if a worlds winning decklist has romp in it, i’ll sell my collection. I feel very safe making this bet.
3 Likes
Sell it to me, cause I’m going to win world with disruptor and exploratory romp, with force of nature
Also, Off The Grid with Crisium Grid on HQ.
disruptor got a bit better with the new NBN current, the one which forces the runner to start the trace-bidding.
Excalibur for the win win win
2 Likes
I wanted to check if the discrete-ness of this composition would make it harder to steal. I used a script I wrote. Here are I think the only reasonable options for non-TFP decks, and some historical options included for comparison.
The number is number of accesses until the runner hits 7.
Compact suites:
44GGG3 → 20.4
3333333 → 18.8
633333 → 19.1
33332222 → 17.0
3GGG2222 → 20.1
Ycombinator, your suite is too good.
NBN suites:
3322222211 → 17.1
22222222211 → 17.1
G222222211 → 18.8 (Calimsha’s suite, counting the 15 minutes as a 0-pointer which I think is reasonable)
1 Like
What about 333GGG2?
Or 333G2222?
Thanks for the data!
If you compare the 44GGG3 with 3GGG2222 (20.4 vs 20.1 respectively), they’re pretty close in number of accesses. However, -2 agendas, +2 useful cards is a big difference as well. I feel like the 44GGG3 composition opens up corp deck-building a little bit.
That said, it has less flexibility in how it scores (and costs more influence), so take the good with the bad.
I’m looking forward to integrating this into my CI OtG deck.
Would it be possible to re-run the simulations also reporting at least standard deviation, but better the 85th percentile confidence intervals?
My suspicion is that the 44GGG3 composition has a little more variance than the others. 20% of the time, they steal 2 of the 3 agendas that let them win in 2 scores. The lower number of accesses in that case will likely increase the variance significantly compared to the case where the runner must steal three (i.e. 50% of your agendas).
Long story short, I think the average is hiding importance differences between the compositions. The same logic would apply to TFP (do you win the psi-game early, or late…).
Thanks for the simulation! Any interest in adding it to the simulations I used for the QuantANR series? That code-base is also python, so it could integrate well.
2 Likes
I modified @jrp’s script a bit to do comparisons of cumulative probability distributions, and ended up with this chart comparing 44GGG3, 3GGG222, and 333GGG2:
4 Likes
I wonder if Early Premiere will help with scoring these more subtly/easily. If you have one out, VP is on the table just as long as GFI or any other 5/3. Too much influence for outside of NBN, but could still free some room up in the same way Food could.
It’s nice to discuss on how to minimize the agenda density in R&D but at some point, you still need to score those agendas.
Looks like that’s something who got overlooked in this discussion 
2 Likes
I tried running the 5-agenda Government Takeover spread. 6x1 + 3x4 + 1x2 to steal on access. That gave 19.2 accesses to win with 9.3 sd, which is an access or two worse than I was expecting – apparently that’s the difference between winning in 2 steals most of the time or winning in 3 (or 4).
Of note is the fact that the worst playable spread with TFP in it is at 22.5 accesses to win, and you can hit 25.9 accesses to win (9.9 sd) with 3x TFP 1x GFI 3x Nisei 1x NAPD, where TFP is stolen 40% of the time and NAPD is stolen 90% of the time.
Looks to me like GFI is going to widen the gap between RP’s agenda suite and the rest of the field, not shrink it.
Edit: Okay, so that’s a little unfair. The worst pre-GFI spreads gain +2 accesses to win, whereas the best TFP spread gains 1.7. So a slight narrowing.
1 Like
Damn, I hadn’t actually considered this but it should work! Reducing the number of agendas means more slots and it’s rare for it to lose agendas early on thanks to its natural resilence. It’s probably a quesiton of influence.
I think there are many previous discussions about scoring 5/3s. We can reiterate that - and the difficulties therein - here, but I’m not sure if any of it is new. The compositions are new.
Conclusion: TFP is OP 
@kevelairn found and fixed a bug in my script that caused TFP to require 1 extra access if it was the last agenda scored. I reran and the numbers didn’t change substantially. Here is the new version if anyone wants to pick up the fix.
I think standard deviation captures the variance pretty well. Do you mean 85th and 15th quantiles? I’m not sure what the confidence interval means in this case. @kevelairn’s graphs of the CDF are informative. (And the current github version outputs the CDF too.)
I agree that average is not a great measurement in general. For what it’s worth I didn’t see a case where avgA < avgB but medianA > medianB, so the average is not behaving too badly here. I certainly don’t endorse reading too much into minor differences in average or median - as you said 20.1 vs 20.3 is probably less important than a card slot, or having to score a 6/4, etc. I get excited when it’s 17 vs 20 though.
Feel free to incorporate the code into your library.
Awesome. On the subject of GitHub, here’s where I keep my R package for parsing the OCTGN data.
Latest results: loyalty to a single ID has zero effect on your skill rating over any time horizon. Play one deck, play lots of decks, the game gives zero fucks.
3 Likes
CDFs are more than enough information ;-). It was good to see. You see the effect I was pointing out at low access numbers where 44GGG3 is higher than the rest, but it quickly drops off and is stronger around 13 accesses. It is a little swingy: You might lose quickly. If you don’t, you’re in a good spot.
Thanks again @jrp !