|Live Action Mafia
|Page 1 of 1|
|Author:||achester [ Mon Dec 25, 2017 11:21 pm ]|
|Post subject:||achester's thoughts|
The other spymasters have committed to not looking at this thread until game's over; posting thoughts here so observers after the game can see.
|Author:||achester [ Mon Dec 25, 2017 11:21 pm ]|
|Post subject:||Re: achester's thoughts|
Day 0 thoughts:
-Green seems to have the strongest team at mafia. Blue seems to have the strongest team at Codenames. Red is probably in between on both, but closer to the weaker end on both. Accordingly, I'm much more scared of green than red.
-There's no HP-reveal rule. Why? The short version is that the spymasters no-trade-theoremed. The long version:
We spymasters didn't think about having an HP reveal rule until after we'd seen our teams, so any change to the rules needed consensus from all three of us. I strongly preferred revealing all HP, since that makes queue formation easier and lets people know how much of each clue they've gotten, both things that I think my team can take advantage of better, and thought giving each team member one target person's HP would be better than nothing, because it gives the same information, just slower and in a way that green is more likely to mafiaishly abuse than blue is. So, I proposed all-HP and mentioned single-target-HP. (Oh, probably also both of those make the game a bit more interesting, independent of their effects on balance.) Python, fitting my expectations of a green spymaster who had the same judgments as I did, gave exactly the opposite preferences (no to all-HP and slightly against single-target-HP). So, we defaulted to not changing the rules.
-Blue's True Names include lots of geography words. I think red got really lucky on this. The only two other True Names that are geography at all (ALPS and PORT) are red; red would rather hit itself than green because it at least helps reds connect with each other if a red gets shot for something that's clearly a red-clued word. Without too much looking, I see AUSTRALIA for 7, SCOTLAND for 6 or 7, or EARTH for 8 with a lot of bystanders. I don't think this elevates red over green as my most feared opponent, but it certainly helps. Green's True Names don't seem nearly as easy, although neither do red's.
-Jackie says she'll update someone's team by a factor of 3 for a witnessed fire (seems about right, although if she's not confident in the person claiming to have witnessed a fire, it should be less) and a factor of 1.2 for being clued. I'm not sure what she could even mean by that 1.2; it should be much higher (e.g. I could clue every name of lukesci's and, if she actually updated only 1.2 for each, she'd still think he was most likely blue, just for having claimed blue to her). She doesn't seem to be thinking much about that one, though, so I expect she'll update once more once she sees actual clues.
-I'm spending too much time on this. For the first day's clue, I generated 88 possible clues, evaluated how well 47 of those fit with the 36 most relevant words (18 green True Names to hit and 18 red True Names to avoid), evaluated how well the top 14 of those fit with the 54 next most relevant names (green's other names), then thought about how the top 6 of those would do on everything else. At least I have things set up well enough in my spreadsheet that I should be able to reuse most of that work on future days.
-Spreadsheet setup: I have a grid, currently 88x90, of how well each possible clue fits each relevant name. I have weights on each name for how much I do or don't want to hit it: 1 for green True Names, -.3 for other green names, -.8 for red True Names, times 0 for DROP and FALL because so many of my best clues hit them that I'm resigned to their being hit at some point, plus .4 for aok's and lukesci's names (both True and not) as greens trying to infiltrate blue. (AOK would get access to a bit of color information, probably not that big a deal. Lukesci I'm worried might get access to jamb's spreadsheet, which would be the real prize since I think a majority of the game's Codenames-solving effort will happen on it, and jamb'll probably solve the other teams' clues thoroughly again.) For future days, I can just change those weights (e.g. move red's weights closer to 0 as blue figures out who's who; zero out the weights of anything blue guesses) and recalculate the clue qualities automatically.
-Thoughts on the top 6 clues for today:
6. GANDALF THE WHITE for 7 (only 10^5.9 google hits, but the other spymasters approved): this casts suspicion on two of jamb's names (MINE and TOOTH) and only a bit on aok (DEATH), so it'll be better tomorrow when aok's outed and jamb's more trusted. It does get PASS, FAIR, and RING well, all words that will be relatively hard to clue otherwise.
5. SHOULDERS for 7: decent on raw numbers, but it doesn't out aok or lukesci, does out ksedlar (which I might dislike a bit because I think she'd do better at running green than pravinas would, and green might not yet trust her). Also, several of the connections are via square dancing, and I don't yet know how active Troy'll be.
4. METALWORKING for 7: hits MINT, which is otherwise pretty hard, and definitely outs lukesci, but probably throws more suspicion on jamb (for MINE) than on aok (for BAR). Maybe another day.
3. DESCEND for 6: definitely outs aok and has a great chance of actually killing him with only his True Names hit. Points at lukesci for DWARF and SPIDER, but those are just weak enough that if jamb thinks he's blue she might ignore them. Also, jamb has MINE, for which people might be suspicious of her too.
2. SLIP for 8: Definitely outs aok and not jamb. Doesn't hit lukesci at all. I don't yet know how active Troy'll be for the square dancing, and it's a generic enough word that people might ignore that connection. Might hit SHOE, but hopefully people realize that if I'd wanted that I could've done SLIPPER?
1. TITANIC for 8 (SINK on aok, CRASH on aok, TRIP on aok because the Titanic was making a trip, TRAIN on ksedlar as another somewhat-old and canonically English mode of transportation, DANCE on Lucy because of the Titanic dance scene (but I'll probably get it tomorrow anyway), DWARF on lukesci because dwarves and titans are both mythological creatures synonymous with extreme height, and PENGUIN on brunnerj and SNOWMAN on Lucy because the Titanic hit an iceberg.): Definitely outs aok; good chance of killing him, but he might get shot for DEATH before TRIP, and TRIP might not make it into today's top 6. Doesn't hit any of lukesci's True Names well (although I hope it'll at least put DWARF on their queue), but clues GIANT so strongly that blue should at least mistrust him for today (that's the biggest reason I'm not interested in instead cluing RMS TITANIC). Doesn't hit any of jamb's (or corwind's) names, so jamb can coordinate and not mistrust corwind. Clues JACK on ahkim and HOLLYWOOD on dylanhen pretty strongly, but at least they're not True Names. Might also hit COLD on dylanhen, ANTLANTIS on lzahray, SNOW on lzahray, and UNDERTAKER on Troy. Also, it doesn't overlap much with the dancing or LoTR clues which are likely to be near the top of my queue tomorrow.
|Author:||achester [ Tue Dec 26, 2017 10:42 pm ]|
|Post subject:||Re: achester's thoughts|
Day 1 thoughts:
-RED SEA for 6? Seems a little small given all his geography things, but pretty good.
-BANGLADESH for 10: What on Earth? This seems to hit a bunch of blue True Names really strongly, and even looking at the red True Names, I can't see more than a few that it clues at all. PORT, I guess, because it's coastal (and amusingly, PORT is probably more clued by both RED SEA and TITANIC). SHOE and BELT, because they make clothes there. EMBASSY, because they're a country, although there are a bunch of actual countries that aren't red. In fact, it seems to hit Jackie's INDIA more strongly than it hits anything red. Sad for team unity. Either Python is up to something really tricky and I should be terrified, or he gave a bad clue and I should stop being scared of him and start being terrified of jasonye (since a lot of blue True Names are going to accidentally get hit by green today). Leaning toward the latter.
-Python said in spymaster chat that BANGLADESH was intended to mostly be about the substring BANG. His team isn't seeing that, and I wouldn't either. Anyway, I think red is now the biggest threat because I think green's going to waste most of their shots today, and disproportionately on blue True Names.
Updates to colors for being clued:
-One of the biggest problems I'm having is that jamb (arbitrarily, as far as I can tell) came up with a factor of 1.2 as her update on someone's color if they're clued and this is way too low. Also, there's no explicit dependence on how strong the clue is, but she seems to have settled on a factor of 2 for the six strongest clues, CRASH, DEATH, SINK, SATURN, SHIP, and GIANT (all of which I did in fact know were very strong, and all of which are in fact green), which is even more way too low.
-I'm going to calculate how big such an update should be. I'll start with RED SEA for 6 as an example: jamb found 48 things that fit at all. As a first-order approximation, suppose that all 6 True Names that jasonye intended to clue are included and the remaining 42 are split evenly between red, green, and blue. That's 20 blue words to 14 green or red, so in that case the update for each of the 48 words should be a factor of 20/14 = 1.4.
-Most important correction: some words are stronger-clued than others. Suppose we divide those 48 into the top 6 (like jamb did for TITANIC) and the remaining 42, and suppose (very conservatively) that only 3 of those top 6 are intended True Names and the others are split evenly. Then those top six are 4 blue, 1 red, and 1 green, a 4x update. (Of the remaining 42 words with the same assumptions, 16 would be blue to 13 for each other team, for an update of 1.2x, which does seem pretty reasonable for random not-too-strongly clued words (like the four of lukesci's that jamb sees some connection from to RED SEA).
-The assumption that the remaining words are distributed evenly is too conservative, because the spymasters can look at them, both to avoid other teams' True Names and focus on their targets' false Names over other teams'. (This is why, of jamb's actual top 6 for TITANIC, all six are green: I was paying enough attention to what bystanders I hit.)
-As a sanity check on update strength, consider LUNAR: we went from thinking that sammyluo was 98% blue to less than 10% blue, an update of a factor of at least 450, and that seems about right for that case
-Updating too much or too little for clues is like having a steering wheel that's over- or under-sensitive for a car driven by the spymaster. If it's oversensitive, I can adjust for it by turning the wheel less/cluing more conservatively. If it's undersensitive, I can't control the car. For instance, if today I clued all three of lukesci's True Names as strongly as TITANIC->GIANT, and Jackie assigned all of them her 2x update, but jasonye's clue hit two more words as strongly as RED SEA->FIRE and RED SEA->DWARF, Jackie would still think lukesci had a greater than 50% chance of being blue, left over from that he said he was blue at the start of the game. What would I even do at that point? Clue some of his True Names as strongly as TITANIC->GIANT a second time? (I realize I'm ranting a lot here; I think I'm frustrated because I don't feel in control of the game.)
-You should probably update more based on your own spymaster's clues than other spymasters'. The biggest reason is that the other teams have information on what words have already been shot and who's what color that you don't. For instance, if green is confident that ksedlar is green but blue isn't, Python can safely give clues that seem to hit some of ksedlar's words, since it won't lead green astray but will lead blue astray. (I won't be able to do the same thing cluing jamb's words after today, because green will also know that she's blue.) Also, the other teams might have made requests to their spymasters that you don't know about.
-The most frustrating thing is how so crucial an error seems to be going entirely unnoticed. That random number of 1.2 is getting used in all of Jackie's many updates, and she's relying on those updates in her spreadsheet, so that random number of 1.2 is determining basically all of her color perceptions, but she doesn't even realize how much impact it's having and hasn't sanity-checked or even explained it.
-Possible clue for today: "1.2 for 0".
-In the last game, blue did lots of great Codenames work that green leeched off of. I unfortunately didn't talk much before the game about how important I think this was; I think my team is better at Codenames and is in a strong position if it keeps all its Codenames skill for itself. The biggest relief of today was that jamb's only sharing her spreadsheet of what things are clued on green, which is basically costless to share and doesn't let the other teams leech off our work. Well done on operational security, Jackie. This is particularly important because the 1.2 error means her trust is bad.
-Oh, wait; Jackie's actual message to lukesci is "I don't think there's much point giving you my red+blue tabs from the other spreadsheets, since there's still some probability you're not blue, but if you have reason to want them I probably trust you enough to share." No! D:
-Good job jamb posting fairly quickly a reminder to everyone not to claim to aok. Sad that it'll relieve pravinas from organizing green.
-Jackie mistrusts Troy a lot because she has a conspiracy theory that he was trying to get information out of her that would only be useful in combination with information that aok wanted. Between this and trusting lukesci just because he claimed blue at the start of the game, I think she's putting much too much trust in costless things people say over the actual information of spymasters' clues. Actually, this is only a little high; it's mostly just way too high relative to 1.2.
-Jamb's asking for odd/even to indicate whether she's blue. That's be pretty low cost for me to implement, so it's good for them to do something like that. I'd sort of like them to pack a little more into it to include Troy as well; at this point in the game I'd be pretty happy with a factor of 3 or 4 constraint on my clue. Whether jamb is blue is unfortunately one of the least useful bits of information I can give my team right now. Yes, it'll make sure information from all blues passes to me, but jamb trusts lukesci enough that all information from blues is at risk of passing to green; jamb's going to be limited by time spent on game and her time would be much better spent Codenamesing and passing the results on to some other coordinator than coordinating things herself (since the team has several people who can do mafia schemes pretty well, but she knows how I think for Codenames clues best by far); and if she's the coordinator I don't know how I can possibly get her to mistrust lukesci or trust Troy, given that her uncaught "1.2x" error makes it really hardfor me to clue anyone's color to her (but other coordinators are likely to have better color judgments).
-It's sad that jamb assigned CRASH to lukesci. If CRASH and SINK were both going to be fired on by actual blues, I could clue TRIP really strongly to finish aok off with almost all of his bystanders unshot. Since not, I don't care about TRIP more than any other True Name.
-HEART's getting shot at. I knew that was a risk, but hoped it wouldn't be in the top 6; it's the only non-green True Name that's clued at all by TITANIC. The other shots are SATURN (intended bystander), SINK (hit), HEART (hits red), and DEATH or SHIP (known bystanders).
Clues for day 2:
-One possible path for today is to clue lukesci strongly enough that it can't be written off as an accident, and for multiple words. The risk is that if Jackie continues taking her 1.2 literally and is proven blue by the clue's evenness, then she'll still trust lukesci and the clue would be wasted. If her teammates realize and she explains this to them, maybe she'd not only uncover lukesci but realize that 1.2 is a mistake?
-The other possible path for today is to ignore the lukesci situation, let it resolve itself in a few days when he gets in enough espionage/witnessing beefs, and just keep cluing names. In particular, in the last game I thought (incorrectly?) that Daniel Grazian overreacted to our trusting sammyluo and that contributed to his MALFUNCTION stumble.
-I think I'll increase slightly the value of strong clues for lukesci's words and decrease the value of weak ones for today.
-One clue from yesterday seems to do everything I want: BLACKSMITHING. It hits DWARF strongly
-Today's top 9:
9. LIVERS for 6: Unfortunately, dwarves having multiple livers (for extra alcohol consumption) seems to be an OoTS-only thing, and I don't think any of my team would get it. There are too many bystanding living things.
8. METALWORKING for 6: Seems strictly worse than BLACKSMITHING for 6 (I think METAL hits more bystanders than BLACK, WORKING hits more bystanders than SMITHING, and METALWORKING doesn't have any chance of finishing anyone off.)
7. PIGLET for 6: the connections to DWARF and SPIDER aren't strong enough given lukesci's status; awkward that it gets TAIL but not ARM when I have several ways of getting ARM or TAIL that get both.
6. STRONG for 4: gets ARM but not TAIL, same problem as for PIGLET, and at least DWARF is not strong enough if Jackie's calling the shots on lukesci.
5. MORDOR for 4: seems strictly worse than FRODO. Gets PASS stronger, but KNIFE and DWARF weaker, and might hit FIRE, SPOT, and some of the places that Mordor is supposed to be an allegory for (e.g. TURKEY).
4. GANDALF THE WHITE for 6: WHITE hits some bystanders, and a lot of the things this hits are stretches. Also, it might hit LIFE on dylanhen, which is bad since blue doesn't trust him.
3. SHOULDERS for 6: gets the bodyparts well, and other clues don't get them. But hits dylanhen's COLD, which will be much better once he's trusted, and doesn't deal with the lukesci situation, so better to save for another day. Also, it might take some time for jamb (who'll be dictating tomorrow) to trust Troy, and he's the one most likely to get the dancing things.
2. BLACKSMITHING for 8: Should very strongly hit lukesci and maybe even kill him; all three of DWARF (a canonical blacksmith), MINT (turns metal into coins), and SPIDER (black) are clued. TRAIN is surprisingly hard to get, and this might. PENGUIN (part black), KNIFE, and RING are stretches. Might hit BELT and SHOE, but hitting red names is a lot less scary given that green seems to be losing, and having my team think that Daniel Grazian's green will at least dispel their thoughts that he's blue. Hits BAR (and DEATH, if they haven't already shot it), which are wasted shots since aok's close to death. Might hit AGENT, NEEDLE, DRAGON, FIRE, and LIGHT.
1. FRODO for 6: Might get TRIP and thereby kill aok. Should pretty strongly hit lukesci, with DWARF and SPIDER. PASS (that is, Cirith Ungol) and KNIFE (that is, Sting) are stretches. Absolutely gets RING, which might shift green's coordination a bit back to pravinas. Might hit EAGLE and SHOE, but hitting red names is a lot less scary given that green seems to be losing. Might hit KING (Return of the), but hopefully aok'll die before they get to that point on their queue. Might hit CLOAK, DRAGON, and LEPRECHAUN.
-Ah, lukesci shooting jamb is another way she could stop trusting him. Clever of lukesci to put that off to the last minute so I couldn't account for it, although I don't think it'd've changed my clue.
-corwind shot at CRASH without being in contact with anyone. That's the name that lukesci was supposed to shoot, so great, it got shot. But the rest of blue doesn't know that, so they might shoot it again. Also, he espionaged INDIA, which is indeed damaged, which might keep him from trusting jamb. Hope he at least gets into contact with someone.
-dylanhen missed his action, but is at least playing.
|Author:||achester [ Wed Dec 27, 2017 10:35 pm ]|
|Post subject:||Re: achester's thoughts|
Day 2 thoughts:
-I'm proud of my team.
Day 1 actions:
-Six greens, five blues, and three reds took their actions. We've heard back already from one blue and one red who just missed the deadline, but ixa might need to step in soon for a red.
-Three blue, two green, and one red True Name were shot (amusingly, in the opposite order from the number of shots the teams targeting them took).
Day 2 clues:
-MILITARY for 4 hits 5 blue True Names pretty solidly with only a few bystanders; good clue. At least it hits corwind so much that someone might think to reach out to him as blue.
-TREES for 6: I see four of them and some bystanders.
Win probabilities (start of day 2):
-Green missed all its shots, but Python says someone on green figured out BANG, so when that information gets around they'll have a great clue; also, their whole team is active. I think Python has been too busy so far to put much time into cluing, though, since BANGLADESH seems ill-considered.
-Red is dealing with many inactive people, but is in the lead on HP, and jasonye's clues are pretty good.
-Blue has fixed just about all of its mafia problems at once: lukesci and Daniel Grazian are gone, Troy, ahkim, jamb, and lzahray are correctly trusted, and I'm hopeful for getting the other two in today. But my clues don't feel as clean as red's, and the fact that fewer red shots were taken means that blue has less information on colors.
-Conclusion: 2^2:2^2:1 red:blue:green.
Blue team members:
-Jamb's taking a bit of a risk with operational security, letting Troy, lzahray, and ahkim in, but I think her (correct) belief that they're all blue is reasonable. Oh, actually, she's not even, only sharing the green clues. Caution's also reasonable. (It does make it less likely that the other team members'll update her on the other true blues; oh well.)
-Lisa, I forgot to mention it yesterday, but your paranoia about jamb yesterday was amusing, appreciated, and a good general principle. (despite turning out unnecessary).
-jamb's spreadsheet has guesses for the colors of everyone but cmcclena and laberca (who didn't do anything yesterday), and all of them are right except lukesci's. Yay! (Jackie: "I hope my alignments tab helps. I'm trying not to be too confident on any one thing." )
-Troy claims that dylanhen claimed a witness on day 1. Either dylanhen mistrusted Troy and lied, or there's a miscommunication somewhere. Hope it gets solved before they conclude that dylanhen is proven nonblue.
-Amusingly, jamb's first reaction to lukesci shooting her for INDIA is to mark him as red. I hope she catches the mistake (should be a more BANGLADESH-clued thing, despite the lies she told him) or at least sees all the names of his that are strongly clued. Holding my breath; this will either be an epic fail or a clean wrapup of yesterday's silliness.
-Update from two hours later: "okay so at the 75% confidence level, adamyedidia, dgrazian, lukesci, yyao are all red": Jackie's perfectly calibrated, although I think there should be enough information about lukesci. Definitely not an epic fail, although I do, annoyingly, have to find some way to clue lukesci. I think the answer is "ignore the problem until enough reds have shot at my team that they reexamine lukesci", or maybe give a clue that includes MINT.
-Update from another hour later: "if Lukesci is green and you're annoyed at us for not having put high enough odds on this based on your excellent clues... sorry, we'll get there eventually.
In general, I think you don't need to worry about hard-cluing people or doing much other than getting maximum throughput with clues, and we'll ask you for one bit per day for figuring out alignments." Thanks, Jackie; that's exactly correct and good to hear. And I think that strategy's correct and I'll follow it. (It's actually going to be more relevant on brunnerj than lukesci, because I think my highest-throughput clue today is going to hit lukesci strongly again, but brunnerj's the last one I'll only have clued weakly, and one of the last blue claimants to shoo away.)
Clues for day 3:
-Whether they figure out corwind in time to hear that he shot SHIP is important for my clues today. I guess I'll put off figuring out my clue until tomorrow.
-Whether they're going to shoot TRIP (which would kill aok) is important for my clues today. I hope they decide relatively early. I think they're doing it, and even better, jamb says "Once we think we've fired at 2 possible names of someone, we shouldn't fire at more than one of theirs per day unless we have good reason, since we don't want to waste shots if they'd die from the first one". I'm glad they thought of this.
-Whether they shoot two of lukesci's words is important for my clues today, since I could try to finish him off. I hope they decide relatively early. SPIDER seems to be the only piece of FRODO they've missed.
-My team's queue is pretty good, so I'll probably be giving a small clue tomorrow to let them spend their other shots on the queue, or even a clue for 0 to remove some of the bystanders from their queue. (But clues for 0 are terrifying, because if they go wrong they need a very specific small clue to fix, so probably not that.) I can implement this preference in my spreadsheet by raising the badness of bystanders.
-Definitely cluing for an even number, because my team said that would confirm that lzahray, Troy, and ahkim are blue. Also, this lets me stop evaluating those peoples' names, which should save me some time.
-Jackie: "I wish I could ask whether that's an annoying constraint or whether you'd rather be more constrained." I think this is about right. I could probably deal with a ternary constraint easily enough that it'd be worth it; but unfortunately, most situations don't come in threes. Maybe mod 4, but where one of the options is "decline to use this channel"?
-Jackie "Can you please give something 0/1 mod 4 if you'd like us to try to get 2+ bits out of your clue #s instead of 1 bit, and 2/3 mod 4 if you'd like us to get at most 1 bit out of your clue #s?" Ooh, this is a tough choice. Today's clue would be better at 4 than 6; 6 means including DANCE (since you use your LEGS) and PENGUIN (since their LEGS are distinctive). My ideal number of bits to give out in a clue is something like 1.5, which doesn't fit either of those options.The fact that I'm having trouble getting to 2 mod 4 now is a sign that maybe I should avoid committing to giving two bits in the future. It'll also be harder to give two bits in the future since I'll be cluing for smaller numbers, and changing between, say, 2 and 4 for a clue is harder than changing between 4 and 6. But one extra bit is a big number. What might they use it for? They should only need one or even 0 yes/no question(s) tomorrow to confirm the whole blue team, then one or two yes/no questions to confirm the whole green team; they'll probably start using it to check things on their queue, and that seems a bit more against the spirit of the game. I think I need to be 2 mod 4.
ixa's being proven red:
-jasonye, I hope you don't feel too bad about it; not a big deal, and mistakes happen.
-I think red benefits most, from having a proven person to coordinate through, and a competent one at that.
-Green probably gets nontrivial color information and learns a target.
-My blues already have correct guesses for almost everyone (cmcclena and lukesci switched), but they were unsure enough about ixa that they were thinking of firing at a name of hers (not a True Name) today, so it saved us a shot.
-I think the pick-one-of-three thing was a clever compensation and about right.
-jasonye picked FOOD for 0, HP LOVECRAFT for 4, and VIRUS for 3. Python vetoed VIRUS (my middle choice) and I vetoed FOOD. I think HP LOVECRAFT is the easiest one to misinterpret, since most of what jasonye wants from it is just HP.
|Author:||achester [ Thu Dec 28, 2017 10:42 pm ]|
|Post subject:||Re: achester's thoughts|
Day 3 thoughts a bit more scattered because I didn't have enough time today (and, unfortunately, it shows in that I missed one or two relevant bystanders).
LEGS really wanted to be for 4 instead of 6. My team's strongly-highlighted ones are the four strong ones and the one bystander I knew I'd hit. Then they have another 26 weakly highlighted, but the only one of them that's correct is DWARF, which was already clued. No great harm today, but might make their clue worse tomorrow. Maybe I'll clue LEGS for 0 once they've shot at the strong ones.
Most of the clues I came up with today were better even. Giving one bit out through that channel is still probably worth it, but today was worse than expectation for it. Also, gosh, is MINT hard to clue.
-MINT strong enough that they don't waste their shots on other things in their lukesci clue.
-Odd, so they know ksedlar is bad.
"we got shot yesterday by Adam, dgrazian, ixa, yyao
Jackie • 1 min
David Corwin (firstname.lastname@example.org, email@example.com, firstname.lastname@example.org, email@example.com)
Ksedlar shot me
David • Now"
David: "Ksedlar shot me" (Yes, but it was on day 1, and for CHINA, which was as plausibly clued by BANGLADESH than by RED SEA.)
Jackie: "also lisa+corwind do you have esp or wit?
oh what'd ksedlar shoot?"
FYI I'm asleep just going to bathroom lol"
Troy: "What did ksedlar shoot? (Then mute us for sanity)"
David: (no response)
David, what are you doing?!
ahkim: "...i have a really silly idea
the last shooter is the only person to know if someone is dead, right?
what happens if we shoot one of our teammates' last word ourselves? will the other team continually waste shots trying to kill us?"
I had this idea too.
lzahray, to just me:
"I don't know i feel like jackie's putting too much weight on what people have claimed to her. Plenty of people claimed things and then turned out not to be those things
I think katie is green
guess we'll see
ok maybe not forgot about her shooting david "
Yes! But don't send this to just me! Jackie's mostly doing a great job running blue, but this is indeed a mistake she's making, and if you pointed it out she might do the Bayesian calculations (e.g. that even of the claims she's witnessed, a lot fewer have been correct than would justify her 3x or 6x updates) and fix it.
green has 14 HP.
red has 14 HP.
blue has 13 HP.
green has 11 HP.
red has 10 HP.
blue has 12 HP.
lzahray: "I think I like knife better since it kind of would make since why frodo was the character clued? But happy to go with what the most people think
Oh noes what if I'm wrong about the drop and the knife? I'm sorry!!"
Be more confident! Yes, this was one of two reasons it was specifically FRODO. Thanks!
Lzahray: "I'm watching lord of the rings now for research purposes, entirely due to my codenames dedication - bilbo indeed was the thief" Thanks for your dedication! Pity you didn't get to the scene with the SPIDER.
Win chance estimates:
Red has sort of fallen apart: cmcclena didn't take actions, jasonye's clue went poorly, and he's doubling down on it with HEWLETT PACKARD, which I think is a mistake.
Blue's doing pretty well; the top of their queue is mostly good, but three bystanders narrowly made it onto their list for today and I already worry about GLACIER 3.
Green's terrifying; they got 4 shots and are organized.
|Author:||achester [ Fri Dec 29, 2017 4:29 pm ]|
|Post subject:||Re: achester's thoughts|
Win probabilities 1:2 against blue, 4:1 for green, 1:128 against red (summing to more than 1 because one of the most likely outcomes is a blue-green tie).
Thoughts for tomorrow:
At the end of today, the health totals of members of green will be:
ksedlar 2 (TAIL not even enqueued)
brunnerj 2, 1 enqueued
Lucy 2, 2 enqueued.
I think I probably have exactly two more clues: green would need 5 hits/day for the next two days to win with only one more clue, and on no single day have they managed 5. On the other hand, they'd only need 3, 3, and 4 hits in the remaining days to win with two more clues; 3 seems like par, and the hit rate should increase slightly since there are things already clued and everyone'll be pretty confident of everyone's colors. Similar calculations apply for me, except that I already know I can't finish in two days because green will have 8 HP left at the end of today. So my goal is entirely to make sure I finish using exactly two more clues (and 12 shots, hitting at most 4 bystanders).
The most important words are LITTER and FAIR, the last two I haven't even clued, and MINT and TRAIN, which my team hasn't seen at all. Those are on four different people. The remaining ones are PASS (2nd in queue), SNOWMAN (4th in queue), CHEST (6th in queue), and TAIL (not in queue, I think due to spreadsheet error). So the easiest way would be for me to give two very precise clues for 2 for the four most important words, let my team hit the four bystanders and three True Names on its queue, and hope they realize TAIL.
To make that easier, I hope to kill a few people tomorrow. The easiest people to kill will be pravinas and lukesci, because they'll have only one HP each (so even if my team doesn't take multiple shots at the same person, they could die). Lukesci is also the highest-priority target because two of the four bystanders in the queue are his (FIRE and LEPRECHAUN). That'd mean cluing MINT and FAIR today, by somthing like CONDITION 2, and TRAIN and LITTER tomorrow, by something like VEHICLE 2 or CONVEYANCE 2. But that leaves me no room for error: if, say, I need to reclue TAIL, I'd need to fit it in precisely with TRAIN and LITTER.
Another possibility is a weaker clue for all four important ones: GROUND 4 gets FAIR perfectly, LITTER (because it's on the ground), MINT maybe (because the plant grows in the ground), and TRAIN maybe (because it's ground transit). But it'd hit BERRY as strongly as MINT (although I've clued MINT before), it'd hit CHURCH and FALL, it might hit PIN, SPRING, POUND, and SPOT, and, oh, it's illegal because it's someone's name.
If I do CONDITION for 2, I think it's probably good enough to hit precisely those two, killing lukesci and pravinas. My team would definitely, if they stick to their queue, fire at CHANGE (miss) and SNOWMAN (hit). Then they'd probably fire something at ksedlar (and their queue for her is empty; I just have to hope the need to fire at something on her makes them reexamine ksedlar and see that TAIL was strong and unadded) and then fire at someone a second time; if they follow their queue that'd be PASS for brunnerj. Then Lucy would be alive but with DANCE, her last, at the top of the queue; ksedlar would be alive with TRAIN (weakly clued by TITANIC, but certainly in need of a reclue), and brunnerj would be alive with LITTER (unclued), and I'd clue CONVEYANCE 2, CARRIAGE 2, or something like that and finish. This goes wrong if they don't follow their queue exactly, which is unfortunately quite likely because jamb says she knows the queue isn't very organized right now but my team doesn't seem to realize it's close enough to the end of the game to be relevant. Also, tomorrow they'll probably still be asking for parity bits to indicate team membership (today's is reasonable because cmcclena hasn't fired on them, which is confusing, but they really shouldn't need tomorrow's); that might mess up this plan.
I suppose another path for today would be to come up with a clue for 2 that's vague enough that it applies to three of the important four words so they shoot at all 3. FINE 2 might do that? But that hits lots of bystanders, and probably any clue meeting those criteria would hit a lot of bystanders.
I think I'll go with CONDITION 2, and the status of the game is that Python and I are both finishing after two more clues (so, the end of day 6) unless something goes wrong, where I have a lot less room for error if anything goes wrong than he does. Each of us should, at worst, finish at the end of day 7; red looks like it'll probably need until the end of day 7 or even 8, so not a factor.
Thoughts on today:
*grumble*, I didn't notice CHANGE for GLACIER; that's a reasonable possibility.
Ack; my team hasn't noticed GLACIER for MINT. I knew that was a risk, but they're mostly being pretty good about checking all possible connections, and if you Google GLACIER MINT you get good results. (To be fair, it isn't a strong enough thing to make it onto Wikipedia's glacier disambiguation page.) Jackie's being epistemically good:
"I notice that I am confused by the following things:
1) We got an odd clue, so either our team is wrong, or ksedlar is green.
Jackie • 1 min
Jackie Bredenberg (firstname.lastname@example.org)
2) Our team has been mostly proven - the 4 of us for sure, the rest by witnessings twice. Additionally, nobody else has made serious claims.
3) If we remove people we think are on our team and dgrazian, today's clue needs to be a subset of PENGUIN, SNOWMAN, CHANGE, SPRING, none of which seem strong enough to make it glacier, though I guess they could be."
4) If dgrazian and ksedlar were green, we'd have way too many green people.
(It is in fact 3, because she hasn't noticed that GLACIER is a brand of MINT, and I'm not sure anyone else is going through carefully enough to notice. They also haven't noticed SEAL, though.)
Ooh, Jackie's doing the Googling! "Things that do not have substantial Google results: "GLACIER TRAIN, GLACIER TAIL, GLACIER TURKEY"... come on! Persist! Make it to GLACIER MINT and this will all make sense! Ah, darn, she got distracted, and maybe she was only doing that for Katie. Oh, now she's back to Googling, but still only for ksedlar's words. (I guess it's reasonable to focus on ksedlar's words because today's clue parity proved her green, and I did want to clue something of hers strongly today to make my team a bit less suspicious of corwind and dylanhen.)
Also, I wasn't paying attention to corwind's and dylanhen's words any more, because it's too much to keep track of and I'd hoped my team would just trust them by now.
My team is obsessing too much over the fact that GIANT was intact, worrying that lukesci is not green. It's one word, and it was in the first clue (when cluing green bystanders was slightly positive for giving my team color information), and lukesci has not shot at any blues in three days. (jamb, later: "maybe we should consider that Pesto might be sometimes cluing people who are green to tell us they're green, regardless of if those are actually their words?" Yay!) I wish they'd just sit down and list out all the things they know about him or something.
A mistake I suspect my team is making and doesn't realize: they have a queue tab of the spreadsheet into which they've copied a lot of names, but not all of them, not even all the ones that they're most suspicious of from their spreadsheet highlighting. I don't think they're going back to check the spreadsheet highlighting, though. In particular, they correctly realized that TAIL was strongly clued by LEGS, but didn't transfer it to the queue document.
jamb, on ESP decisions for today: "that's fair... PENGUIN is not so strong that it gives us much info though
so if you wanna esp something of his, PANTS?
my thought about SPRING was that knowing the esp result would be particularly helpful in figuring out if we had solved GLACIER or not though
since it's only for 3
so I would rather do that one than spider"
(Both of those changes are from actually-damaged words to strong bystanders. My poor team will be so confused. )
MEDAL/METTLE/MEDDLE/METAL homophones came up in a crossword I was doing today. I kept a straight face and pretended to be slightly confused, because they're still on my list of possible clues.
SNOWMAN didn't make it into my team's top 6 for today; DROP and AGENT both beat it. (AGENT is such an easy word to clue by adding -ER to almost anything; it seems to have made it onto the queue on the theory that FRODO is an AGENT.)
I'm generally in support of the strategy of maximizing clue throughput and letting color information sort itself out later, but my team's underconfidence about colors is starting to really restrict my clue throughput, since they if I give clues that aren't dead-strong for the greens they're not confident about, they miss them and spend a lot of time thinking about the blues that they're underconfident about, which include some good bystanders because I'm just ignoring all the blue words now. (At least they started firing at lukesci! )
After two days of some difficulty with parity requests (not that they aren't a good system), today's looks like it'll be easy; all five clues currently at the top of my list are even.
I wish my team had a longer queue. It currently only contains 7 words. I'd like to remove some things from it, but I'm not sure what they'd hit next. Maybe cluing for 0 and removing a few things from their queue would make them reexamine it and thereby solve the problem that they forgot to put TAIL on it? There are still a few True Names I haven't clued at all yet (LITTER, FAIR) or that my team doesn't think I've clued at all yet (TRAIN, MINT), so I think I should do those instead of queue-clearing for now.
|Author:||achester [ Sat Dec 30, 2017 2:39 pm ]|
|Post subject:||Re: achester's thoughts|
Day 5 thoughts
HP left at the start of the day:
Green: 8 (4 at the end of today)
Green is almost certainly finishing in two days; odds 2^7:1 that they win. Red is almost certainly not; odds 1:2^3 that they win. I think blue's chance of also winning is about 1:2, depending on which mistakes they see and how they correct them.
The TAIL saga:
Background: My team correctly identified TAIL as strong for LEGS, but then forgot to transfer it from their day 3 tab to their queue tab and so forgot about it, and this has been a significant issue for me.
jamb: "well, sads that we seem not to have killed anyone, pending dylan and corwind
in other news why are we not getting clues for ksedlar?
we didn't kill her and she's provably green"
lzahray: "yeah she's also not at the top of our queue at all"
Troy: "Maybe we're dumb?"
jamb: "but ideally Pesto would be contending with our dumbness and helping us win anyway"
Yep, that's what I'm trying! This is exactly what I hoped you'd think, and so check your spreadsheets and realize that you actually do have things for her that didn't make it into your queue.
jamb: "I know we're not doing a great job going through previous clues, but we searched through them pretty thoroughly and they're on our queue, right?"
me: *is very sad, but with a straight face*
me: *is then very happy, because it means that two of the words I'll most need to clue are LITTER (the last unclued word) and TAIL, for which KITTENS is a perfect clue*
jamb: "Huh, it looks like we never put TAIL on our queue after LEGS. That was bad. I'll stick it on our queue".
me: *is very happy, but with a straight face*
me: *is then very sad, realizing that this reduces to close to 0 the chance that KITTENS will actually be a strategically optimal clue tomorrow*
jamb: *puts TAIL on the queue after SPINE*
me: Augh! No! SPINE only went on the queue at all just a few minutes ago because you couldn't think of _anything_ for the ksedlar queue and needed something; it's only clued by CONDITION (and you correctly realize it's weaker for that than FAIR or NURSE, which you're _downgrading_ because they're outside the top two for that queue); TAIL was actually strongly clued and you realized it!
me: *is then tentatively happy, realizing that this might make KITTENS strategically optimal again*
(Only tentatively happy, because I'm not sure it'd be enough to win, and as much as I want to clue KITTENS, I want to win more.)
I clued CONDITION 2. There are five words to think about:
I intended MINT and FAIR. (Oh, maybe UNDEBASED'd've been better? Well, too late now.) I think DOCTOR and NURSE are worse than FAIR (because, e.g. HERBALIST could get things off in that semantic direction), but ok. Then jamb realized that TRAIN fits well, which I didn't realize, but I'd've actually had to clue CONDITION 2 either way, because my team requested an even clue.
Someone on my team (Troy?) realized yesterday that parity requests mean there might be an extra True Name that fits the clue that I couldn't include. Today, my team has decided that anything below the top 2 for CONDITION is not worth shooting at, even though the same point applies. (They've forgotten it.) They're even spending a lot of time thinking about espionaging things clued by CONDITION to know to not shoot at more.
In Live Action Codenames, it was an advantage to be the target of the losing team, because your best solvers wouldn't die, and you could optimize for winning and stopping your main competitor simultaneously. In Dead Action Codenames, I think it's been a disadvantage for us to be the target of the losing team, because being shot at no longer kills you, but does give you information. For instance:
-All the blues have been shooting at greens, but not all the reds have been shooting at us (cmcclena didn't shoot for the first three days, some other reds didn't shoot on the first day, and more of red's shots went to non-blues), so green had better color data.
-Each team knows their own health. The most relevant healths to know are red's (because it'll end the game) and green's (because it'll determine whether we also win). Blue knowing its own health is irrelevant because red won't win.
-Strong clues on greens help greens identify each other; those existed early. Strong clues on reds help reds identify each other; those existed starting on day 2. Strong clues on blues would've helped blues identify each other, but the clues on blues haven't been as strong.
There will be four people alive, each with one HP, with the following queues:
brunnerj - FLY > PILOT > other (*LITTER*)
ksedlar - SPINE > *TAIL*
Lucy - NURSE > *DANCE* > FALL
pravinas - *FAIR* > SPOT > other
So, it'd definitely help to clue LITTER and TAIL to promote those two on the queues. (And LITTER's the last word I haven't clued at all, so I need to do it strongly.) That leaves Lucy and pravinas. NURSE and FAIR my team has said they'd demote if:
-(version written in my team's queue tab) at least 2 of DOCTOR, CHANGE, and TRAIN espionage as broken (which won't happen)
-(version discussed in chat that was mistranscribed to that) at least two of DOCTOR, MINT, and TRAIN espionage as broken (which would happen)
-(observation from chat yesterday that they haven't remade) not, because the clue could have wanted to be for one more and been constrained by the parity request.
So, I don't really have any idea what they'll do with their queues for Lucy and pravinas. But it looks like for just about any interpretation, the correct answer should be in the top two. So, if their interpretation of being clued for someone is that the clue is for their last True Name (which I think is probably the right interpretation), then they'll have two shots to make on each of Lucy and pravinas, and we'd finish tomorrow, as we need to.
One thing I need to be able to do is clue someone's last True Name strongly and have my team shoot at just it and not waste extra shots on the same person. My team was thinking of shooting at FIRE in addition to MINT for lukesci, but realized they shouldn't for this reason; that's one realization that'll make my job tomorrow a bit easier.
jamb, last night: "so if we want to write a parity check I think it should be of the form:
for all subsets of people who could be alive tomorrow on green team:
<queue 0> <queue 1>
clue even if you prefer queue 0, odd if you prefer queue 1"
Sure, this sounds like a great idea.
jamb, this morning: "I would honestly just go for trying to interpret Pesto's clues as best we can and not constrain them"
Ok, not constraining my clues is reasonable (although I could give out a bit), but I at least need some context for your clue-interpretation: e.g. whether you plan to fire more or less at clued people. I could work with just about any convention, but I need something. (Sort of like in regular Codenames, where stating any queue is better than stating none, or in Hanabi, where you can work with any well-defined convention as long as you have one.)
Anyway, my clue absolutely needs to hit LITTER dead-on because I haven't clued it yet. It also needs to get at least one of TAIL, DANCE, and FAIR. KITTENS gets TAIL, FLOOR gets DANCE, and GROUND gets FAIR, but KITTENS gets LITTER strongest. I don't see a good way to get even three of them, and I don't even know which three I'd most want to hit because of the unclear status of the queue. So, KITTENS it is.
Last thing to figure out is whether there's a way to clue LITTER and TAIL that doesn't also get STRING. SLEEPY_KITTENS? (Illegal.) FURBALLS? (Not strong enough for LITTER.) FURRIES? (Not strong enough for LITTER.) Hope my team decides that LITTER+STRING=CONFETTI, TAIL+STRING=SUFFIX, and TAIL but not STRING had been clued before (but was still in need of cluing because it wasn't on top of the queue)? I worry they might ignore such arguments on the grounds that "Pesto would clue KITTENS anyway". Even if they miss that and shoot at all three of them to be sure, they could still guess correctly which of pravinas and Lucy to shoot at twice. (Which one is correct to shoot at twice depends on the resolution of the queue mistake.)
I think that's the best I can do.
I really hope we win on KITTENS; that'd be great. I really hope we don't lose on KITTENS, because then my team might think KITTENS was suboptimal but that I couldn't resist cluing it anyway.
|Author:||achester [ Sat Dec 30, 2017 10:51 pm ]|
|Post subject:||Re: achester's thoughts|
Addenda to day 5 thoughts
Now lzahray has decided to shoot at TAIL (correct, killing ksedlar) instead of DOCTOR (incorrect) or TURKEY (incorrect), giving me another pair of emotional swings in opposite directions: made happy at another hit, but sad that KITTENS is again probably not strategically optimal.
Ooh, and lzahray did some repair work on the queue. It's now the case that they'll push DOCTOR, NURSE, and FAIR way down the queue if mint and train espionage as both broken (which they will). That'll make the top of Lucy's queue be (correctly) DANCE, and I just need to clue LITTER (really strongly) and FAIR (maybe not as strongly).
Added complication: corwind said that about 3 hours before day end he's getting on a trans-Pacific flight. It's now less than 3 hours before day end, and he didn't submit his assigned shot, which is PASS. So brunnerj will have two HP left. I guess I have to hope that they shoot at PASS tomorrow anyway, thinking I wouldn't know it hadn't been shot at? That's an extra reason to make the clue unambiguously hit LITTER, though, so my team doesn't think that brunnerj should already be dead and so doesn't need to be clued.
brunnerj - FLY > PILOT > other (*LITTER*)
Lucy - NURSE > *DANCE* > FALL
pravinas - DOCTOR > *FAIR* > SPOT > other
SUPREME CAT SHOW
HP at end of today:
green 3 (corwind made his shot! )
|Author:||achester [ Sat Dec 30, 2017 10:53 pm ]|
|Post subject:||Re: achester's thoughts|
By the way, I considered vetoing SCOTTISH AUSTRALIAN (and knew I had the right to), but let it through because I wasn't scared enough of it.
|Author:||achester [ Sun Dec 31, 2017 9:40 pm ]|
|Post subject:||Re: achester's thoughts|
My team did it! There was a close call today with corwind (who was assigned to shoot the last True Name) saying in chat he'd shoot and then not submitting the form, but it resolved when he submitted half an hour later.
Red made a valiant attempt to catch up at the end; as of 20 minutes before day end blue had only 4 HP left, so red would also have had a good chance of finishing tomorrow.
|Author:||achester [ Sun Dec 31, 2017 9:46 pm ]|
|Post subject:||Re: achester's thoughts|
My personal spreadsheet's at https://docs.google.com/spreadsheets/d/ ... sp=sharing .
|Page 1 of 1||All times are UTC - 5 hours|
|Powered by phpBB® Forum Software © phpBB Group