QuantANR: Analysis of a Changing Meta

[SneakySly]

Originally published at: QuantANR: Analysis of a Changing Meta - StimHack

Discuss the latest StimHack article by @YCombinator here.

[Dragar]

I’d be really interested in seeing how much of it is Day Job, and how much of it is just adding another economy card. For instance, if you add copies 4-6 of Dirty Laundry or 3 Daily Casts, how does that compare to Day Job (in PPVP and without)?

[moistloaf]

“Or will its card art make it tier one, independent of any other factor?”

accurate.

another very insightful article. stimhack is lucky to have a contributor like you @YCombinator

[endgame]

You could summarise the warning paragraph as “All models are wrong. Some models are useful.”, which looks like a quote or paraphrase from a George E. P. Box.

[YCombinator]

@Dragar

I’d be really interested in seeing how much of it is Day Job, and how much of it is just adding another economy card. For instance, if you add copies 4-6 of Dirty Laundry or 3 Daily Casts, how does that compare to Day Job (in PPVP and without)?

Great point. Quantity vs. quality seems like a great topic for the future.

@moistloaf

another very insightful article. stimhack is lucky to have a contributor like you @YCombinator

Thanks! I’ll provide them as long as people are interested.

@endgame

You could summarise the warning paragraph as “All models are wrong. Some models are useful.”, which looks like a quote or paraphrase from a George E. P. Box.

So true. In fact, that probably needs to go into future articles

[kenobob]

My hunch is it’s the power of another econ card. Lucky find is more efficient than Day Job, yet the curves are about the same.

If we got a Restructure for the Runner, which I think we should, that would really open up the possibilities in prepaid decks.

With all that said, it seems these analysis are kind of an ode to prepaid Kate. I haven’t been able to glean from these the benefits of static econ like Cyberfeeder, Stealth rigs, and what builds fit best with drawless econ like Kati or Magnum Opus (this may have been covered in the last article though).

Also, probably beyond the limits of these simulations, is the economic benefit of things like Datasucker, Yogasaurus, and Ice Destruction.

[Ajar]

Day Job vs Lucky Find has pretty significant playstyle implications for Prepaid Kate. That deck loves to Lucky Find with VoicePAD credits and then run with an SMC on the board. In addition, sometimes you just have to wince and SOT an LF to get credits immediately. Neither of those is possible with Day Job unless you also add some means of gaining clicks, which I have a hard time seeing deck slots for in Prepaid Kate.

Edit: Still, nice work! Always interesting reading.

[bblum]

Influence in kate is also a real concern. Calimsha’s inf spread is 3 lucky, 2 stimhack, 1 parasite, 1 mimic, 2 legwork. What of this can you cut to make room for being employed? I think -1 lucky, +2 day job is the only justifiable change.

[mediohxcore]

@bblum

Influence in kate is also a real concern. Calimsha’s inf spread is 3 lucky, 2 stimhack, 1 parasite, 1 mimic, 2 legwork. What of this can you cut to make room for being employed? I think -1 lucky, +2 day job is the only justifiable change.

You could cut one Lucky for 1 DJ if you needed an influence for whatever reason. Maybe you could play 1 Mimic 1 Femme or cut a Legwork for a Medium or something.

[bblum]

Yeah, the deck has always been rich enough to install femme, it just has never needed to. But it will once people wise up and start running Susanoo in RP.

[Nordicstrike]

The runner still has an infinite hand size right? This assumption may give decks that draw a bit too high quality draw efficiency, since “quality draws” that you have to discard are worthless. How much do the results change if you enforce the hand size?

[quote]
How can we take tempo and opportunity cost into account? The cost of
event cards doesn’t impact the curves if they never go below the event’s
cost. In the event economies, if we never go below 5 credits, then we
never see the downside of Sure Gamble‘s cost.[/quote]

Runners also need “quality draws” and free clicks to install the necessary programs and to make runs. A good model should take this into account. Just as an idea for the future, it might be possible to play a simplified game in an average sense. As a starting point one could assume

• fixed number of accesses to win the game (maybe 10, sometimes access 1+ card).
• fixed cost for an access (3-4 credits maybe?)

and then compare the number of clicks (or turns) each economy+draw package needs to win the game.

While this is a crude approximation of a real game, tempo hits from playing expensive cards should become visible since the runner needs to find the economy to win the game.

[YCombinator]

@kenobob

My hunch is it’s the power of another econ card. Lucky find is more efficient than Day Job, yet the curves are about the same.

This may be right. Perhaps it is worthwhile just looking at each econ card in a vacuum.

@Nordicstrike

The runner still has an infinite hand size right? This assumption may give decks that draw a bit too high quality draw efficiency, since “quality draws” that you have to discard are worthless. How much do the results change if you enforce the hand size?

Yup. I believe the assumption is roughly equivalent to never discarding econ.

I don’t know how the results change if you have a limited hand size. If you can think of a way to automate what to discard, and what not to discard, and how to count that WRT quality draw, I’m all ears. I don’t have many good ideas there, unfortunately.

@Nordicstrike

As a starting point one could assume

fixed number of accesses to win the game (maybe 10, sometimes access 1+ card).
fixed cost for an access (3-4 credits maybe?)

and then compare the number of clicks (or turns) each economy+draw package needs to win the game.

While this is a crude approximation of a real game, tempo hits from playing expensive cards should become visible since the runner needs to find the economy to win the game.

A good proposal. A question it raises is how to really take into account tempo considerations? Just fitting multiple 4 credit runs under the curve would yield a number of accesses, but I don’t think this is accurate.

A similar way to look at it is to consider the parts of the deck that don’t require corp interaction. These include econ, program/hardware installs, and dependencies between these installs (due to memory constraints, or simply program install ordering preferences). Once this is done, and the credits for the installs are taken into account, then how much money is left?

Given this new “quality credit” curve, then we can try and answer some of your questions, in a similar manner: how many 4 credit runs fit under this curve? How early can these runs be made, and how does that impact the rest of the system. The weakness of this is that it will ignore early-game opportunity, where you make early runs, to the detriment of program installs, and using more econ. Next, we can manually add these in, and see the impact.

Thanks for the ideas!

[moistloaf]

@Nordicstrike

fixed cost for an access (3-4 credits maybe?)

I wanna play in your meta.

[bblum]

I don’t; sounds like it’s all neh blitz!

[Nordicstrike]

@moistloaf

I wanna play in your meta.

Well, early game accesses are cheaper… I pulled out that average cost for a run from a hat, but parasite and stimhack business should be kept in mind, since this discussion is mostly about Shapers.

But let’s make an estimate:

• according to the results here, PPVP Kate can get to ~38 credits in 30 clicks
• need 10 clicks for runs
• maybe 5-7 clicks to install icebreakers, clone chips, and other stuff

If you pay (on average) 4 credits per run (40 credits) and 5-7 credits (let’s say 7, with Kate discount) to pay for programs & hardware, then you need 47+ credits to win. You’d need 47 / (1.2 credits/click, I’m generous) = 39 clicks to generate the credits and 15 clicks for the runs and installs, that is, 13-14 turns.

So according to this back-of-the-envelope calculation, you can win the game in

• 3 credits/run: 11-12 turns
• 4 credits/run: 13-14
• 5 credits/run: 15-16
• 6 credits/run: 17-18

Seems reasonable?

[YCombinator]

Holy hell, that’s what you meant by your access analysis! Fantastic! This is also a framework in which multi-access cards can be considered as they cost setup and money, but increase the access rate per run.

Thank you so much for adding this level of specificity. This is a great idea and might be the third layer I’ve been looking for:

1. Econ = draw + credits (what these articles have been about so far). I need to add some notion of a minimum credit pool required to get the resulting efficiency (i.e. 5 credits to play SG).
2. Installs = consistency of installing a set of breakers considering dependencies between them (2 card combos, MU) which requires the economic analysis. This will also be required to consider the economy of Noiseshop and Desperado+security testing+john Criminals.
3. Access efficiency = how many accesses one can get over time, given varying costs to break through ice. This could consider multi-access cards.

The main part that would be missing is the economic impact of trashing assets, and a co-simulation of the corp would be required for that. We’ll call that (4).

@Nordicstrike: sorry for not quite getting the full implications of what you said before, but again, thanks for adding this here. Sometimes examples are the best way to convince someone, and I’m on board.

…now when to find the time to get all of this coding done

[PeekaySK]

@Nordicstrike

Seems reasonable?

The general thought process seems legit, but I’d like to challenge the specific numbers:

It’s sort of been established knowledge that on average, you as the runner need to access 17 cards to win the game. I’m assuming you take this to still hold - so, seeing as how you’re counting with 10 runs, it would appear you’re expecting the runner to slap down some sort of multi-access after 3 single-access runs, at the latest.

So, my question would be - are we really assuming that a complete breaker suite and a console and any circumstantial hardware (Feedback Filter, Plascrete) and multiaccess tools fit within 7 credits? That seems overly optimistic to me.

edit: one more thing the model could bear would be accounting for at least some, minimal form of interaction with the corp - namely, trashing assets. Based on what’s getting played these days, I’d aim for about 5-10 credits spent per game, minimum (Jackson, DBS, Adonis, SanSans), how does another 7 credits added to the cost needed to win sound?

[YCombinator]

@PeekaySK

So, my question would be - are we really assuming that a complete breaker suite and a console and any circumstantial hardware (Feedback Filter, Plascrete) and multiaccess tools fit within 7 credits? That seems overly optimistic to me.

Agreed that some of these assumptions might be off. At the very least, the values will be different for different decks. The answer is to add into the simulator (see (3) above) knowledge about playing breakers and building board state.

The question will be, when the code gets to this state, what sets of progs/hw should I consider? Opus is obvious. Typical shapers with 3 breakers are obvious (with an emphasis on when the sentry breaker comes out). Kit with one breaker is obvious. Crim with faery. What else? What install packages have dubious consistency, which don’t? An article is more interesting, I think when we can study things that we already have an intuition about, and provide a firm reasoning for why the intuition is justified, or not.

@PeekaySK

edit: one more thing the model could bear would be accounting for at least some, minimal form of interaction with the corp - namely, trashing assets. Based on what’s getting played these days, I’d aim for about 5-10 credits spent per game, minimum (Jackson, DBS, Adonis, SanSans), how does another 7 credits added to the cost needed to win sound?

I couldn’t agree more. See (4) in the reply to @Nordicstrike. A co-simulation of the runner and corp is the last step in all of this being significantly more useful.

PS – @PeekaySK: Your comments and decks were some of the most influential for me when starting ANR. Thanks! I’m still a CI fanboy. CI is also the most algorithmic identity (least required interaction), so will probably be the first corp to study.

[Fricassee]

Can we get an idea of what priority you put on Day Job? Did you put it before everything else, or would you drop a PPVP first, pushing Day Job back a turn?

[Nordicstrike]

@PeekaySK

It’s sort of been established knowledge that on average, you as the runner need to access 17 cards to win the game. I’m assuming you take this to still hold - so, seeing as how you’re counting with 10 runs, it would appear you’re expecting the runner to slap down some sort of multi-access after 3 single-access runs, at the latest.

edit: one more thing the model could bear would be accounting for at least some, minimal form of interaction with the corp - namely, trashing assets. Based on what’s getting played these days, I’d aim for about 5-10 credits spent per game, minimum (Jackson, DBS, Adonis, SanSans), how does another 7 credits added to the cost needed to win sound?

I wasn’t even trying to be too precise here. I agree that 10 runs is too little, 15-20 runs is maybe more like it if asset trashing is taken into account.

Generally the runner needs to make N runs to (multi)access cards, kill Jackson(s), trash assets, etc. The credit cost to break ice and trash stuff assets then depends on decks. Just as a starting point, I’d go for a simplified model, where the different costs are accounted for in an average sense, and then see if @YCombinator’s model starts to have tempo issues.

Making a run might force the runner to click for credits in order to play an economy card, and playing an expensive “quality draw” card or ProCons might prevent a run on the following click. So, “loss of tempo” should then increase the “turns to win the game” (or some similar metric) OR the runner might have to let some assets stay on the table.

@PeekaySK

So, my question would be - are we really assuming that a complete breaker suite and a console and any circumstantial hardware (Feedback Filter, Plascrete) and multiaccess tools fit within 7 credits? That seems overly optimistic to me.

It probably is, I actually pulled that number from my PPVP Kate. Lady, Mimic, and 1-2 Cyphers with discounts are 6-7 credits, SMC is free, and Chips you can usually install for free. Might have to add some mandatory installs to the toy model to really bring up the tempo issues. For example, assume that for each three runs, you need to install a card for 2-3 credits (so the runner would generally run-run-run-install, run-run-run-install, …). Just to throw out some not-too-complex ideas

I’m just generally cautioning against trying to put too much stuff into the model, it’ll soon end up in http://www.simulationcraft.org/ level of detail.