ALSO, this is especially ridiculous and if it’s not a priority people should feel free to put it off? But she has four different remote-consoles at her workstation right now, even though one of her consoles theoretically has the capacity to control five different mechanisms, its default programming doesn’t cover the weird ones like the vasodilator-constrictor. Can…someone…maybe grab another of the standard consoles on the unit and…reprogram it? So that she can tie all of the machines that she’s controlling personally to one piece of hardware?

(She hasn’t actually dropped any of her remote-consoles on the floor yet but she keeps feeling like she’s going to, and the CAMERAS ARE ON, there are Very Serious People watching her.) 

OH NO Merrin has just managed to remember that she is green. And red-eyed. And wearing a bathing suit. AAAAAAAAAH.

She has already decided that she has limited mental space available and 'agonizing embarrassment' does not get to take up any of it. It's weirdly easy to just...do that mental motion...when the stakes are so obviously stratospheric. Merrin wonders vaguely if that trick generalizes. 

     "So the image our hospital is presenting to all these Very Serious People is that our top opper is a Sparashki in a swimsuit.  Now that things have quieted down slightly, I just thought I should take a moment and say that."

"Uh huh.  And?"

     "Are we comfortable with that?"

"Why wouldn't we be comfortable with it?"

     "I mean, I'm fine with our sentience-embracing employment practices, I think it's a great image, myself -"

"Okay, good, because for a second there I thought you were about to go all speciesist on me.  If our best employee for the job is a Sparashki, a Sparashki gets the job."

     "Look, I'm just trying to establish common knowledge about whether we are all, in fact, fine with this."

"I literally can't think of a single valid reason why we shouldn't be."

    "Combine it with the secrecy requirement, and Merrin's apparently superhuman endurance, and there could be people in those secret markets wondering if Sparashki are real and the whole standard-alien business is a smokescreen to hide any who go out in public."

"Almost certainly not.  But if anybody is actually wondering that, what part of that would not be unutterably awesome?"

     "Fair point.  It's just -"

"Just what?"

     "Why is Merrin an aquatic-stage Sparashki today?"

"I'm sure if we were cleared to know that, Exception Handling would have told us already."

It goes wholly without saying, at this point, that anyone at the hospital who knows about Merrin's aborted friend-date - or even the fact there's an Alien Trade Con today - is saying absolutely nothing about this fact to anyone who doesn't already know.

All of her fellow employees are vigorously maintaining to anybody outside the hospital itself, should the question arise, that Merrin has always cosplayed as a Sparashki while on duty, in fact nobody's ever seen her out of costume; sure it's a little odd, but lots of people are a little odd.

(This is not considered a lie, in that it would be universally understood and expected that no one in this social circumstance would tell the truth.)

Their patient is definitely making Merrin - and dozens of other people on site, not to mention thousands off site - work very hard for those precious hours. 

The human body is, in many ways, a pinnacle of engineering, an exhibit of how blind evolution can, eventually, surpass the abilities of even the best dath ilani engineers working together with the power of prediction markets. Second by second, hour by hour, hundreds of thousands of proteins (even dath ilan doesn't have a precise count) are expressed, in varying ratios depending on the cell type (of which there are around two hundred), and the surrounding tissue and, on a larger scale, organ. And an endless, fractally complex orchestra of signals, hormones in the blood, electrical signals echoing through neurons and muscle tissue, minute gradients in electrolyte concentration, and even more close-in, signalling molecules interacting in and around the complex organ-like structures found in every. single. cell. 

(The human body is, as per the usual design style of evolution, a spectacularly messy kludge of spaghetti code.) 

Put all of that together, and the result is a system that - with lower power consumption than a bright incandescent lightbulb - maintains a highly ordered internal state against the forces of entropy, corrects for perturbances, with interlocking control systems far beyond the capabilities of even the Basement's computing tech. 

And while thousands of different measures, from core body temperature to the exact potassium concentration in the blood, are usually maintained within absurdly narrow parameters, the human body actually has rather wide tolerances on what it can handle and recover from. A person starting out in good health can survive a complete lack of water intake for at least three days, or zero salt intake for - at least that long, possibly up to a month. With the first tiny flicker in sodium levels, still undetectable to the most sensitive medical equipment, and half a dozen internal alarms go off, the hypothalamus and pituitary gland react, and the kidneys do their work to hold the system stable as long as possible. The human body can adapt to outside temperatures of well above 45 C, as long as the humidity is low enough for sweating to shed heat. It can lose up to 40% of blood volume, and the heart will beat harder, the blood vessels will contract to raise blood pressure, and the patient will certainly be having a bad day, but if they have the cardiovascular reserve and aren't simultaneously facing too many other perturbances, will probably survive until the EMTs arrive. A patient with a body temperature as low as 30 C will be drowsy and confused, but probably still conscious, and able to regulate back to a normal body temperature with no treatment more invasive than warm blankets. The human body has an absurd degree of redundancy. One kidney, if it's healthy, is more than sufficient to meet the body's needs. 

The compensation mechanisms that come into play in abnormal situations aren't without side effects – high fevers do help the body fight infection, and are also not very good for brain cells – and when one system starts to fail, it taxes others, eventually exceeding their tolerance thresholds, and damage accumulates in a cascade through that intricate interlocking web. 

High blood pressure deals small but cumulative damage to blood vessels, hardening and scarring them, and after long enough, inflammation attracts plaque and clots, weak spots in major arteries bulge and stretch, kidneys gunk up and gradually lose the ability to compensate for a high-potassium meal, and sooner or later a coronary artery blocks off or a blood vessel bursts in the brain.

Eat one sugary meal, and a healthy pancreas will groan but put out a flood of insulin; eat a high-sugar diet above the maintenance calorie needs every day for a decade, and the cells will gradually respond with less vigor, even as the overtaxed pancreas starts to fall behind, and fat deposits (fat is far from inert - it's an endocrine organ of its own, in a way) secretes its own hormones, and baseline blood sugar creeps up and up – and, again, inflames the lining of blood vessels, gunks up and eventually cuts off circulation to extremities, gradually damages peripheral nerves, and makes a very tempting meal for any bacterial infection that starts to sneak in, ignored by an immune system unable to reach it through those sticky damaged capillaries. 

The power to regenerate organs is a mixed blessing; every time a cell divides, however high-fidelity the DNA copying, and however thorough and paranoid the cellular machinery for error-checking and error-correcting, there's still an opportunity for a new mutation. Most mutations do nothing, or just kill the cell, but some instead leave it deaf to the instructions from the rest of the coordination-symphony of biochemical signals, and the cell sets off on its own rogue mission. The immune-system police are paranoid and careful, and very few of these cellular criminals make it past their watchful eyes and ears, but with two trillion cell divisions happening daily, eventually something slips through the cracks. Despite everything they've thrown at reducing cancer risk, about thirty percentage of dath ilanis will be diagnosed with cancer during their lifetime (though usually caught early enough for treatment), and likely many more people die of something else before the rogue enclaves are big enough to show up on a scan. 

Because, of course, the body has a limited operating lifespan. Eventually, even if nothing specific goes wrong, the awesome regenerative powers start to fail one by one. Most people don't wait around for that gruesome slow-motion breakdown to reach its natural conclusion; the brain deteriorates along with other organs, and besides, it's unpleasant to experience. 

And that's only what it looks like when the control-system symphony breaks down slowly. 

It can happen very fast, from a single catastrophic disturbance, if the tolerances are exceeded drastically enough or affect enough systems at once. Drop even a tiny wandering clot in the wrong place, and the finely-tuned electrical orders conveyed across the heart will fall apart into chaos in seconds.

The healthiest adult can still bleed to death in less than five minutes – and even if the EMTs are on-scene in time to pour replacement blood into them, very little has to go wrong to set off a different failure cascade. If blood gases get out of whack, too-high CO2 leaves the blood acidic, with all sorts of nasty downstream effects. Confused clotting signals, aware that there's an injury somewhere but too rushed to coordinate properly, can spark a distributed clotting process everywhere at once, denying circulation to gunked-up capillaries and, more relevantly, instantly using up all of the body's reserve of clotting power. The stabilizers used in stored blood can even accumulate to toxic levels. 

It may take days to die of dehydration just from not having water, but one bad enough case of gastroenteritis, out of range of prompt medical help, and the fluid and electrolyte losses can hit critical levels in hours. Even with treatment, it's easy for the acid-base imbalance or low potassium to get out of hand.

An infection hits the bloodstream, and the immune system kicks into gear, marshaling its armies and supplies for a pitched battle - but a response that works very nicely at the local level can be actively unhelpful once it spreads too widely. With chemical orders flying left and right, coordination breaks down, immune-cell battalions overwhelmed by the confusion start attacking tissues indiscriminately, and the feedback loop of emergency alarms explodes out of control. Fevers can spike high enough to start denaturing proteins. Tissues everywhere send out chemical-radio alerts begging for aid workers, blood vessels helpfully dilate to open a quicker path, and oops, now the systemic blood pressure is 50/30. The heart tries to make up the difference, beating faster, but inflammatory cytokine messengers actually weaken the contractions of the heart muscle cells, and each beat accomplishes less and less. Inflammation and random friendly-fire damage to lung tissue leaves the endothelial cells more permeable, leaking fluid and blocking gas perfusion at the interface, and now the rest of the body, already burning energy at emergency levels, is facing a dropping O2 saturation in their blood supply. 

Trying to make up the energy shortfall, the endocrine system hears the inflammatory alert and responds by cranking up blood sugar - which maybe does some good somewhere, but also multiples the production of oxygen free radicals, and the oxidative stress on cells results in another round of inflammatory signals (yes, this is what is generally known as a positive feedback loop.) In desperation, energy-starved cells resort to less-efficient anaerobic metabolism, dumping lactic acid left and right, which adds on to the reduced CO2 clearance and sends the blood pH into a nosedive. Clotting factors respond to the generalized alarm of Something Wrong Somewhere and see if clotting it off is a helpful contribution to the war effort (spoiler: it really isn't, but they're trying their best.) Soon the kidneys and liver, overwhelmed and under-resourced, start to fail, adding accumulating waste toxins to the mix - and since the all-purpose-factory-extraordinaire liver is responsible for a huge fraction of the signaling hormones that coordinate all these immune forces, the chances of getting any of this back on the rails are dropping fast. 

Meanwhile, the central nervous system is having a bad time. On top of a protein-mangling fever, wildly outside-of-tolerance blood acidity, and a rapidly collapsing supply chain for oxygen delivery and CO2 collection, the inflammatory signals directly futz with neurotransmitter expression. Delirium, seizures, coma, and (if it continues for long) permanent brain damage are the inevitable result. 

These are some of the conditions under which the normally-resilient homeostasis control mechanisms of the human body are no longer particularly in control. 

These situations are still potentially survivable - survivable without brain damage, even - because a prepared medical team has a lot of options for taking manual control. Fevers can be lowered. Failing hearts can be supported with drugs, replacement electrical pacing, and, in extreme situations, mechanical force. Loose blood vessels can be sent loud, unmistakeable chemical signals to quit that right now, putting peripheral tissues on strict rations and prioritizing vital organs. (With the cost that if you push it too far in the other direction, the heart has to labor against more resistance, and also this is not necessarily good for fingers and toes.) Clotting signals can be regulated from outside, while used-up platelets and clotting-factor proteins are replaced. Insulin can be administered to keep blood sugar in a reasonable range. Kidneys can be replaced with dialysis, degraded liver function propped up via a massively inconvenient team effort.  

It's critical to intervene early, and fast, before the cascade of system failures gets very far. The levers available for manual control are a lot clumsier than the systems they're trying to replace, and reversing a system failure, while dancing the delicate dance of avoiding harmful side effects, is an enormously harder problem. 

This makes it especially important to predict when a patient is on track to deteriorate, before it actually happens. Fortunately, this is something that prediction markets, and a supply of experts around the world, are very good at covering. 

Of the factors that are currently making Merrin's job easier: 

- They are intervening pretty early. The patient may be in some degree of organ failure for just about every organ system, but they got the manual levers in place before anything could spiral very far out of control. 

- They're likely to see a new complication coming in advance, thanks to the insane liquidity on the diagnostic prediction markets. 

- The conditions for making the right judgement call on when and how to apply known treatment protocols are almost ideal, thanks to the liquidity and update frequency on the treatment planning prediction markets. 

- They have almost unlimited resources to throw at the problem, up to and including 'invent a new treatment protocol on the spot.' 

- The patient is cold. Which is of course causing all sorts of exciting problems by itself, but it does buy them a little longer to correct a mistake before the consequences spiral out of control. 

- They have explicit permission to attempt treatments with high risks of serious side effects - even potentially fatal ones - as long as the treatment planners and the markets think that it increases the likelihood of survival-with-brain-intact. They're even willing to risk minor permanent damage as long as it's not impairing, and major complications can be addressed later, once they know the outcome. If the guy needs a heart transplant after this, it's worth it. 

(Though Merrin is very, very badly hoping it doesn't end up being her call on whether to use a treatment protocol that ends with an organ being needed. She knows where the organ might come from, and she knows that the people who make that decision are making it for a reason, but it's not a reason that has ever made sense to her, and she doesn't like it.) 

Also on their side: before he fell in a river, the patient was in good physical condition. (Now that he's been ID'd, they have a medical history in the chart; it's heavily redacted on details, but they have the basics on his last decade of health screenings.) 

His non-brain organs have reserve to spare. Which is good, because while they're trying to match oxygenation and perfusion needs to the tissue metabolic rate, a blood pressure of 50/30 reaching the brain - which they don't want to exceed - isn't enough to get much circulation to his extremities, especially given that they're dealing with a complete lack of vascular smooth muscle tone, and resulting tendency for blood to just hang out permanently in his feet, by giving potent vasoconstricting drugs, and their ability to differentially address blood pressure in different parts of the body is very limited. The risk of major tissue death is small, but he could lose toes, and cumulative peripheral nerve damage is likely. Losing a bit of sensation in his feet is probably acceptable. His visceral organs are getting more bloodflow and oxygen, but also need more, and will take more damage from stretching this out for many hours. 

He might need a heart transplant. The mechanical-assist device gets blood pumped through the heart, and they can increase that if necessary, but it comes with the downside of applying compression to the coronary arteries right as the systolic pressure is rising. They're already detecting a flicker of elevated troponin, byproduct of damaged or dying heart cells. It's also entirely possible that his electrical activity just...won't come back...when they start rewarming him. They won't know until they get there. 

His kidneys are...actually sort of working? The basic concept is just a filter; as long as some circulation is happening, some urine will come out. Absolutely no regulation is happening; a lot of that is neurologically-controlled, and even if his kidneys were getting the appropriate signals, the cells are going to be very sluggish at responding. They are very carefully monitoring exact quantity and electrolyte concentration, and trying to replace whatever is coming out. Meanwhile, there's already detectable protein in the urine, indicating cell damage. There is an ongoing debate and updating prediction market on whether to administer the usual kidney-protective drugs, or whether the unpredictable response at this temperature - and delayed breakdown, for anything not covered by one of the liver functions Merrin is providing manually - means an unknown benefit and unknown risks and they should wait. 

He's definitely going to lose the entirety of his gut lining after this. Those are fast-dividing cells, and fast-dividing cells don't respond well to multiple hours of hypoxia. Fortunately, they're not cells that particularly matter. It's a risk taken as a matter of course with chemotherapy patients, and it's not much fun for anyone involved, but it does grow back. He'll need a few weeks of IV-only nutrition, which is a much smaller risk than it would have been a few decades ago; the latest generation of IV nutrition is very thoroughly optimized to provide everything the body needs, customized to a particular patient and responsive to changes in their condition, and it's only slightly harder on the liver than regular tube feeds. Besides, if they have to, they can get him a new liver. 

Infection risk is a bigger deal, and they won't even notice it until they start rewarming. Data on patients cooled to 28 C for 24-48 hours shows a serious immunosuppressant effect; the extrapolated models currently being generated have a wide range of uncertainty, and there isn't enough time to do any new animal testing, but it's possible the patient has no immune activity right now. Fortunately, a bacterial infection will be slowed down as well, and plausibly a lot of the organisms adapted to be nastiest in the human body, at normal human body temperatures, won't actually have cell division happening at 20 C. 

Hospital-acquired infections just don't happen, as a general rule, with the standard precautions, but this case is different, and they can't fall back on 'minimize invasive procedures' because they...sort of need them. They're pumping him full of prophylactic antibiotics and antivirals, and a team somewhere is working out a cost-benefit analysis for kicking the UV sterilization up to double the usual setting. They're considering doing something to kick his gut bacteria, since bacterial translocation from a damaged intestinal lining is a risky potential source of infection, but adding any more procedures right now is a risk, and this call isn't urgent on the scale of minutes, so no decision has been made yet. 

His lungs are a mess, and probably the next biggest problem other than the low body temperature. They can at least put off worrying about a local inflammatory cascade leading to acute respiratory distress syndrome - the immune dysfunction works in their favor, here - but this also makes it pretty hard to get a feedback loop on whether what they're trying is having any effect, positive or negative. It's not great, though. They suctioned out as much of the lake water as they could, went in with an endoscope and rinsed everything with saline to clean out all the river-silt, but that wasn't the first priority on arrival - given the low O2 sat goal, lungs full of gunk weren't actually a threat on the level of minutes - so it waited on all the key monitoring equipment was in place, and wasn't done until forty-five minutes in. Long enough for the fresh water to have some fascinating impacts on the delicate mucusa lining of the bronchioles and the even more delicate alveolae. The bedside chest X-ray was horrifying, and only a little bit because Merrin snapped at them not to even THINK of repositioning him for it (X-rays are usually done with the torso as vertical as possible.)

It's still....actually fine for oxygenation? With 45% O2, only a minimal increase since what the EMTs set, they're easily meeting the goal set for the interim treatment plan. But CO2 is getting to be a problem. There are a few different things they can change: peak pressure to force air in, lowest pressure (to keep alveolae from collapsing), volume, or rate. The last one has the lowest risk of causing additional damage, and low-volume-high-rate is recommended, so they're now up to, like, fifty breaths a minute. 

Merrin is fine. This is fine. They're coming up on 150 minutes in (two and a half hours) and she's...getting the hang of this? It's objectively harder but it feels easier, as she's able to better chunk the pieces of it and get repeating subroutines automated by world-class programmers. 

(She has a library of her own, though most of the little automation scripts she wrote for situations involving five machine at once have never seen the light of day outside simulations, and also Merrin is objectively a terrible programmer. Her code is hideous and full of bizarre workarounds, though it's at least conscientiously well-documented. She....does eventually scrape together the dignity to ask if someone can look through a subset of them anyway and make sure they're safe to use for real and will behave the way she wants even in edge cases, or, uhhhhh, maybe it's....easier at that point....for someone who isn't stupid to just start from scratch and give her a non-stupid version?) 

To a random observer, it might be utterly mysterious what Merrin is even doing with her brain and how, but to a rank-two Keeper with explicit training in attention-splitting, who is watching her closely with the aim of replacing her as smoothly as possible once she's exhausted, it's probably not hard to figure out. Merrin talks to herself a lot anyway (or subvocalizes, at least) and given that she has an admin transcribing notes for her and taking requests, she's trying to actually be coherent in her narration. 

Merrin doesn't have attention-splitting training, and certainly doesn't have the raw thinkoomph to split her attention five ways, so she's not even trying. 

One key part is that she's not trying to hold very much state in her head - if anything, she's trying less than she would in a sim, because she trusts her backup, and trusts that everything is being documented in detail and a dozen experts are checking over and analyzing every decision she makes. (If she frames it right, this is reassuring instead of mortifying.) She focuses on tracking a gestalt sense of how things are going - better or worse than expected - and is ongoingly making mental notes when something isn't working the way she expected it to. It helps a lot that, in this case, she doesn't have to task-switch to make those actual notes, she can just subvocalize everything that goes through her head - and intermittently, when she has a minute or two of breathing room, check in with her admin to get fed back the list of accumulated notes, and hopefully some commentary from the observers. 

Everything else is purely reactive, stimulus-response, over and over and over again. For any individual machine, she's spent dozens of hours drilling in sims; she's intimately familiar with the alarm tones, the patterns they form, what those patterns mean - and the link from 'hear alarm' to 'adjust controls on console' is down on the level of motor memory. For 80% of the adjustments she makes, it never even reaches the level of conscious attention, any more than someone carrying a box up a hill would have to stop and use explicit planning to lift their foot high enough to clear an obstacle while re-balancing the box in their arms. For alarms or alarm-patterns that call for a slightly more complicated response - one that involves pre-emptively toggling another machine, for example - she has the OODA loop down to less than five seconds, and once she's done it, she drops it from her mental stack entirely. 

Besides alarms, she has her screen set to notify her with a (distinct) audio tone for non-urgent Treatment Planner updates for her protocols or targets, and a different tone for urgent updates. 

Even with the alarms set to peak sensitivity and narrow parameters - and it's a lot more than five of them, she's watching the direct sensor data for the machines she's managing herself, but also a lot of other sensor data, and changing machine settings for all the other supportive devices that are going to impact with hers - but even then, she's not constantly reacting. She often has fifteen-second pauses, and every so often she lands in a nice settled equilibrium and gets an entire five minutes before some underlying process changes and starts rippling, or - more likely - someone ELSE decided to tweak their settings. 

The downside is that it takes her much longer to notice trends, if they're not linear and obvious, especially if she's in the middle of a bad run of "overcorrecting for previous overcorrections". To actively flag her attention, it has to reach the level where it's standing out in her vague gestalt sense. She makes the most of her moments of breathing space, though, and a lot of that is instinctive stimulus-response as well, and her internal process might be described like so: IF no interruption for ten seconds, flip through set of last-thirty-minutes-history screens for all her machines, IF obvious problem then troubleshoot ELSE refresh updated diagnostic predictions, refresh updated treatment plan - 

She usually doesn't get all the way through her cached list of Things To Orient To, but when she does, she stops and takes a deep breath and Thinks Ahead. It's very obvious when she's doing that: she goes quiet, often closes her eyes to help clear her mental cache of all the cruft, and relies on the alarms to interrupt her if she's run out of time. Merrin may be slower at thinking, and certainly much less able to track complexity, but she's actually around average at task-switching. 

She doesn't get a lot of longer-than-two-minute blocks to Think Ahead, but she makes the best use of them that she can, and these are generally the times that she notices she could ask for outside support on something. Or, like, a sugary drink. 

Merrin also has some odd but charming habits, mostly developed during sim time when no one was watching her. She wants to disappear into the floor the first time she calls her vasodilator-constrictor a FLAMING NUCLEAR TURD, out loud rather than subvocalized, but regardless of how many Very Serious People are watching her from afar right now, it's honestly not worth trying to rewrite a bunch of random habits closely-adjacent to the ones she's leaning on so hard. She'll just...try to keep the actual noise pollution down.

The real problem is that, while she has a lot of sim hours with the vasodilator-constrictor machine controls, she's using it to hit a nonstandard target, and so it's flagging her attention a lot more often. ...Of course, an hour later, having sort of gotten the hang of some reactions and gotten the stupid repeating pattern automated, she's patting it and calling it a good girl. 

To expert eyes - especially to the eyes of a world-class expert in the particular organ system and interactions she's running, sometimes the person who designed the current-generation equipment - it's pretty obvious that Merrin's performance is imperfect, and not just because her reflexes is slower.

She actively makes mistakes. She overrides-and-ignores alarms if she's in the middle of something else, and sometimes forgets to come back to them until the two-minute silence override runs out or until someone actually interrupts to remind her. She occasionally tries the same thing three times in a row, accumulating frustration about how it's not working, before her conscious attention comes in and points out that it wasn't even what she meant to do.

Based on the responses she picks out from her repertoire of standard ones (because she's basically never thinking that through in real time), even the ones she spends ten seconds choosing, her mental model of the underlying territory is massively simplified. In any given case, she's not doing much reasoning more sophisticated than 'if X go up, Y go down'. Every so often, she loses focus for a second, or just tries to make an adjustment without looking down at the controls because her eyes are busy on the screen, and toggles something in the wrong direction. (She invariably catches herself ten or fifteen seconds later, and squeaks 'sorry!' to the room at large while she fixes it.)

But her performance kept gradually improving, rather than degrading, for the first ninety minutes, and then leveled off - mostly because her rate of noticing and actually having the attentional capacity to implement her minor iterative process improvements was now having to keep pace with a slightly higher rate of unrelated problems, like sensors failing or her vasodilator-constrictor controls deciding to freeze on her for two minutes. (It's not really specced to be used almost-continuously for two hours.) Her reaction times start to slow if she's been in react-only mode for ten minutes straight, and her rate of making obviously-incorrect adjustments climbs, but two minutes of breathing room is enough to reset it. 

And, of course, she is just slower - not to mention less able to step back and see larger patterns, or think ahead and improvise a better set of next steps than her memorized set provides - than the counterfactual Merrin with +2 SD thinkoomph and the same accumulated hours of sim practice. But counterfactual Merrin isn't exactly waiting to take over. The rank-two Keeper with even twelve hours of intensive sims could probably improve on Merrin's performance on real-time management while vastly outperforming her on planning ahead, but they - unlike Merrin - definitely can't do twelve hours of sims in a single day at all, let alone keep going after. As of right now, the benefit if any isn't clear, handover costs are clear, and 'planning ahead' isn't centrally Merrin's role anyway, given that this isn't a sim and they have huge quantities of human capital going into the planning process. 

The markets - and the hospital administrator - are still not seeing any reason to replace her until she asks to be relieved or her performance noticeably deteriorates. 

They're edging up toward three hours in. Merrin is obsessively refreshing the treatment plan dashboard in hopes of a new and better plan on what to do about the patient's disaster lungs, because the current state of affairs is giving her Creeping Unease, but she's also kind of desperately hoping they don't go for high-frequency ventilation. Not even for a good reason - it's a reasonable next step, balancing risk and invasiveness with benefit - but she HATES the noise it makes. 300 breaths a minute (delivered via an oscillating membrane in a control chamber) sounds like– not like anything else does, really, but it sets her teeth on edge, the vibration will throw off other sensors, and also it's just DISTRACTING. 

And maybe there will even be an updated timeline on their 100-custom-proteins so she can plan whether to ask for a bathroom break or push through it. She doesn't have to go very badly yet or anything, and asking is mildly awkward, and she's worried that whoever covers for her on the liver machine will mess up the nice stable equilibrium she's been in for the last half-hour. 

A note on the heroic and innovative treatment plan now being improvised in real time, with venture-capitalist funding and the support of most of the secrecy-cleared relevant domain experts on the plan: 

As with everything else, the problem has to do with cascades, and the exact sequence of multiplying and self-reinforcing problems that ensue when a brain is briefly deprived of oxygen and then circulation is restored. 

The brain is metabolically hungry, has minimal local energy stores, and has very limited capacity for anaerobic metabolism even as an interim measure. It takes about fifteen seconds of interrupted circulation for a person to lose consciousness. Almost immediately, a number of changes result. Synapses depolarize early, coordinated internal regulation of various chemical gradients fails. Cells, almost immediately depleted of their main energy source, start to lose ground on the ongoing active work of keeping sodium, potassium, and other ions in the right places against osmotic gradients. The shifting balance in electrolyte concentration between cells and the fluid around them results, very quickly, in water entering the cells, causing them to swell. 

Eventually, a cell runs out of all its last-resort options to compensate, and dies. But cells are sturdy little factories, and killing them takes much longer - minutes, not seconds - and even longer than that in a cold brain. It's plausible that most of the neurons in their patient's brain are, currently, damaged but not destroyed. 

The problem is that the signs of damage are in the blood and extracellular fluid all around them. Nothing happens immediately, because any biochemical response to a perceived injury is one that takes energy. If you don't start to provide some energy to the struggling cellular factories (and your patient is still at room temperature rather than stored at -200 C), then damage is still accumulating. But if you do restore circulation and oxygenation, it's a little like a family coming home, turning the lights on, and only then realizing that some miscreant has vandalized it in their absence. You can imagine a reasonable response to this, the one an actual dath ilani family would have - 'stay calm, everyone, let's get everything back to normal first and then plan our next steps' - but the human body does not have dath ilani rationality training. The response is an immediate, forceful, panic-driven overreaction.

Oxygen is highly reactive - one of the reasons it's such a useful player in energy-generating cellular respiration - and this has its downsides, especially when the situation is already unstable. Chasing and fighting the vandal who trashed your house is rarely good for you or the surrounding infrastructure, especially if every other house in the city has also been vandalized and your neighbors are running around shouting and adding to the general confusion. Intensely reactive oxygen "free radicals" - about as indiscriminately violent as the name sounds, if the usual firm control and disposal mechanisms aren't yet up to speed - are generated everywhere, and go around breaking more things, setting off a cascade of even more cellular incident reports. Glutamate is released into the extracellular fluid, where it makes the neurons irritable and excitable, exactly at the time when they don't need to be messily burning more oxygen and pissing out waste products into a street where the maintenance workers aren't yet back on the job. 

All local and higher-level coordination is failing, lost in a cacophony of random signals. The usual inflammatory response to injury ramps up and up, every incoming piece of information convincing the cells that they SHOULD be panicking. 

...And a mechanism that plays a key role in the very low rate of new cancers comes into play. Immune-cell police are already on high alert, tracking down damaged offenders and destroying them before they can wreck the whole neighborhood, but cells have their own internal code of honor. Damage, especially DNA damage from oxygen free radicals and all the other nasty things flying around, makes them a risk. It's possible that most of the neuron deaths are cells following a deeply-programmed protocol to shut themselves down. 

At which point, of course, every dead cell dumps its waste contents in the public street, and the situation is not about to start improving anytime soon. In the worst-case scenario, the tissue swells enough to block off the recently-restored circulation, and it's game over. But even survivable outcomes are not generally going to be good. 

They're in a better-than-usual situation for reperfusion, because of various factors but mostly the 'freezing river' component. With high quality medical treatment available, the cascade will probably stabilize before playing out to its gruesome final conclusion. 90% estimated odds of survival. 

The odds of a full recovery - of a cascade that stops before the damage exceeds the brain's ability to repair damage later and compensate for the odd dead neuron here or there - were initially estimated at 10%. The space in the middle is wide and varied, there are better and worse intermediate outcomes, but that doesn't actually matter. They know what the patient wants. A full recovery, neurologically intact, at whatever cost (whether denominated in money, favors called in, or awkward but temporary gastrointestinal adventures) - or, if every recourse fails, to go into the cold, and wait for a future time when, hopefully, Civilization will know how to fix it. 

(From here, they have a theoretically straightforward path to immediate cryosuspension if they run out of options, and hopefully an actual protocol will be finalized before and if it comes up.) 

10% obviously isn't good enough, but it's too late to interrupt the earliest stages of the cascade, because they happened before the patient even reached the hospital. 

So interrupt the later steps, then, right? Seems straightforward enough? It's exactly the same principle as basically everything else they've been doing so far. The issue is that here the tolerances are tiny, and it's not good enough to seize manual control of three or four levers, or even a dozen like the liver machine. 

The relevant knowledge exists, in bits and pieces, spread between dozens or hundreds of early-stage research programs. It would have been obvious before today that pulling it off on a human patient, with current research and available tech, will cost way more than most people's insurance would pay, and it might not even work. (The prediction market on the likelihood that this particular plan will work is still settling; they don't have a lot to go on, given that the plan mostly doesn't exist yet and is being written in real time.) 

Civilization would love to have solved this already, to have a protocol ready to go. But sometimes problems are just really hard, and their existing system of efficient market incentives to coordinate experts' effort hasn't, yet, solved this one except in fragments. Most patients won't choose the same stark risk-benefit analysis that this patient asked for, and certainly most patients don't have that much capital being thrown at their specific case.

They're about to find out if urgency and money are enough to pull together all those existing fragments, and make something work in time. 

Thus: a hundred custom proteins, and a team of some of the top world specialists, and a plan - that doesn't exist yet, but it will by the time the proteins do, hopefully - of what to infuse, in what order, to preemptively prepare all the walking-wounded cells to keep calm and carry on when the relief effort arrives, rather than, like everything always does in human biology, instead making everything worse. 

In the meantime, they're in a holding pattern, because they can't afford to restore full circulation and oxygenation but they definitely can't afford to let any more damage accumulate, and so they're going to navigate a careful and maybe-impossible balance. Just enough energy to keep up with the most basic cellular maintenance, stringing damaged-but-not-dead cells along a little longer, but not enough for the cell-home security alarms to go off, or for the cellular neighborhood police to start reporting for duty. 

The single most important sensor result - displayed on the main wall screen in bold, big enough to take in at a glance - is from a tiny, very expensive sensor. It's hanging out in the patient's dural venous sinuses, where venous blood from the brain, and only from the brain, collects before returning to the heart. Choosing to place it was a real risk, given the patient's immunocompromised status and the very bad things that happen when infections reach the brain, but it's essential for the plan - or, perhaps more specifically, for the decision on whether to keep following the current best plan at all, or abort and frantically try something different.

Like with damaged heart muscle, there are chemical byproducts specific to damaged neurons and neural support cells. A few dozen of them are known; the continuous-measurement sensor suite that could be flown over on short notice from the nearest specialist lab can measure nine of them. It's not quite as sensitive or accurate as the full laboratory suite in the hospital, but there's a limit to how often they can draw blood and send it, and they can make up for some of the deficit by placing it where the concentration should be highest, as-yet undiluted by mixing with the rest of the circulating blood. 

There was, to no one's surprise, a period of measurable but gradually dropping levels on three of the indicators. Exactly what the markets predicted from the known cellular and metabolic changes that happen during a period of interrupted circulation. Slower, in a cold brain, but they don't actually know how long the patient was in the water; the models run on 'assuming normal body temperature, time required for the relevant heat loss' are guessing 15 to 30 minutes.

Then, for a few minutes, there was a just-above-the-minimum-detection-threshold level of a few byproducts specific to reperfusion injury, but the consensus is that this doesn't, yet, indicate any kind of irreversible damage. The brain does still have considerable capacity to heal. 

There's also an ultrasound skullcap for regular checks of circulation and signs of swelling in the brain - lower-accuracy than either a CT or MRI, but it skips the radiation risk and the incredibly inconvenient need to remove all the metal on the patient and haul him off to the MRI suite for a half-hour adventure, and running it frequently helps account for any noise in the measurement. They may want to risk getting a baseline MRI at some point, but it's going to call for things being Actually Under Control for at least the last half-hour, and this hasn't happened yet. 

There's a saying in Civilization:  "There are three hard problems in system design:  Robustness against intelligent optimization, generalizing 100% correctly to a never-previously-tested problem on the first try, and maintaining the consistency of a Network-distributed data structure."  The joke/warning being that the third problem is allegedly that hard; the difficulty of the first two problems is considered obvious.

Hard problems are hard, often, because they have subtleties; gotchas that zap you when you wouldn't first-order expect them.

Of the first hard problem, a gotcha is that things blow up in the face of intelligent optimization even when those intelligences are not adversaries.  When you run a prediction market, the trading parties aren't adversaries, per se, but they want money out of your system and will take weird execution paths to get it, if any such execution path is available.  Even the programs the traders run, if they're executing simple means-end searches for maximum extractible value, can be considered as a kind of intelligence within the system against which the system must be robust.

Of the second hard problem, correctly generalizing out-of-distribution with 100% accuracy on a first try, a famous subtlety is that Reality has been known to be really ridiculously strict about what counts as a 'new and untested domain'.  Reality will kick you over the cliffside without hesitation even if you think you're in a domain very similar to a past domain, if you didn't test exactly that set of conditions.

And of course while it's sometimes hard to generalize correctly to a new domain at all, it's often a lot easier to do it right the second time, or if it's okay to get a few outputs wrong here and there.

Civilization has a lot of medical prediction markets.  It has a lot a lot of medical prediction markets.  Most medical markets don't experience this kind of attention, sure, but some do; it's not unprecedented for a rich patient to bid a million labor-hours on their outcomes.  Civilization has run thousands of medical prediction markets that saw higher bids than Merrin's patient, more trading, or more complicated and weirder medical problems.  If it was just that, the problem would be as 'inside of a previously tested domain' as any unprecedented complex event inside reality ever gets.

Few of Civilization's medical prediction markets are secret prediction markets, however.  There are routine secret prediction markets, some with a lot of liquidity, but those routinely secret markets are not medical markets for complicated medical conditions with lots of liquidity and traffic.

The case with Merrin's patient is, in fact, the very first time that a secret market operated by Exception Handling, in the medical domain, has dealt with a case that had this much liquidity, this much trading, these many medical complications... and, now, a particular malfunctioning sensor.

The programs underlying Exception Handling's secret medical prediction market have unit tests, function tests, they've been tested on vast amounts of simulated data, they've been run over live data from nonsecret medical cases.  Just because you can't test exactly the future domain you're concerned about, doesn't mean you can't deliberately get as close to that as possible, under special training conditions meant to try to materialize any obvious kinds of anomalies that might turn up later.  That's just common sense.

The problem is hard anyways because Reality can be such a huge asshole about what counts as having tested your code inside a domain that's close enough.

There's a number of medical experts participating in the prediction market on Merrin's patient.  They're not expert market traders, though, they're medical experts; almost nobody has enough time to become world-class at both of those domains simultaneously.  The way medical experts participate in markets is that they have a deal with a professional market-trading entity that takes in the medical professional's expression of their opinion: its felt confidence, their guess as to how much evidence supports it, how much they'd shift their opinion if another medical professional said the opposite, a half-dozen human outputs that closely match the native format in which a human medical expert's brain generates its felt sense of the problem.

It's then the trading entity's job to run an algorithm that takes in that advice, weighs it against the bidding history of other trading entities, integrates those human opinions against algorithm-calculated statistical advice derived from a huge database of patient histories, and finally places a buy or sell order.

The human experts on this problem behave essentially the same individually as they would in a larger open market; that component isn't being run out-of-distribution in any way that'll turn out to be significant.  The human experts are fewer by a factor of around 32, compared to total market liquidity, than the number who'd be participating in a corresponding open market on the same patient; but individually they're behaving almost exactly the same way as they would for any other patient. 

But there are fewer separate codebases for the final bidding entities, fewer distinct algorithms being used by the actual traders and market participants, than there'd be in a corresponding open market.  Only three distinct codebases, in fact, because it's expensive to develop the software and algorithms and database to be a full-fledged medical trading entity, except that it operates on a secret subNetwork only and doesn't consult public feeds.  Exception Handling did not have infinite subsidy available for producing expensive secret infrastructure to use on secret medical problems.  Even so you obviously wouldn't want just one codebase used by all the trading entities, that would be insane; three such codebases gives you something like a majority vote in case one of them fails.

All of that secret infrastructure gets extensively tested against the routine open markets, of course.  It loses money, because it hasn't been optimized for nothing but open medical trading; but so long as it only loses a little money, the system is proving itself almost-correct.

But Reality can be ridiculously strict: those three algorithms trading in open nonsecret markets, are being tested under noticeably different conditions than those three algorithms alone trading in secret markets.  In particular, the secret trading algorithms being tested in open markets are seeing many more external buy-sell orders than they'd see inside a secret market.  You can program the secret trading algorithms to ignore that info, for purposes of a test run, but then they really do get eaten alive by open traders who can see the larger market history.

The three trading codebases cleared for secret medical prediction markets (each with several distinct instances, owned and operated by different professional traders, separately integrating the advice from subsets of the human experts) were of course requested by Exception Handling to be written separately from each other.  There's no redundancy if the programmers just copy each other's code.

But even then, sometimes there's a pretty obvious solution, such that three programming shops working separately will all pick that solution.  K-nearest-neighbor, for example, or lasso linear regression, or considering 3081 ratios between 79 measurements.

One of the sensors included in the suite of continuous blood-test measurements, inserted a few hours ago into the patient’s vena cava, has malfunctioned and started showing high levels of a particular cell-damage indicator, one specific to the connective tissue involved in the lymphatic system.  The reading would make a lot more sense in a patient who wasn't being kept at 20C with minimal to no immune-system activity, and hence doesn't immediately trigger the first-pass garbage-data detection checks.  Worse, this individual sensor is part of a sensor complex producing enough distinct individual readings that humans are usually looking at summaries rather than every byte of data.  Only the machine algorithms are looking at every byte of data.

This sensor error doesn't set off the local anomaly-detector program, because that individual reading would be reasonable for a patient at standard body temperature.  The sensor error doesn't set off global anomaly detection, because those warnings have mostly been shut off and filtered; a patient at 20C is already chock full of unusual readings and ratios, already in a far strange corner of the state space.  So there's no simple way of noticing the anomaly, when the readings on the patient shift into a different far strange corner of the state space; a strange corner which, unfortunately, bears a nearest-neighbor resemblance to a totally different class of weird cases in the historical dataset, and sends a straightforward linear regression over quantities and ratios of measurements into bold new territory.  The complicated anomaly detectors have been shut off hours earlier, because they were just ringing all the time.  The entire system, in fact, is operating in something of a jury-rigged state of 'stop complaining about everything that's weird because the patient is at 20C, while still complaining about anything that looks weird given that'; this jury-rig is not perfect.

No human has eyes on this particular subelement of a wide stream of raw data.

An open market would be wide enough to have trading entities that specialize in noticing complicated algorithmic anomalies produced by sensor errors, but this market is too narrow for that.

And finally: in the underlying dataset, the resulting weird ratios between sensor outputs, produced by the sensor failure and run through a linear regression to yield other predictive factors, happen to be nearest-neighbor in the resulting metric to some cases of post-infectious autoimmune syndromes, in which the human-expert inputs often missed a treatment retrospectively deemed prudent.  This is a good occasion to 'correct' those disagreeing human-expert-opinion inputs! according to the algorithm anyways.

This sort of algorithmic error is a kind of hiccup that happens all the time, inside trading entities in the bigger open markets.  It's just that in those cases, there's a lot of diverse other trading activity you can look at and suspect that your own algorithm's weird prediction is mistaken.  Traders in real markets get more practice, and have a larger budget to spend on proprietary patches to their algorithms, to try to avoid losing money over things like this.  The result of a few players' local hiccups is just that their trading entities will lose some money before somebody there notices the anomaly, leaves the market temporarily, and returns with a patched system the next day.

In this smaller, secret market - the trading algorithm would have been less confident if other trading programs had seemed to disagree.  But since nobody counterbids its bid, and indeed some competitors soon start to bid similarly, the trading algorithm rapidly doubles and quadruples down.

Really, you could be impressed with how the thinner secret market got as far as it did, before breaking down.  The secret market processed tens of thousands of bids, thousands of prices, before outputting a gross error in one particular treatment-planning policy market.

It's just that, sometimes, 99.9% correct isn't good enough, and you also can't just leave the market and come back a day later.

Of which it's therefore sometimes said that there are three hard problems in system design:  Robustness in the face of optimization, consistency of distributed systems, and generalizing 100% correctly to a not-exactly-identically-tested new domain on the first try.

A lot of secret people with secret jobs are going to be secretly very upset, after the exact origin of this problem has been tracked down.  They won't just be annoyed that the secret market failed, but that it failed in such an obvious, straightforward way.

And annoyed worse, that when the system did fail, the alternatives immediately available to fall back on were so paltry.  Why didn't Exception Handling just request for the system to have a comprehensible-reasons-only operating mode that the trading software could fall back to, if the complicated statistical stuff started producing obviously bad outputs?  Shouldn't you obviously do that, if you're a Governance project only pretending to be a real market using a tiny handful of custom-written codebases, when broken software can't just leave and come back later?  Somebody just coded up a system in blind imitation of a real illegible medical market, when those systems should have at least had a legible failsafe fallback mode that could immediately kick in!  Obviously!  There should've been a legible system like that checking all the illegible 'market' results, considering how the secret 'market' was ultimately a fake one that lacked the pseudo-adversarial inexploitability of a real market!

They will feel that this failure lacked dignity.  One of the sharper comments will be that if there actually had been a Sparashki present, Civilization would have embarrassed itself in front of aliens by failing like this.

Dath ilan lacks some context.