exploit.
Indeed I wouldn't. It's not the AI's deficit alone that makes it an exploit, you the player must gain from it as well to make it exploitish. In this case you are still paying 600 gpt x20, so you're not gaining anything "for free".
I see very little difference between these two trades. In some ways the first trade, if limited, is LESS exploitive than the second. In the first, if limited, the AI gets a Tech and since I don't think they'll ever trade more for a Tech then they think its worth, they'll pay what they think is equal value, or less. And the player, in exchange for a possible trade rep hit, has gotten more of what the AI values the Tech for in exchange. In the second the AI gets NOTHING, and the player gets a free tech. This just manufactured an amount of free gold equal to the player's cost to research the Tech. Since we have always allowed the 'buy a Tech from the AI for gpt, then cut the payment' tactic, getting 'free gold' is not a good enough argument to ban the first trade.
The problem in this case is that the AI is so incredibly stupid that it will buy whatever it can afford (or sometimes not afford). No they won't pay more than what they think its value is, but the AI simply cannot make such valuations. They will happily pay tons of gpt for Music Theory even when Bach's is completed elsewhere. And by taking their research gpt (which is effectively what you do when you "free up" more of his available gpt) he will also have lost the ability to research something else. My previous comment about door salesmen and senile old women still stands.
Also, turn it and see what the AI loses. In the first case the AI loses nothing (except the ability to trade that tech to you), in the second case the AI loses a lot of gpt, potentially crippling it.
That's the first step of the manipulative gpt trade, and I thought everyone feels that part was OK. It's really the 3rd step, when the gift/trade of gpt is broken, where the exploit can occur. The problem is there are lots of ways to get from 1 to 3 without being exploitive, or that aren't covered in any of the rules discussions; even the spirit of the law may be clear to most but vague

to others. So I'm proposing a comparison that looks just at Step 3, and if we can come to a general agreement on how much of the total gpt available to the AI can be snagged, we can avoid causing a negative gpt and crippling of the AI.
It is really the combination of the steps that creates the exploit. I don't think cutting a luxury deal is an exploit, but it would be if you received bloated gpt by cutting it. That's what I mean by speculation - the combination of artificially increasing the AI's available gpt and then trading for it. If you don't cut the trade route or alliance, there's no speculation, so the first step would be ok. If you haven't bloated the AI's available gpt, again there's no speculation, so the third step would be ok.
I think this is fine as long as the AI doesn't pay more than it thinks the Tech is worth to it (and I don't believe it ever does that), and you don't take more gpt than the AI can afford. Checking the 999 number and applying a generic ratio is not a sure thing, but I believe it can generally prevent the negative gpt, AI crippling situation, and give the player a guideline on how far they can go.
See again the comment on stupidity in AI tech valuation.
At the moment it looks like there are two parties here: the ones who want to ban Emsworth Agreements altogether (as described by Niklas) and the ones who just want to ban those Emsworth Agreements that criple the AI "too much" (as described by civ_steve). I would like to add a few more ideas to the current discussion (let me use the abbreviation "EA" for Emsworth Agreement in the following):
- Some have said, that if we allowed EAs, then everybody, who wants to be a competitive player, will need to use them. Well, what is so bad about that? These days everybody, who wants to be competitive, needs to know about the "tech broker strategy", about 4-turn settler factories or palace jumps and many more strategems. EAs would enrich our arsenal of strategic weapons and would make the trading and diplomacy aspect of the game more interesting/colorful. If EAs are well documented in the strategy forum, everybody can use them. I don't have a problem with that. This would be a point in favor of civ_steve's idea.
Actually, that's a very good point, but...
However I don't like the idea of 999-checking, calculating ratios and determining how much I have to pay back to keep a clean consciousness...

But if that can't be avoided, so be it.
... I really don't like this part of it. I simply don't think we can find a limit that would be suitable, since even at low ratios you can potentially create incredible amounts of gold, and cripple the AI.
I did some digging on the 999 trick (so yeah, now I know it, how quaint) and it seems to me it contains
a lot of cans of worms. First of all, just so we're clear what we're discussing, as Marsden pointed out the value listed is the Big Green Number (BGN). That means the total of all beakers, gold, happy faces
and corrupted commerce that your towns generate each turn, plus income from taxmen (but not clowns and scientists), gpt from other civs and Wall Street interest.
A first thing to note about this number is that it is not fixed but depends on the slider settings. A civ with lots of libraries but no markets has a higher BGN running a high research than a low research. The lowest value is of course when running 100% luxuries since there are no multipliers for that. Sampling a few of my own games, the difference between 100% research and 100% taxes varies 5-15% of the (highest) BGN (20% of the lowest). Technically it could go higher, but that's unlikely. But considering a Sid AI who has Literature but not Currency (and builds libs ridiculously fast), the value could potentially vary close to 20% (of the highest BGN).
Now, what would be a fair number for the AI to have left? Again sampling some of my own games, the value I would have available for trading, that wouldn't make me start losing units or make all my towns riot, varies greatly. In some games it is as high as 70% (0% lux rate, high income from other civs), in others it's as low as below 30% (high lux rate, lots of units OR a large gpt deal to an AI). And these are clearly not extreme values since I just sampled a few saves I had at random. And furthermore, I rarely (never) play as extremely as the AI does, building all improvements it can in all towns and building lots of units as long as it can afford them. And paying gpt deals to the player or other AIs, since those expenses are counted after the BGN. A redeemeing factor though is that the AI rarely uses a high lux rate, preferring clowns for some reason (at least that's my experience).
So what
would be a fair number? I think 25% of the BGN is too much, given that even I in my fairly moderate unit-focused moments can go below 30% (and that's of the
lowest BGN, meaning it could be more if we compare to the visible BGN - it just takes one library for the values to differ - though in my sample those values were equal). And it just takes one gpt deal paying off most of the available gpt to some other AI for the actual ratio to drop dramatically, closing in on 0%.
So to sum up my thoughts on this - if we are to set a limit, it has to be 0%. And what would be the point of it then? Then we would outlaw plenty of "fair" cases, like PaperBeetle's Byz-Aztec-Japs scenario, or when an AI stupidly puts himself in the situation by signing peace and breaking an MA that you have would preferred to retain. So I don't see how the BGN/999 trick could help us out here.
I stand by my previous opinion, that we need to outlaw the practice and not try to impose limits on it.
Please consider the following scenario: It is perfectly legal to buy an expensive tech for gtp and then declare war on the AI (or provoke it into declaring on you). Let's assume you are planning to do this, and you need a certain tech very badly, but you are not yet in a position to declare immediately, because your defensive troops are not yet ready. So you buy the tech now, pay 200 gpt for the first two turns, move the last remaining troops into position and then declare at the end of turn 2 (so you can at least save the remaining 18 payments...)
That would be perfectly legal as well, wouldn't it? But in the meantime your victim might have made a couple of deals with other AIs, for example spending your 200gpt for a few other techs. So when you declare, you might drive him into negative spending!
Yes, this certainly becomes a problem. But at least a redeeming factor is that the "free" gold won't go to the player but to other AIs. But of course you could then set up gpt deals with them and get it via proxy... So yes, I agree this is a problematic case. Maybe we should outlaw it as well? Meaning that if you want to declare and cancel a gpt deal, you had better do it immediately.
What I'm trying to point out here: I don't see a big difference between a) canceling a deal in which you give gpt to the AI while another deal is active in which that AI gives gpt to you and b) canceling a deal in which you give gpt to the AI while no other deal with that AI is active. In both cases you are stealing the same amount of money from the AI! And in my opinion buying an expensive tech and then not giving them the money which is rightfully theirs, is more "morally reprehensible" than giving them money for a MA and then no longer paying it when the MA is no longer active...
I do see a big difference, in that you don't actually take something from the AI in case (b) except virtual (promised) money, not actual money as in case (a). But as you pointed out above, that can indeed be bad enough, if he uses that virtual money to sign deals with others.
PS: I think the root of the entire dilemma is the fact that deals are "bound together" to a single unit. This looks like a bug to me. For example what I don't see is: if you buy a tech from the AI and pay with a resource and an amount of gpt and then cancel the resource part of the deal, why the gpt part of the deal should be canceled too. If we could fix that and have every element of a deal been treated as a single independent item, the problem would be gone. Do you think we could get the source code from Fireaxis? After all they are no longer fixing and selling it. And from what I've seen on this forum here, it appears that some of you have really close contacts to some Fireaxis developers...?! We could make a patch and require it as prerequisite for the GOTMs.
I doubt we'll get the source code or any some such. But even so, it's not really a bug. For instance if you were paying gpt+luxury for a luxury from the AI, and cut your own luxury, do you still think the whole trade shouldn't be cut?