MajorJuggler

Members
  • Content count

    5,248
  • Joined

  • Last visited

6 Followers

About MajorJuggler

  • Rank
    MathWinger
  • Birthday

Profile Information

  • Location
    Massachusetts

Recent Profile Visitors

5,392 profile views
  1. In any given wave, there has historically been a very large variance in how good some ships / pilots are compared to others, even in the same wave. When anyone refers to power creep, they are typically referring to the top-end power curve that is set by the BEST ships in the game. A new ship that's better than an old garbage ship is not power creep unless it is also better than the previous BEST ships. Nobody talks about a ship and calls it power creep because it's better than a wave 1 TIE Advanced. Every ship in the game is better than a wave 1 TIE Advanced. People talk about power creep when a new ship is better at jousting than the Howlrunner TIE Swarm. That's powercreep. In this case, the Gunboat + HLC is a better straight-line jouster than any other generic ship in the game. Now, generic jousters have been dead for about 2 years now (i.e. they have no contribution to the game's power curve), so it's possible that the Gunboat won't meaningfully contribute to the game's overall power curve. But it probably will, at least a little. I'll spend more time deep-diving on the gunboat mathematically in another month-ish. For some reason, FFG forums aren't letting me upload PNG files. Otherwise, I would upload the custom "Aft Cannons" Lambda Shuttle only modification card that I made for Community Mod, which grants them an ATT2 rear arc. (Bring Your Own Cardboard, aka Firespray).
  2. It's in the queue, I will have to do that eventually once I rebalance TIE Bombers, K-wings, TIE Punishers, etc, for Community Mod. ETA is December or January though - working on rebalancing everything else first. It turns out that rebalancing 12 waves of content in one wave's time, in your spare time, using more rigor than the original designers, is a somewhat optimistic goal.
  3. [edited] Depends how often you still have the evade token. If you have it 75% of the time then it's around 111%, assuming the targets have generic-ish type action economies for defense. All Vessery numbers obviously depend how often he's getting a target lock as well. The #'s here and above assumed a 50% trigger rate, so your mileage may vary.
  4. Generics yes, Vessery / Ryad no. Generics are "only" in the 97% - 99% range re: wave 3 power curve. Vessery and Ryad are both around 107% (although Vess's EPT is still free!). All numbers use an overly simplistic assumption that x7 triggers on 75% of turns you're getting shot at. Also, x7 Defenders get a white K, so, there's that! Edit / PS: the efficiency style is obviously very different between HLC gunboats and TIE/x7s. TIE/x7s are super tanky, gunboats are on a survival clock and are skewed towards being glass cannons, comparativly speaking.
  5. I have yet to fly gunboats myself, but this is what the math is predicting. Harpoon boats need to burn their EPT on deadeye to be on the same footing as HLC gunboats, i.e. just point and shoot. Low PS jousters like this are fundamentally just point and shoots -- good players will find ways to maximize their approach angles, and certain ships (B-wings, Skurgg) can reposition to help this along, but at the end of the day it comes down to just getting the other guy in arc and rolling dice. As long as you can get the target in arc reliably (sorry Lambda), jousting efficiency reigns supreme and is still the best predictor of this archetype's success. In this regard, HLC is better than harpoon. Harpoon missiles themselves are insane, but their overall average damage can't keep up with the HLC due to the lower firing duty cycle. Reloading is a great option to have, but if you're relying on it in the middle of a heated battle (which the early engagement almost always is), then spending all those points on low PS jousters and then not shooting with them => you probably lose. Plus, deadeye boats can't spend their focus on defense, which matters quite a bit for AGI 2 -- again, lowering their jousting efficiency, which is not what you want out of any ship, let alone a jouster whose only real trick is "smash you in the face with lots of dice". This isn't to say that harpoon boats can't win and aren't dangerous when lining up opposite them -- but in the meta-game sense, on the whole, I don't expect them to be as good as the HLC boats, just based on the raw numbers. The FFG Devs did a pretty good job here, but I am concerned that it is still quantifiable power creep, even if it is milder than we have seen in waves past. It basically matches the jousting efficiency of a lambda shuttle... and can turn easily and also has SLAM. Dee Yun @Mynock Delta made an astute passing comment on a recent Mynock podcast, essentially that he "doesn't even know what the power curve is anymore". I think the gunboat is another example of the developers nudging up the game's overall power curve.
  6. Can confirm. MathWing 3.0 jousting math for PS2 + XG-1 + HLC + Linked Batteries puts both the PS2 (28 points) and PS4 (31 points) at `104% to 105% jousting efficiency, relative to Ye Olde Wave 3 Baseline. Cons: I'm assuming they always have an HLC shot, which is somewhat optimistic, as they will occasionally get stuck with a 3 dice primary shot at range 1. The game's overall power curve has gone up significantly since wave 3, so having the jousting efficiency about as good a lambda shuttle (about 106% re: wave3 baseline) isn't as amazing as it used to be. No K-turn. Pros: SLAM! The PS4 still hasn't used its EPT slot in this analysis (hello crackshot) Long Range Scanners (not modelled in the above) They set the new standard for point-and-shoot non-ordnance generics, but that archetype has been dead for a long long time now. So, how good do I really think they are? That's a harder question to answer, but the baseline power curve that I have been using for Community Mod can help shed some light. The power curve for Community Mod isn't exactly the same as stock, but it's pretty close for the top-end stuff. Almost everything gets buffed, with a few ships like Soontir and Fenn staying put where they are. Using the Mod-Wing power curve, the efficiency drops to around 88% to 89%. That's still not too bad, and basically says that in the stock game you'll need to use the SLAM action and Long-Range Scanners (and EPT) to increase your efficiency, which is entirely possible. As with any arc'ed ship however, its weakness is going to be arc dodgers, which should be seeing more play now that bombs have been nerfed. This is just a really quick initial estimate that literally took only took a couple minutes to update scripts for and recalculate -- about as much time as it took me to write this post. I'll do a deep dive later when I get around to fully rebalancing the ship (as needed) for Mod-Wing, which right now is at the end of my queue. Probable ETA is December or January.
  7. Nice, added to the Index of Useful Links!
  8. Done!
  9. Working on it! X-wing Community Mod is currently in closed alpha. Although, I am getting a very limted amount of data, and I really need my first "wave" of ships playtested, so I might open it up soon.
  10. Sure, I can add this to the pinned thread. Here's another category. Rebalanced for standard 100/6
  11. Writing an AI that can play the game intelligently would be the cornerstone of the holy grail of automation. Everything starts there. Once you have a machine that can play the game against itself, and play well, then you can leverage that for a lot of different things: Seeing how effective certain lists are at other lists Backing into those results to get an indirect measurement of how good certain ships are There's a few ways to go about this. Conventional wisdom is to write the AI code yourself and hope it's good enough. But a better way is to just feed the rules to a Neural Network, give it 0 strategy, and let it play against itself 100,000 times, to let it self-train the optimal strategies. This is the approach that the current Go Neural Network used, and it managed to beat the previous bot (that had dethroned the humans) quite handily. Once you have a bot that can play the game, then the real fun begins. Then you can create a AI / Neural Network layer at the list building stage, which can try building new lists and find new combinations. You would use the first bot to test the lists and see how feasible they are.
  12. Right, this is exactly why I wasn't going to push the point during the interview. I could have corrected Frank's response during the interview, but it wasn't worth getting into. FYI, Jay Little has said essentially the same thing in a back and forth online Q&A I had with him a few months ago. I think both Frank Brooks and Jay Little fall into the trap of thinking that since they are really good at game architecture and content development (which they are!), that they are also the most experienced/knowledgable people in the discussion with regards to technical balance, just by virtue of being employed in the games industry and having experience in the industry. Unfortunately this leads to all the kinds of design issues that I mentioned offhand a few posts earlier. Math is just a tool. It doesn't "do" anything, it's just a reflection of reality, or more specifically, what reality will be after you release your new content. So if you want to mix things up in the future, or to create perpetual (but controlled) power creep, then you can use math to accomplish those goals too. Edit: every complex system still has a steady state solution. I.e. the meta settles out. There's no such thing as an "unsolvable" system, the community itself works it all out pretty quickly after product gets tested at tournaments anyway.
  13. In short yes. Identifying combinations is a very different problem than quantifying them. Sometimes after doing analysis some things become obvious, like Commonwealth Defenders, but other times you either need Jeff Berling or deep machine learning to find the Dengaroo style interactions. Bombs are actually pretty straightforward, although I haven't yet sat down to rigorously go through them yet. But you know that Cluster Mines + sabine autodamage on a 3 hit point 35 point ship is going to be worth a lot. Most of properly quantifying bombs goes back to step 3 above, i.e. by analytical playtesting find out how often the bombs are actually going off and hitting their targets. Then it's much easier to go back and calculate what they are actually "worth" in squad points.
  14. A change in the meta, say between high AGI targets and low AGI targets, or a change between high firepower ships and low firepower ships, will generally move a ship's efficiency by a few percentage points, which is noticable, but not enough to dramatically affect a ship's overall viability. Looking under the hood, what the different meta assumptions are actually doing in my scripts, is changing each ship's average action economy, expected durability, and expected damage output. This is not the same thing as hard counters, like PS10 Bombs, or PS8 Miranda action bombs, either of which murder low hit point PS9 aces. Nerd sidenote: solving the the action economy coefficients everytime I change the meta assumptions (or generating a new meta testbed) is a high-order multi-dimensional problem (i.e. there's a variable for every possible attack/defense permutation). The probability of spending focus when attacking is essentially unchanged regardless of meta, but the probability of spending a focus on defense is very much a function of what's shooting at you. This in turn affects how likely you are to have focus on offense, which in turn affects how often everyone else has to spend their focus on defense, and this cycle keeps repeating. So solving this set of equations is an iterative problem until all the action coefficiencts settle out at a final result. So, in my scripts I literally have a step labelled "calculating hyperspace convergence". Edit: Maybe this would have made for an even better Star-Wars-ey sounding question: "Are you aware of how calculating hyperspace convergence can affect game design?"
  15. Historically the opposite has been true, math has had better predictive power of game balance than the designers' intentions. There's a very long list of pilots / mechanics (especially when I was evaluating most everything during waves 3-7) that this was true for either pre-release, or very shortly after once I ran the numbers: TIE Fighters being better than anything else in wave 1 by a wide margin (post-release but I was the first to actually "prove" it) B-wings vs X-wings (also technically post-release, but I was ahead of the meta by about a year here) Original Defenders (nailed this even before the dial was spoiled, much to the hostile reaction of several playtesters) Autothrusters (hello 35 Fel) Palpatine (technically post-release, but did the math after one game with him, and demonstrated how bonkers he was with 35 Fel) TLT x7 Defenders (related: I created/popularized Commonwealth Defenders right after the preview article.) Inquisitor (it turns out that jousting efficiency > 110% is really good at PS8 with reposition, go figure) Parattanni (was late running the numbers on Attanni anything, but when I did it demonstrated that it was the best jousting list in the history of the game). ... and more that I can't think off of the top of my head. Having worked on Community Mod for a couple months now on and off, I believe I am the only person who has rebalanced (or designed) X-wing using a comprehensive mathematical analysis as fundamental to the design. So, I have a unique perspective which has yielded several conclusions: You need to really get the high level theory down before you can even think about getting useful results from any sort of mathematical implementation. There's a lot of components in here, like: deriving what the fundamental power:cost curve looks like (and why); quantifying the general-case for how PS and the glass cannon/tank ratio specifically affects the raw efficiency numbers; how firing duty cycles, efficiency, and 'break even points' are all related; what efficiency numbers should be ideal for any given archetype and why. There's enough content in here that's worth writing an academic journal article, which in theory (no pun intended) I would like to publish someday. If you can't get past this point as a designer, then math might be useful, but frankly you really don't know how to use it. Just knowing how probability works, but without understanding all of the above would be like a carpenter having a hammer and banging some 2x4's together, without having a blueprint of the house that he's building. You'll get something, but don't be shocked if it doesn't do what you wanted. A proper and thorough implementation of the above theory is also not trivial, but of a different kind of not trivial. Think more code-monkey elbow-grease kind of not trivial. For context, I have >10k lines of MATLAB code. The mathematical implementation (i.e. matlab scripts) has to assume some trigger rates for every ability that you're evaluating. How often does x7 trigger? If you're taking Backstabber, then how often does his ability trigger? There's a very long list. You can't just pick numbers out of thin air and expect it to be right -- you need analytical playtesting to see what the trigger rates are. The best you can do is set up a framework to evaluate everything, and then get some data analytics from real-world testing to see what your mathematical coefficients should be to best reflect reality. Everytime there is a new mechanic introduced, you need to think about it very carefully to see how it can be modelled mathematically, and with what degree of certainty. Usually after some analytical playtesting, you can get a pretty good approximation of "ability X" that reflects reality. After their responses to my question of "what if there were an equation that could help balance the game", I wasn't going to turn around and offer a literal graduate level course lecture on how they actually could execute such an approach. We were already poking at them enough, and the point of the interview was to get them to share their insights. Plus I don't think it would really matter unless they would be willing to listen - in general they seem to be very confident about what can and cannot be done mathematically, despite the fact that they "don't know what they don't know" when it comes to mathematical analysis. Frank's response that (paraphrasing) "there isn't an approach that simultaneously works for jousters, arc dodgers, and turrets" was a little suprising, considering that a) he has an engineering background, and b) he has to know that I have pioneered that field. I liked Alex's response much better, pointing out that math is good for evaluating existing mechanics that are well understood. However, I am much more optimistic that it is possible to understand a new mechanic and its implications before you are done playtesting, especially if you have the right tools that can realistically model the mechanic.