D&D.SCP: Anomalous Acquisitions
post by aphyer · 2022-02-12T16:03:01.758Z · LW · GW · 25 commentsContents
STORY (somewhat longer than usual, still skippable) DATA & OBJECTIVE NOTE ON SOURCE MATERIAL None 26 comments
STORY (somewhat longer than usual, still skippable)
When you graduated top of your class from Data Science School, you didn't care where you ended up working, you just wanted to find the highest-paid job possible. (Your student loans may have had an impact on this decision).
(Yes, I know a past scenario background [LW · GW] has also begun this way. It's a versatile backstory. And one that I think may resonate with many of my players.)
Somewhat surprisingly, at the last minute Goldman Sachs was outbid by a firm you'd never heard of before called "Marshall, Carter and Dark Ltd." They hired you (for a substantially higher salary than anyone else was offering) to analyze sales data for "our ultra-exclusive chain of clubs, facilities, and banking services, catering to the finest gentlemen since 1952".
You've been working for them for several years now, in the course of which you've seen a variety of...very strange objects pass through their firm. It seems that, in addition to their upper-class business, they acquire and sell a variety of items that don't strictly seem to obey the laws of physics.
In the course of the last few years, you have excelled beyond your employers' expectation at all the technical aspects of your job...while perhaps performing less well on the social aspects.
It might have been turning down an invitation to drinks with your boss Mr. Alden because you don't like alcohol...
...or spending forty-five minutes explaining a particularly interesting piece of your data methodology to a client who had asked an innocuous question (how were you supposed to know he would be such a philistine as to find that boring?)...
...or asking a few too many questions about the various strange items that passed through the firm...
...or (what seemed based on your colleagues' reactions to be worst of all) pointing out that when you acronym "Marshall, Carter and Dark" to "MC&D" it looks very like "McDonald's".
Whatever the reason, when you were called into the office of Mr. Marshall for a private interview, you assumed you were going to be fired.
(You weren't too worried. You're a data scientist. You can find another job in less time than it'll take you to leave the building. And given the severance package in your contract, firing you would basically amount to paying you a large bonus.)
With your usual level of social acumen, as soon as the door to Mr. Marshall's office closed you asked him: "Am I being fired?"
He looked surprised to hear you say that (which would have fooled you three years ago, but since then you've learned that the top-tier client-facing staff of MC&D change expressions the way you change clothes.) "Fired? Of course not! Mr. Alden speaks very highly of your technical ability."
"My technical ability yes, my non-technical ability...uh...not so much." (You're sure a salesperson would have said that in a better-sounding way).
Marshall puts on a smile that you couldn't imitate if you spent months in front of a mirror. "Quite so! In fact, that's why you're here." He steeples his hands and leans forward slightly. "It's come to our attention that you're being...under-utililized in your current role. As you say, your salesmanship skills are...perhaps a point for future improvement. However, your technical skills are top-notch, and we think they're being wasted in your current position. And as it happens, we have a recent opening as head of Acquisitions that my colleagues and I think would benefit from a more data-driven approach."
You blink in confusion. It sounds like they're...promoting you? That can't be right...can it?
"I'm pleased to inform you that this position would come with a 20% base salary increase over your current compensation, with the additional potential for a noticeable element of performance-based bonus compensation. Additionally, I know you've shown interest in some of our more esoteric offerings," he smiles again, "and I'm happy to say this position will offer you a role dealing directly with them, which I'm sure you'll find fascinating. If you'll sign this non-disclosure agreement, I can give you more detail."
You sign, and he smiles again. It might be your imagination, but it looks like a different smile, with less warmth, like the smile of a shark that's just wandered into the middle of a school of tasty, slow-moving fish.
"Marshall, Carter and Dark is a globally connected firm. Among the advantages this lets us offer our clients is our knowledge of certain esoteric artefacts that act in manners beyond our current understanding of physics - or, in some cases, in manners outright foreclosed by our current understanding of physics. We seek out such objects from across the world and acquire them for our clients, or in some cases for direct usage by ourselves. While this is a source of revenue for us, and your performance as head of Acquisitions will be evaluated principally by revenue generated, the true benefits lie largely downstream from that - the connections thereby formed and relationships generated with our clients provide returns in excess of the initial sale price."
Blah blah client relationship blah blah. You get it. How do you find these anomalous items anyway? Perhaps some kind of analysis of mana flows underlying the world? That sounds like the sort of problem you could turn your data science skills to.
"The majority of such anomalous objects are secured before we can identify them, by an extralegal organization referring to themselves as 'the SCP Foundation'. They consider it their mission to 'Secure, Contain and Protect' - to locate these artefacts, priceless relics that offer unique capabilities or even a chance to advance human technology, and sit on them lest anyone do anything productive or profitable with them. As such, the main role of Acquisitions lies in identifying items currently in the custody of the SCP Foundation that we can realistically acquire and put to better use."
"Wait, you're stealing them?" Surely he can't be trying to-
"I would prefer to say we're using these objects in a more efficient manner. The market indicates a substantial demand for these items, and by removing them from circulation the Foundation is-"
"No. This is obviously illegal. I'm not helping you do it. I'm calling the police."
He's still smiling, and it's freaking you out. "I do seem to recall that less than two minutes ago you signed an agreement not to disclose-"
"An NDA to cover up a crime isn't enforceable."
He sighs, and sounds genuinely disappointed. "A crime? What crime?"
"You just told me-"
"Told you that you were going to help steal 'magical' objects from a clandestine organization no-one has ever heard of. Strange physics-breaking artefacts that no-one believes exist. By all means. Call the police. I'm sure everyone will be quite distressed to hear that a valued employee of Marshall, Carter and Dark has suffered a nervous breakdown." He sounds genuinely sympathetic. "It must have been the strain of the job. Of course we would never dream of overworking our employees, but the young always think themselves invincible. Don't worry, officers - under the circumstances Marshall, Carter and Dark won't be pursuing any cause of action for slander."
You look down at the phone in your hand, with a slow realization that maybe you did not entirely think this through.
"Secondly, I think you are underestimating the capabilities of Marshall, Carter and Dark in a litigation environment." He smiles, and it's all teeth, white and cold as a winter mountain. "Under the choice-of-law clause in both your employment agreement and the nondisclosure agreement you just signed, you agree for issues between you and the firm to be resolved first by our in-house arbitration committee, and should that fail by escalation to courts in our corporation's home jurisdiction. The firm is incorporated in the Cayman Islands. There are a grand total of four judges in the Cayman Islands. Three of the four are former employees of Marshall, Carter and Dark, still drawing generous pensions whose value is based on the company's stock price. The fourth golfs with me on weekends." He lowers his voice conspiratorially. "I usually let him win, you know. It's good for building relationships."
"And thirdly." His voice has gone cold now, matching the smile. "Under ordinary circumstances our retrieval teams are employed to locate and acquire anomalous objects. Highly valuable artefacts, defended by a quasi-governmental agency with a vast budget, anomalous resources and weapons of its own, and very few constraints on its actions. And in more than half of cases, a team sent to acquire such an object succeeds." He leans forward, locks eyes with you. "While the firm would of course prefer to avoid any unpleasantness, should it prove necessary one or more of those teams can be redirected to...resolving a loose end before it can grow into a potential legal and public-relations issue for the whole corporation."
"Now sit back down."
"Good."
"Your main role as head of Acquisitions will be selecting targets for our retrieval teams. You see, despite our frictions with the SCP Foundation, we believe that they are doing good work. Many of the anomalous objects they contain in fact should be contained. We are interested only in that subset of objects whose value outweighs their danger. So by all means let them continue to contain the giant indestructible reptile driven solely by hatred for humanity. We would prefer to acquire, oh, say, the magical pills that can cure any illness."
"Sadly, our information about SCP objects is limited. We have access only to the Foundation's internal classification system, identifying objects by their danger level and a few tags. In the past this has led to certain unfortunate incidents where Acquisitions in fact imposed costs on the rest of the firm, to clean up the messes caused by acquiring something we rather shouldn't have."
"You will be supplied with a list of all SCP objects identified as targets by our teams, and we have also collated for you a comprehensive history of our past attempts and how successful they were. As I understand your latest soliloquy on the matter, that should provide enough information for you to employ data science tools, and hopefully to identify targets more effectively. I wish you more luck of it than your late predecessor."
"One last thing. Our firm's assets are our employees, our capital, and our reputation. If any of these is ever diminished, the last is the most difficult to restore. Marshall, Carter and Dark has grown on the back of our reputation as an acquirer of anomalous items. However, recent ineptitude in the Acquisitions department has led to very poor performance. Our firm's reputation has suffered."
"I and my fellow partners appreciate competence in our employees. Turn around the lackluster performance of the Acquisitions department, restore our firm's reputation, and we will see you well rewarded. Fail in this, and...well." He sighs. "Mister Glover is a major client, with several billion dollars invested with our banking subsidiary. He found your long explanation of recent machine learning advancements tedious in the extreme, and has expressed his displeasure to us. I'm sure a successful quarter in Acquisitions will give us interesting objects to demonstrate and renew his pleasant relationship with the firm. Failing that, perhaps he would settle for the ability to express his displeasure upon you in a more personal fashion."
"Security will show you to your new desk." He gestures behind you, where two men each larger than the room's doorway have somehow managed to enter without making any noise. "Don't let me keep you, I'm sure you're anxious to get to work!"
DATA & OBJECTIVE
- You have been allocated 9 retrieval teams to acquire SCP objects for your employers:
- 3 Infiltration Retrieval Teams
- 3 Legal Retrieval Teams
- 3 Paramilitary Retrieval Teams
- You need to select (from among those SCP objects identified as potential retrieval targets this year) a target for each of these teams.
- For the avoidance of doubt, you may not send two retrieval teams after the same SCP object.
- A valid solution might look like: 'Send infiltration teams to retrieve SCP-111, SCP-222 and SCP-333; Legal teams to retrieve SCP-444, SCP-555, and SCP-666; and Paramilitary teams to retrieve SCP-777, SCP-888 and SCP-999'.
- Your objective is to maximize the profit made from acquisitions:
- A failed acquisition makes no profit.
- A successful acquisition makes some amount of profit, which might be positive or negative.
- Your overall profit is the sum of all profits of all successful acquisitions.
- To help with this decision, you have been provided with a dataset of past SCP objects seen, MC&D retrieval teams sent, success/failure of retrieval operation, and profits/losses made. (The dataset seems to reach back further than you thought the firm had even existed for.)
- This dataset also includes a variety of 'tags' indicating yes/no whether an object possesses a given property. For example, the 'is_organic' tag indicates an SCP that is to a substantial extent composed of organic material, while the 'is_location' tag indicates an SCP that is anomalous location (a building? a town? a forest? a particular geographical latitude and longitude?) as opposed to an anomalous object.
- The SCP objects identified as possible targets this year are as follows (data format matching the main dataset is available here):
SCP | Foundation Site | Object Classification | Tags |
SCP-537 | Site 2: Washington D.C | Safe | Mechanical/Mobile |
SCP-1282 | Site 4: Moscow | Euclid | Infohazardous/Mechanical/Virtual |
SCP-1466 | Site 3: Kinshasa | Euclid | Humanoid/Organic/Mechanical |
SCP-1720 | Site 3: Kinshasa | Safe | None Applicable |
SCP-1785 | Site 7: Shanghai | Euclid | Location/Infohazardous |
SCP-1838 | Site 3: Kinshasa | Keter | Location/Predatory |
SCP-1970 | Site 2: Washington D.C | Euclid | Organic |
SCP-2116 | Site 7: Shanghai | Keter | Predatory/Mechanical |
SCP-2122 | Site 4: Moscow | Safe | Location/Virtual |
SCP-2178 | Site 7: Shanghai | Keter | Replicating |
SCP-2253 | Site 3: Kinshasa | Safe | Humanoid/Predatory/Mobile |
SCP-2325 | Site 6: Geneva | Euclid | Location/Mobile |
SCP-2603 | Site 6: Geneva | Keter | None Applicable |
SCP-2626 | Site 4: Moscow | Keter | Predatory/Mobile |
SCP-2628 | Site 6: Geneva | Keter | Humanoid/Mobile |
SCP-2699 | Site 6: Geneva | Euclid | Predatory/Replicating |
SCP-2719 | Site 6: Geneva | Euclid | Organic |
SCP-2720 | Site 7: Shanghai | Keter | Humanoid/Organic |
SCP-2797 | Site 7: Shanghai | Keter | Humanoid/Infohazardous/Organic/Mobile |
SCP-2883 | Site 7: Shanghai | Euclid | Infohazardous |
SCP-2898 | Site 3: Kinshasa | Keter | Humanoid/Mechanical/Mobile/Replicating |
SCP-2942 | Site 2: Washington D.C | Euclid | Replicating/Virtual |
SCP-2964 | Site 7: Shanghai | Keter | Humanoid/Mechanical |
SCP-3212 | Site 4: Moscow | Safe | Organic/Mobile |
SCP-3273 | Site 8: Tehran | Safe | Mechanical/Mobile/Virtual |
SCP-3279 | Site 7: Shanghai | Keter | Predatory/Mechanical/Replicating |
SCP-3339 | Site 3: Kinshasa | Safe | Organic |
SCP-3440 | Site 8: Tehran | Safe | Mechanical |
SCP-3577 | Site 3: Kinshasa | Keter | Location/Mechanical/Mobile/Replicating |
SCP-3597 | Site 4: Moscow | Safe | None Applicable |
SCP-3656 | Site 8: Tehran | Keter | Humanoid/Mobile |
SCP-3668 | Site 6: Geneva | Keter | Humanoid/Infohazardous/Mobile |
SCP-3699 | Site 6: Geneva | Safe | Location/Organic |
SCP-3781 | Site 2: Washington D.C | Keter | Mobile |
SCP-3850 | Site 4: Moscow | Safe | Location/Organic |
SCP-3936 | Site 4: Moscow | Euclid | Humanoid/Infohazardous/Organic |
SCP-4004 | Site 7: Shanghai | Keter | Location/Infohazardous/Organic/Mobile |
SCP-4026 | Site 4: Moscow | Euclid | Humanoid/Mobile |
SCP-4027 | Site 6: Geneva | Euclid | Mechanical/Mobile |
SCP-4036 | Site 7: Shanghai | Keter | Predatory |
SCP-4222 | Site 2: Washington D.C | Keter | Virtual |
SCP-4271 | Site 4: Moscow | Safe | None Applicable |
SCP-4370 | Site 2: Washington D.C | Safe | Location/Infohazardous/Virtual |
SCP-4390 | Site 2: Washington D.C | Safe | Virtual |
SCP-4412 | Site 8: Tehran | Keter | Mechanical/Mobile |
SCP-4424 | Site 2: Washington D.C | Keter | Humanoid/Mechanical/Mobile |
SCP-4449 | Site 3: Kinshasa | Safe | Mechanical/Virtual |
SCP-4550 | Site 4: Moscow | Euclid | Location/Infohazardous |
SCP-4565 | Site 4: Moscow | Keter | Humanoid/Mechanical/Virtual |
SCP-4579 | Site 2: Washington D.C | Keter | Mobile/Virtual |
SCP-4625 | Site 8: Tehran | Safe | Location/Predatory |
SCP-4654 | Site 4: Moscow | Keter | Location/Replicating |
SCP-4709 | Site 7: Shanghai | Safe | None Applicable |
SCP-4834 | Site 7: Shanghai | Euclid | Humanoid/Organic/Mobile |
SCP-4931 | Site 6: Geneva | Euclid | Infohazardous/Predatory/Mobile |
SCP-4957 | Site 4: Moscow | Safe | Mobile |
SCP-5058 | Site 4: Moscow | Keter | None Applicable |
SCP-5087 | Site 6: Geneva | Euclid | Location/Organic |
SCP-5117 | Site 7: Shanghai | Euclid | Infohazardous/Organic/Mobile |
SCP-5136 | Site 8: Tehran | Safe | Location/Infohazardous/Organic |
NOTE ON SOURCE MATERIAL
This scenario is based (per request from simon, winner of the PVP competition in an earlier scenario [LW · GW]) on the SCP Foundation canon.
A few brief comments on what you should and should not assume to hold from that canon in this scenario:
- 'Safe', 'Euclid' and 'Keter' work the same as in canon - they are levels of danger for an SCP, with 'Safe' being the least dangerous and 'Keter' the most.
- Specific numbers of SCPs do not match up with those in canon. If you see a row for SCP-255 in this data, you should not look up SCP-255 on the SCP Foundation wiki and assume that it is the same thing.
- I have attempted to make this scenario and dataset incorporate certain elements present in the SCP canon. No further detail on this point will be provided.
I'll aim to post the ruleset and results on February 21st (giving one week and both weekends for players). If you find yourself wanting extra time, comment below and I can push these deadlines back.
Thank you to abstractapplic for providing feedback on a draft of this! For clarification, abstractapplic has no inside information on the scenario and is still free to play it.
As usual, working together is allowed, but for the sake of anyone who wants to work alone, please spoiler parts of your answers that contain information or questions about the dataset. Given that this scenario is SCP-canon-based, this rule is more important than usual.
25 comments
Comments sorted by top scores.
comment by Randomini · 2022-02-14T00:43:42.954Z · LW(p) · GW(p)
Okay, this comment's for stuff I found. Will edit this post with things as I find them.
MC&D when they burn millions to acquire keter items https://i.imgur.com/GlTx4HQ.png
I guess they must be a loss leader, it's not like their only source of anomalous stuff is stealing from the Foundation.
I'm filtering out any items which have more than five bits on in the binary section. Two rows are likely corrupted by infohazards - I thought it might have been time travel, but seeing as they also have every bit set to 1 EXCEPT for the infohazardous bit, they stand out like a sore thumb. I found those first two by plotting SCP number VS time and noting the outliers. They're also the only times an item was acquired when no team was sent. There are others that I'm on the fence about but I think eliminating those with excessive intersections is a reasonable data cleanup decision. Plotting year VS row number has a few jumps back but time travel's a thing in this setting so whatever.
Not sure where the nonlinearity here comes from. Maybe the simulation has a certain number of SCP numbers assigned at any given time, and then randomly draws from them as targets.
Splitting cumulative profits out by both SEK classification and site number shows variations between sites, though I'm not normalising here for the number of times a site was targeted or the SEK distributions at any given site.
comment by abstractapplic · 2022-02-20T13:38:52.338Z · LW(p) · GW(p)
Findings:
- As noted by others, there are weird everything-but-infohazardous rows that don't seem to make sense with the rest of the data (and sometimes just don't make sense in general, such as by being acquired without a team being sent). I filtered these out.
- Attempts by Paramilitary units to capture Virtual assets NEVER work.
- (once you account for the above, success rates for Paramil/Infil/Legal teams start looking eerily similar)
- My predecessors never sent a Legal team to capture a Humanoid, or an Infiltration team to capture a Location. This makes intuitive sense, so I'll follow their lead.
- Infiltration teams work best in Sites 2 and 6; Paramilitary work best in sites 3 and 8; Legal work best in Sites 4 and 7 (Sites 1 and 5 are conspicuous by their absence).
- Infiltration teams make most of the profit, Paramilitary make a little, Legal has actively been losing money.
- The Safer (or, failing that, more Euclidean) objects are, the more profit they net. Aside from that, it's pretty difficult to predict how much a given anomaly will be worth, though it being Organic and/or Humanoid seems to help a little.
- There's a weird tension between the last two points. Being extracted Legally seems to actively devalue an anomaly compared to being extracted by other means, in ways I can't explain by correlations with success rates or other factors. Maybe Lawyers are skimming a lot of the good stuff off the top, while Infiltrators are much more loyal/terrified? Or maybe SCP is secretly really good at resisting legal challenges, and just lets MCD's lawyers win when-and-only-when it's over something they don't mind losing? I can't help but feel I'm missing something here.
- Paramilitary and Infiltration efforts seem to have become less profitable over time (while in contrast, Legal has always sucked)
- MCD's lifetime profits (ignoring extra costs like employee salaries etc, which we know are huge) is a little over 120 billion in today's money. That's a lot, but it's still kind of cute that a century of effort and the use of literal dark magic got them about halfway to Musk's current net worth, even with insanely generous assumptions.
Final allocations:
Paramilitary units target SCP-3339, SCP-4625, and SCP-5136
Infiltration units target SCP-4390, SCP-2719, and SCP-537
I was debating having the legal teams stay home, but if I point them away from Keter stuff they may actually help the company's bottom line for a change, so . . . they can target SCP-3850, SCP-3212, and SCP-4957.
Thoughts on the Long Game:
MCD's legal department are a bunch of incompetent, possibly-corrupt clowns who keep losing their company money. I should probably retain some lawyers of my own and see what they have to say about this NDA I signed ASAP.
comment by Randomini · 2022-02-12T22:59:49.449Z · LW(p) · GW(p)
...odd to see characters I created being used by someone on LessWrong, and references to my stories to boot.
I liked the surprises scattered throughout the dataset - which, for the curious, can be pretty easily identified by just scrolling through and noting which lines stand out.
woah info hidden behind a blackbox this is just like scp again
also I think MC&D has stolen a few items before they got their designations, and that the rows are in order of execution - which, when time travel is possible, is not necessarily temporal order.
also "We would prefer to acquire, oh, say, the magical pills that can cure any illness" marshall you haven't tried to get SCP-500 since 1927, and I know that one's canon in this puzzle because SCP-682 is also very conspicuously missing from the dataset
↑ comment by aphyer · 2022-02-13T00:32:28.930Z · LW(p) · GW(p)
I don't know what parts of it specifically you've written, but yes, I've gleefully stolen leveraged a lot of things from the SCP wiki and related works in writing this. Thank you for your assistance!
[DATA EXPUNGED]
↑ comment by Randomini · 2022-02-13T00:40:28.507Z · LW(p) · GW(p)
I wrote the wiki's MC&D hub, designed the modern MC&D logo, and coined the given names that you used here during a contest back in 2014. Thank you for making the puzzle - I have found some interesting things so far!
Such as the virtual predatory organic-mechanical mobile replicating humanoid locations, which are for sure accurate representations of the real world and are definitely not infohazardous, since the data said it wasn't.
↑ comment by Pattern · 2022-02-13T04:43:09.956Z · LW(p) · GW(p)
What's MC&D, and where can it be found?
Replies from: aphyer↑ comment by aphyer · 2022-02-13T05:14:00.432Z · LW(p) · GW(p)
Within the SCP canon, 'MC&D' is 'Marshall, Carter & Dark,' start e.g. here. I've tried to be faithful to the image in the story above - a quick summary would be 'very rich, very upper-class, and very very capitalist people who want to use SCP objects as things to sell to the rich and powerful.' Start with investment bankers, as they are envisaged by the far left, and then throw in a dash of SCP stuff.
If you want a few things to read, my recommendations for the short things you can read that are most MC&D-ish in my head are:
https://scp-wiki.wikidot.com/thank-you-for-your-time
https://scp-wiki.wikidot.com/scp-604 (trigger warning: yes)
↑ comment by Max Morawski (max-morawski) · 2022-02-13T17:00:26.227Z · LW(p) · GW(p)
In that case this is the closest I'll ever be to interacting with a celebrity, that's awesome. I love MC&D.
comment by SarahSrinivasan (GuySrinivasan) · 2022-02-17T15:55:13.415Z · LW(p) · GW(p)
Notes on found & todos:
Data cleanup reveals everyone is super sinister.
Searching for obvious anomalous data reveals there is likely one data-munging SCP and checking surrounding entries narrows its id to a very small range.
Checking for missing SCPs reveals the id of the above anomaly. Truncating data past various times suggests that the foundation finds and numbers SCPs in numerical order, at some point SCPs become inaccessible perhaps by destruction or superb lockdown or otherwise, at some point (how?) MCD becomes aware of them, and randomly (how?) known available SCPs are potentials during each quarter.
(Very WIP) checking for teams sent, seems very variable? But generally at least 2 of each? But not quite always? But now I'm suspicious that some of our data has been erased?
Next steps: evaluate teams sent for patterns and in particular missing teams. Model finding and losing access to SCPs. Figure out how we allocated teams in the past, otherwise acquisition and profit will be (?) unacceptably selection biased. I guess maybe at some point try to figure out how likely various acquisitions and profits will be and what to do about the incentives. ;)
I didn't make much progress on next steps. There does seem to be a large time-dependency on perf, so probably MCD is going through allocators pretty fast :( or maybe certain teams are really good at what they do and stick around?
↑ comment by SarahSrinivasan (GuySrinivasan) · 2022-02-18T08:16:35.192Z · LW(p) · GW(p)
My solutions. Probably won't get any reasonable chance to revise them.
Plan: become invaluable, see more data, find a way to escape eventually.
Legal, SCP-1282
Legal, SCP-2122
Legal, SCP-4370
Paramilitary, SCP-3212
Paramilitary, SCP-3936
Paramilitary, SCP-4834
Infiltration, SCP-4449
Infiltration, SCP-3273
Infiltration, SCP-1466
Plan: poison MCD against ever using "data driven" approaches.
Legal, SCP-3781
Legal, SCP-4036
Legal, SCP-2626
Paramilitary, SCP-1838
Paramilitary, SCP-2116
Paramilitary, SCP-3577
Infiltration, SCP-3279
Infiltration, SCP-2178
Infiltration, SCP-4654
Plan: try to escape, but try to be valuable given my immediate escape desires.
Legal, SCP-1282
Legal, SCP-4370
Legal, SCP-3936
Paramilitary, SCP-2883
Paramilitary, SCP-2797
Paramilitary, SCP-4004
Infiltration, SCP-3668
Infiltration, SCP-4931
Infiltration, SCP-5117
↑ comment by aphyer · 2022-02-21T02:20:13.610Z · LW(p) · GW(p)
It looks like your second 'poison' plan sends two teams (one Legal and one Infiltration) after the same SCP.
Replies from: GuySrinivasan↑ comment by SarahSrinivasan (GuySrinivasan) · 2022-02-21T03:00:03.553Z · LW(p) · GW(p)
:chagrin:
I have replaced the first legal team target with a... legal target.
↑ comment by SarahSrinivasan (GuySrinivasan) · 2022-02-18T08:36:24.253Z · LW(p) · GW(p)
Method, sadly, was just:
clean the data
compute smoothed fraction of acquisitions per attempts for each team type given set values for two tags; average over all pairs of set values
same but mean profit given successful acquisition
handpick best expected profit, worst expected profit, and best expected profit for infohazards that seem maybe useful or volatile
comment by Yonge · 2022-02-20T22:16:35.301Z · LW(p) · GW(p)
Infiltrate:
3273
4449
4027
Legal:
5058
3850
2325
Paramilitary
3440
2719
4027
comment by Max Morawski (max-morawski) · 2022-02-13T16:59:20.481Z · LW(p) · GW(p)
It is very rare that I can't resist being "that person", but Safe, Euclid and Keter refer to required containment resources, not danger level.
Replies from: aphyer↑ comment by aphyer · 2022-02-13T19:09:30.333Z · LW(p) · GW(p)
I don't think that's consistently true across the wiki? SCP-668 or SCP-2317 spring to mind as examples of objects whose classification appears to be based on danger level rather than on required containment resources? SCP-2521 is also a Keter-class object, despite the fact that its containment procedures consist entirely of
comment by Pablo Repetto (pablo-repetto-1) · 2022-02-21T17:25:17.565Z · LW(p) · GW(p)
I came in pretty late, so I haven't gotten much to share.
I split my analysis into expected-profit-if-obtained and chance-of-obtaining-per-team. It probably isn't true, but assuming that team selection does not directly affect profit simplifies things a lot.
- Predators incur losses, except Euclid humanoids which are profitable.
- Safe objects are profitable.
- Euclid objects are profitable, except locations and replicators which are neutral.
- Keter objects incur losses, except humanoid organics which are profitable.
- When objects we obtained in the past are split by this criterion, "unprofitable objects" incur a mean loss of 25.5 (stderr 1.85), "neutral objects" make a mean profit of 5.4 (stderr 2.23), and "profitable objects" make a mean profit of 22.6 (stderr 0.86)
- As abstractapplic pointed out, infiltrations do well on sites 2 and 6, legal teams on sites 4 and 7, and paramilitary teams on sites 3 and 8.
- Paramilitary teams never get virtual targets, but do well on organics.
- Infiltration teams do well on predatory and virtual targets (in particular, they do better than paramilitary teams against predatory Euclid humanoids).
Final allocations
- Infiltration: SCP-4370, SCP-4390, SCP-537
- Legal: SCP-3212, SCP-3597, SCP-1282
- Paramilitary: SCP-3339, SCP-3440, SCP-5136
comment by Measure · 2022-02-15T15:51:04.555Z · LW(p) · GW(p)
My method:
I assigned a score to each target-team pairing based on average profit gained/lost by that team type attempting similar items. Specifically, I summed a site score, a classification score and nine flag scores where each sub-score is the average profit of attempts by that team type for targets sharing that attribute. I excluded team-attribute pairings that are rare in the dataset (legal teams targeting humanoids, infiltration teams targeting locations), which I guess would predictably fail to acquire for obvious reasons. I also noted that paramilitary teams are extraordinarily bad at acquiring virtual targets even though they attempt fairly often. I would have excluded these if it was relevant.
I want to find the overall team assignment that maximizes/minimizes the sum of the scores for each target, but there are 24.8 trillion possible team assignments. To make the problem tractable, I first reduced the target set to "viable" targets, which are the nine top scorers for each team type. Some of these overlap, so this reduced the target list from 60 to just 13 minimal targets and 17 maximal targets (1.2 million and 70.8 million possible assignments respectively). I iterated over these reduced target lists to find the best/worst overall assignments.
Maximal assignment:
Infiltration: (SCP-1466, SCP-3339, SCP-5117)
Legal: (SCP-1282, SCP-3850, SCP-5136)
Paramilitary: (SCP-3212, SCP-3936, SCP-4834)
Minimal assignment:
Infiltration: (SCP-2116, SCP-2178, SCP-3279)
Legal: (SCP-2626, SCP-3781, SCP-4036)
Paramilitary: (SCP-1838, SCP-3577, SCP-4654)
comment by Measure · 2022-02-14T19:49:45.845Z · LW(p) · GW(p)
Possible data error:
1933Q2,SCP-1604 shows a profit even though the team wasn't successful. Is this an error?
↑ comment by aphyer · 2022-02-14T19:56:50.100Z · LW(p) · GW(p)
The row in the file is definitely what the dataset says.
To clarify a potential minor misinterpretation of the dataset - in actual fact that row shows a team being successful (a 1 in the column 'Acquisition Successful'), a profit being made (1.3M in the Profit column), but no team being sent ('None' in the column 'Team Sent'). This does indeed seem very strange. Nevertheless that is what the dataset your character has been given shows.
comment by Measure · 2022-02-14T23:20:04.197Z · LW(p) · GW(p)
I haven't finished my analysis yet, but Darke et al. will find that their particular mixture of positive and negative reinforcement yields rather different results from what they are expecting. I hope that my successor feels the same way.
Replies from: aphyer