A Comparison of Event-Based Runner Economies in ANR

Mr li is way worse than quality time, but quality time costs 1 influence a pop while mr li costs 0. Earthrise hotel and John masanori however are much better than Mr. li as well so andy has no excuse to play mr li

Ugh, itā€™s so mathy. How do I use it? Can someone show me how to plug the below econ/draw package into the formula?

Anarch deck/45 cards:

3 Sure Gamble
2 Day Job
3 Liberated Accounts
3 Cyberfeeder
2 Daily Casts
3 Earthrise Hotel
3 Steelskin

Thereā€™s certainly a reason I started with the Shaper econ ;-). Patience, and Iā€™ll try and get to the others. However, there will always be assumptions that must be made that might not be identical to those made in a real game. So always take this analysis with a grain of salt, and as one of the multiple inputs you use in the game.

It is a difficult balance between limiting the behaviors to something that is both tractable, and realistic. I think that the community at large can help quite a bit with this. One of the big problems is that we donā€™t really have a statistical understanding of how the corp and runner interact. Getting finer-grained information from OCTGN than who won, and how many turns it took would be a goldmine. How many turns are spent drawing? Getting money? When in the game? How expensive is ice to get past at various points in the game. We want our models of the game to be close to the actual game, and a good way to do that is to see what actual games look like from 10K feet.

A guy can dream, canā€™t he?

This economy is more complex, but with assumptions similar to those you propose, I believe we can get some useful information out of it. I sincerely appreciate everyone giving suggestions on how this can be more useful. Time is a finite resource, but Iā€™ll try to make continual improvements. For the time being, however, the next two articles are already done, so donā€™t expect immediate feedback.

Yes. In practice it would be equivalent to never trashing economy cards. This is obviously not a perfect assumption. However, I donā€™t think it is too far off from reality. At the very least, when you do trash economy, you know that the curves will go down.

This analysis wonā€™t help with that package. Youā€™ll have to wait for the Noiseshop article for anything that is similar to that. Do look out for an anarch econ article in the future, but, again, more based on event econ. Sorry! Hopefully in time, youā€™ll have your questions answered.

Again, thanks all for the feedback!

6 Likes

Thanks for the response. I wonder if itā€™s possible to eventually take this type of analysis and design a formula/program that takes a runnerā€™s econ package as input and spits out an approximation value of credits/cards per click. It would be limited and never able to give a true value, but it may be useful, or at least fun to see how close we can get to the sweet spot of 1.4

Something it occurs to me for you to do:

Do a similar analysis in which the corp draws economy and ice and installs ice evenly across centrals and (optionally) one remote. Track the rate at which the average cost of a central (optionally remote) server access grows, and track the times at which the runner first requires a given breaker.

This allows you to make a rough estimate of an interest rate, which should help to reason about what, exactly, tempo hits mean for a runner economy.

One can even dream about approaching the question ā€œis this access worth it?ā€

Fascinating analysis and great work, you can see a lot of effort went into it, Greatly appreciated!
I second the interest in something on GitHub if possible, could be fun to tinker try out other Econ engines, although running 4000 simulations could add some more time to deck building.

Your 10% worse plot is interesting, I am wondering in addition what is the relative distribution for each Econ package over the 4096 games? I.e. Are all of the distributions denser close to the average or are some Econ packages more feast or famine?

Maybe itā€™s just me, but I keep scrolling up and down between the credit efficiency plot and the draw efficiency, it might be nice to have a plot where you could look at them together. Would it be possible to do a scatter plot where credit efficiency/click is on one axis and draw efficiency/click is on the other? You could do snapshots for different clicks, i.e. At click 5, click 10, click 15, etc

Looking forward to the next articles

1 Like

I agree, this would be huge because for each engine and you would know the average game lenght in clicks and the clicks distribution wrt a specific corp archetype. Essentially you could then learn how the simulation results are affected by the interaction with the corp and consequently improve the former.

Good Job! Hereā€™s couple comments for starters:

Whatā€™s going on with ProCon in ā€œ10% percentile of gamesā€ graph? The credit/click ratio=1 after you draw ProCon, yet this level is reached only at ~22 clicks, and the curve is flat between 15-20 clicks. Even without a mulligan, you should see a ProCon within the first 9 cards (of which 5 are in starting hand) in 50% of games.

As a related note, 4096 games is not really enough to draw reliable conclusions in this graph. With a large number of games, the distribution of gained credits in each click bin should approach Gaussian (or be very close to one). The ā€œ10% of averageā€ level is at 2.15 sigma, that is, roughly within two standard deviations of the average value. The number of samples lying within 2 sigmas of the average is 95%, so the statistics in this graph are based on ~100 games (0.05*4096=205, but half of the games are above the average).

3 Likes

A really good article, even if I think itā€™s almost a pure Shaper article.
Looking forward to any follow-ups!

Would be interested in hearing otherā€™s thoughts about drawing into econ in general. My gut reaction would be that I often draw to get other things with the side-effect of getting econ, not specifically drawing in the hope of getting econ. That said I have mostly played criminal / noise decks and can see why shaper builds could be different.

Enjoyable article, thanks for posting!

One nitpick: as I understand it your analysis charges both Sure Gamble and Dirty Laundry 1 click to play the card.

This actually shortchanges Dirty Laundry significantly - Dirty Laundry also gives you a clickless run valued at 1 click. The majority of the time, Runner spends no additional clicks to gain the money from Dirty Laundry because they were planning to run anyway.

Iā€™d at least give Dirty Laundry a discount factor - if 25% of DL runs are just on some unprotected server for the credits, and 75% are valuable runs then DL costs 0.25 clicks to play on average.

I agree this might be cool, but I worry it would be misleading. Iā€™d encourage everyone to not simplify this to a headline number of x credits/click (even though I do that in the heading solely for the tldr crowd). The shape of the curve is in some ways just as telling and just as important. It is important to understand the efficiency curve in additional to the eventual efficiency, and match that to your deck. This is doubly so for the trade-off between credits and quality draws.

I also plot (though I didnā€™t publish) the standard deviations. It was surprising to me that that they were relatively consistent and around 5 credits for all economic packages. This would imply they are all somewhat dense. The outlier here was the procons package that had an increasing standard deviation over the number of clicks. That is explained by the 10% graph. I cant think of a good representation for the distribution as each click represents a histograph. Thus, why I focused only on that worst 10%.

Fair enough! The 2nd and 3rd article will probably have this problem as theyā€™re already effectively done, but Iā€™ll consider ways to avoid this scrolling after that. I do have a representation that gives the ā€œresources per clickā€ which is credits+quality draw. It loses the breakdown for each independent resource, but provides higher-level efficiency information. Iā€™ll post these in the future. A scatterplot would look awesome, but I worry that it would not be useful for directly comparing different engines as the dots would hide each other significantly.

This demonstrates inconsistency in the games. If you donā€™t draw into procons for a long time, youā€™re getting 0 clicks/credit. The simulator draws until it finds procons, then uses it. This is not 50% of the games. This is the worst 10%. This is one of the reasons that a simulator is useful as it can give the worst 10% rather than just typical measures such as average.

Great note, thanks! I agree that 4096 games is probably not enough. To be honest, I chose that number as it provides a reasonable trade-off between time to run the simulation, and having a high enough quality of results that I can draw conclusions with a reasonable level of confidence.

Iā€™m not convinced that each clickā€™s distribution should approach Gaussian. The probably to draw specific cards is Gaussian. Their impact when you draw them creates huge impacts in subsequent clicks that wonā€™t look Gaussian as they will essentially be based on conditional probabilities. The procons graph is the best (accurate) indication of this. That said, Iā€™ve proved nothing, so you could be completely right. Statistics != intuition.

The interesting thing to me about the simulation approach is that it actually lets you tease out those approaches that are not, in fact, gaussian, and investigate the tail (which actually is another argument for running more samplesā€¦but it is hard to know how many more). Translating to normal person speak: we can investigate the consistency of decks. This article (and the next 2) investigate consistency for event-based econ as we donā€™t have a good way to look at them currently. Sometime after that, we can also study the consistency of playing all three breakers. Drawing multi-access. Or conditional effects such as playing a sentry breaker while staying above a credit threshold (to avoid sea source).

Thanks all!

2 Likes

Great point. There are many games where DL is played simply as an econ card, but that is obviously not its potential. To tease out the case where DL is played within an actual (non-econ focused) run, the click to play it would have to not be measured as a resource click. This is trivial to do in the simulator, and I just blanked on doing it. Great suggestion.

[quote=ā€œYCombinator, post:31, topic:2689ā€]
The outlier here was the procons package that had an increasing standard deviation over the number of clicks. That is explained by the 10% graph. I cant think of a good representation for the distribution as each click represents a histograph.[/quote]

That makes sense as it can take a long time to draw the ProCon during which time youā€™re making 0 credits, so the distribution in net credits/click is probably quite wide.

It took me a while to figure out why your ProCon curve in the 10% graph looks so weird. The worst case scenario really is that you donā€™t draw ProCon for a long time, so you should stay at 5 credits. However, both ProCon curves start at 0 credits. The reason for this is that youā€™re taking the (average of) minimum number of credits over all games, which includes the ā€œbest possibleā€ cases where you play ProCon at turn 0 and drop to 0 credits.

Very very interesting article! Iā€™m looking forward to the next instalments.

Iā€™ve been running a shaper economy of pvp + procon + evts lately. Itā€™s a bit of a bummer that itā€™s missing from the analysis. It gets me insane credit amounts, but it is very slowā€¦ Iā€™m playing Kit, so I can leviate the tempo hit somewhat by depending on only one breaker type for most of the early game though. Iā€™d be interested to see how much worse my package performs in comparison with regular pvp + draw with regards to draw efficiency.

I have to say im quite disappointed you left out one of the most solid Econ engines in the game " Hard at work" with 3 copies of this in play you get 6 credits a turn with no more cards needed. Add in one Wyldside and you get 2 cards as well.

so you get 6 credits and 2 cards per turn. Now thats Efficiency!

4 Likes

Some questions

You didnā€™t explain what these decks actually ARE, are they 3 of each of the cards + the rest being quality draw cards, or is it more like 5 lucky find 7 sure gamble 5 dirty laundry 3 quality time 3 Diesel
If it is the first, then I think this analysis would be improved by adding more econ to the packages outlined something around 6 draw cards and 18-20 econ cards is the right amount.

What cards of the following would be easy/hard to compute value for?
Day Job (4 Clicks to use for 8 gained)
Liberated account (5 clicks to use for 10)
Armitage codebusting (7 clicks to play for 12 credits gained)
Magnum Opus (install and mash the opus button)

@edmund_blake_nelson Thanks for the questions! A redditer pointed out I didnā€™t specify what it meant to have those cards in the deck. I dropped the ball there.

Normal deck building rules are followed. 3 of each of the credit, and 3 of each of the draw cards. Many prepaid decks include only the 9 credit economy cards, and I wanted to roughly emulate what they do.

In the second and third article Iā€™ll look at Day Job. It is as difficult/easy as Lucky Find.

I donā€™t include these currently. This is not because they donā€™t have their place, but because they are easier to understand in terms of credits/click. They arenā€™t trivial to add, but not too difficult either. The simulator just needs to understand when they are spent. These are probably the furthest out in the timeline.

I do have this working, but the analysis of it must be more nuanced as realistically decks also need to draw cards. Iā€™m working on this now, and it will probably make the 4th article.

Thanks for the questions!

2 Likes

When I exhaust the current set of articles that are finished/planned, Iā€™d be happy to take suggestions. This would be a trivial one to crunch, so please do suggest it at that time. Currently it just doesnā€™t fit well into any of the outlayed articles. I didnā€™t include it in this article as there is a limit on how many lines you can include while people can still easily discriminate between them.

1 Like

I may be dumb, no problem with that : I donā€™t understand why thereā€™s no cost in playing procon (or MO, or Kati Jonesā€¦).

Like saying Sure Gamble are Easy Marks + 1ā€¦ Itā€™s not. Thereā€™s a threeshold.

Obviously for one shot event the threeshold get refunded + benefits, for procon thatā€™s not the case : so for these actually horizontally their curbs should be move to -5 for procon for example.

My problem is all data starts when cards are refunded.
When you play Sure Gamble, you gained 4. When you play procon, you paid 5.

So thereā€™s something like 9 ā€œressource clicsā€ between a sure gamble and procon : so what are odds to draw a second sure gamble in those 9 clics ?

The cost to play procons is included. Thatā€™s why the credits over clicks graphs dip at the start for the procons economies. When looking at the mean, youā€™ll never see a 5 credit dip. The curve is the average over many runs. Some of those runs havenā€™t found procons yet, thus are still at 5 credits (or more in the procons + evts curve). Long story short, these costs are taken into account.

However, I completely agree that this doesnā€™t tell the whole story with respect to the cost to play cards. In real games, you might find yourself at zero credits with SG in hand. This analysis doesnā€™t measure that. It only provides information about some startup-cost, and efficiency. It doesnā€™t try to analyse all aspects of all cards (yet ;-)).

To look at the cost of cards, there are a number of ways to look at this, and I really donā€™t know what is best. Some conversations with fellow netrunners have helped to clarify this somewhat, but more groundwork is necessary to understand how to usefully look at this problem.

If you play procons, and have SG in hand, you will have to click 5 times to be able to play the SG. This is taken into account. However, once the amount of credits you have is large, none of that really matters anymore. Looking at the first five clicks is in some ways most informative.

Long story short, I do consider the costs of event cards. However, as there is no interaction with the corp, or other costs, after a number of clicks, and the runner has enough credits, then the play costs of many cards become irrelevant. This does not perfectly reflect reality, so should be taken with a grain of salt. However, you can still get some information about startup-costs, and card vs credit efficiency of different packages. Iā€™ll try and make it reflect reality more in the future, but for now I hope that it provides more information than we had before, and informs some deck-building decisions.

Thanks for the response, and for reading the article!

1 Like