The Twenty-Six Month Interval

The Sentinel Arc, Book Two — Part Three: Mars

by N.W.

1. Aonia, 2069

The station stood on the northern edge of Aonia Terra where the old highland breaks down into a plain, and it had been designed by people who assumed it would be photographed. It had not been photographed in four years. What it looked like now was what everything looks like after four years on that surface: the same colour as the ground, because the ground had settled on it and stayed.

Twelve people. One physician.

Nadia Okonkwo-Lind was thirty-nine and had been on the surface for nineteen months of a thirty-one month rotation. She was Medical Systems, which at Aonia was not one of two but one of one, and the difference between those two numbers was not arithmetic. At Shackleton, when she had been wrong, someone had been asleep down the corridor who could be woken and could be wrong with her. Here the corridor ended at a supply lock.

The distance to Earth that season was sixteen and a half light-minutes. A question and its answer were thirty-three minutes apart at best, and in practice a day, because the person on the other end had a shift and a queue and a family and no way to know that the thing she had asked was the thing she was standing in.

She had stopped calling them consultations. What she called them, in the log, was correspondence.

2. What the child wanted

There is a photograph somewhere in her mother's effects of a girl of nine in Gothenburg in a cardboard helmet, and the girl is not smiling, because she is concentrating.

The thing the child wanted was to arrive.

That is what the word astronaut means when you are nine: a person who has arrived somewhere. It does not occur to a nine-year-old that arriving takes eight months, that the eight months are the easy part, or that the whole discipline of keeping people alive at that distance is not about arrival at all. It is about intervals. It is about the length of time between the moment you decide something and the moment you find out.

She had arrived. She could look out of a window at a plain that no one had walked on and it moved her about as much as a corridor.

What she thought about instead, most days, was that on Earth a wrong decision announces itself in hours. At Shackleton it announced itself in six days and there had been a vehicle. Here the interval was twenty-six months, because that is how often the geometry lets anything leave, and inside twenty-six months a wrong decision does not announce itself at all. It simply becomes the situation.

3. Ferrán

Marisol Ferrán was forty-seven and had been on the surface longer than anyone, which at Aonia meant she had opinions about the air.

Her title on the roster was Agronomy and Closed-Loop Life Support, and the two halves of that title were the same job, a fact that the station's own documentation set out clearly and that almost nobody thought about. The plants were the food. The plants were also a substantial part of the atmospheric processing — transpiration, carbon dioxide draw-down, particulate loading, the slow negotiation between what a person exhales and what a leaf will take. There were mechanical systems too. The mechanical systems were sized on the assumption that the growth module was doing its share.

Ferrán's work over the last three years had been the individualisation programme, and it was, by any standard anyone had, a success.

The premise was simple and had come out of a decade of ground work: if you can model a person's metabolism well enough to predict their response to a drug, you can model it well enough to decide what to grow for them. Not supplements. Food. On a station where every gram of mass had crossed the same interval as the people, the difference between growing what everybody eats and growing what each person needs is not a wellness product. It is a mass budget.

So the module ran a rotation that was not a farm rotation. It was twelve rotations braided together, and the braid was recomputed every ninety days against twelve fitted physiological models, and it worked. Iron status across the crew was better than the ground cohort. So was serum folate, so was vitamin D. The programme had a name and a citation and a follow-on funding line, and Ferrán, who did not care about any of that, had a wall in the growth module where she had written by hand the twelve names and what each one was short of.

Nadia had eaten in there more often than in the galley. It was the only room on the station where the air smelled like anything.

4. The rebalance

In March, Ferrán rebalanced the braid for Harjit Kaur.

Kaur was thirty-four, a geologist, and had come out of the transit with a folate picture that had not recovered the way the model said it should. It was not dramatic. It was the kind of finding that on Earth would produce a conversation about diet and a repeat test in three months. At Aonia it produced a rebalance, because a rebalance was cheap and available and precisely the thing the programme existed to do.

Ferrán increased the leafy fraction. The recomputation was automatic; the approval was hers; the log entry was complete and correctly timestamped and named Kaur and named the target and named the expected serum trajectory. Nadia read it the same week and thought it was good work, because it was.

The leafy fraction transpires more.

Over the following six weeks the module's humidity setpoint drifted up and was corrected, drifted up and was corrected, and then the environmental control logic did what it had been designed to do and re-tuned its own setpoint band to reduce cycling. That change was also logged. It was logged in the environmental system, in the environmental system's own format, as a control-band adjustment of four per cent with a note that the load profile had shifted.

Nobody wrote the sentence the medical models were fitted against the old band.

Nobody wrote it because it was nobody's sentence to write. Agronomy had documented a nutritional intervention. Environmental had documented a control adjustment. Both documents were complete. Neither document was about the other one, and the space between them was not owned by a department, a person or a review.

It is worth being exact about the failure here, because the temptation is to describe it as carelessness and it was the opposite. Every individual act in this chain was competent, documented, approved and correct. The station had no negligent step. What it had was a seam.

5. Drift

Nadia found it in Ferrán first, which she later understood was not a coincidence and at the time took to be one.

The station ran continuous physiological monitoring, and the monitoring fed the twin, and the twin produced, for each of the twelve, a rolling set of estimates that nobody looked at most weeks because most weeks they said nothing. Her job was substantially the job of not looking at things that were fine, so that she would have attention left for the thing that was not.

Ferrán's inflammatory picture had been climbing for eleven weeks.

Not steeply. The kind of slope you could draw a flat line through if you wanted to, and the model did not want to; the model flagged it at week nine and Nadia had marked it observed and let it run, because a slow inflammatory drift in a forty-seven-year-old at 0.38 gravity after fifty-one months of cumulative surface time is a finding with about nine plausible explanations and eight of them are boring.

At week eleven the twin stopped offering a slope and offered a probability.

Thirty-four per cent, within two hundred days, of a specific reactivation syndrome that had a name and a literature and a treatment, and which, untreated at this distance, was the sort of thing that does not kill a person quickly and does not let them work.

The estimate came with its uncertainty attached. It always did now. That was the reform, and the reform had her name in the acknowledgements of the guidance document, and she had spent six years of her life on it.

The uncertainty said: environmental envelope sensitivity, ±9.6 points.

6. What ±9.6 means and what it does not

An envelope sensitivity is a statement about how much the estimate would move if the conditions the person is living in were at the edges of the range the model was fitted against, instead of in the middle of it.

It is an honest number. It is the number Nadia had fought for. It says: this model was built on people living in a certain band of temperature, pressure, humidity, particulate load, radiation exposure and gravity, and you are asking it about a person who is somewhere in that band, and depending on where in the band she is the answer is between twenty-four and forty-four.

Twenty-four and forty-four are different decisions.

But that was survivable. She had been trained for exactly that, and the training was the thing Bakker had told her on a screen eight years before: the machine can be right and useless at the same time, and your job is the part where it is useless.

What she noticed at eleven o'clock on a Thursday, reading the envelope specification because she was thorough and because there was nothing else to do at eleven o'clock on a Thursday, was that the band the model had been fitted against was described in terms of the station's environmental control profile, and the profile was named, and the name had a version number, and the version number was the one from before March.

The sensitivity said ±9.6 points within the band.

It said nothing about being outside it.

7. The one thing worse than a number without doubt

She sat with that for a while.

The reform she had won said: a number must carry its uncertainty, so that the person acting on it knows what they are holding. She had believed that completely, and at Shackleton she had learned the first hard thing about it, which is that knowing what you are holding does not tell you what to do with it.

This was the second hard thing, and it was worse.

The stated uncertainty is itself a claim.

±9.6 was not an observation. It was an output of a procedure, and the procedure took as input an assumption about the conditions, and the assumption had been true in February and was not true now. So the estimate was thirty-four with an error bar of nine point six, and both of those numbers were computed correctly, and the second one was describing the wrong world.

She could not say the estimate was too high. She could not say it was too low. The direction was unknown. Increased humidity and altered particulate load could plausibly push the biological process either way and there was no literature on the question because there had never been a population to ask it of.

What she had was not an uncertain number. It was a number whose uncertainty was uncharacterised, which is a different object, and which the display had no way of showing her because the display had been designed on the assumption that the envelope was known.

There was no red flag for this. There could not have been. A system cannot flag a condition its designers did not know was possible, and the reason they did not know was not stupidity — it was that the environmental profile and the physiological model had been procured, validated and accepted by two different programmes, each of which had done its work properly.

She wrote in the log, that night: the model is not wrong. The model has been moved.

8. The credibility file

In 2049, when she was nineteen and had nothing else, she had read a credibility file that eighty-one people had signed off against without opening. That was the whole origin of everything she had done since. It had made her, in the way that a single grievance can make a person, and she knew that about herself and had made a kind of peace with it.

So she opened the Aonia files. All of them.

It took four days. The physiological model's validation dossier ran to nine hundred pages and she read it, including the appendices, including the parts that were plainly boilerplate lifted from the lunar programme, including the eleven pages on partial-gravity pharmacokinetics that admitted in careful language that the measurements did not exist.

The dossier was good. That is the part that she found hardest to explain afterwards. It was a genuinely good document, written by people who had thought carefully, who had stated their assumptions, who had listed their limitations in a section headed Limitations and had not hidden anything in it.

Section 4.2 stated the environmental envelope. It named the profile version. It said, in a sentence that was not buried and was not evasive: estimates are conditional on the station operating within the referenced environmental control profile; material deviation requires revalidation.

The document had told her exactly what she needed to know.

It had told her in a document that the person who changed the profile had no reason to read, in a sentence that assumed someone somewhere was comparing two version numbers, and there was no someone. There was no process that took the environmental system's control-band adjustment and asked whether any physiological model referenced the profile it had just adjusted. Such a process would have been three lines of logic and one shared identifier.

Nobody had failed to read the small print. She had read all of it. Both documents were complete and both were correct and the thing that mattered lived between them, where completeness is not defined.

9. Correspondence

She sent it to Earth on the Tuesday.

Sixteen and a half minutes out. The consulting board at Toulouse picked it up, and because it was flagged it went to a physician rather than a queue, and the physician read it properly, and the reply came back nineteen hours later and was two thousand words long and useful and did not help.

They agreed with her analysis. They agreed the envelope was invalidated. They said that revalidation required an observation set from the current profile, that the minimum viable set was somewhere between sixty and ninety days of stable data across a population, that the population available to her was twelve, and that with twelve the confidence interval on the revalidation would be wide enough to make the exercise close to decorative.

They said they would begin work on a ground analogue. A ground analogue takes months to stand up and its results would arrive at Aonia across an interval.

And then, near the end, in the paragraph that she read four times, they said that in their view the decision could not be deferred to them, that they were sixteen and a half minutes and one shift-handover away from every fact she had, and that they were recording their concurrence with her analysis and explicitly not recording a recommendation, because a recommendation from Toulouse would function as authority without carrying knowledge.

She recognised that. She had seen the other version of it eight years earlier, in a corridor, where a man had written concur and meant it and had been careful to say what it rested on.

This was the honest form. It was also, she noticed, a signature on a document that made the decision hers.

10. Forty-one days

The transfer window opened in forty-one days and stayed open for about three weeks and then did not exist again for twenty-six months.

The return vehicle was there, fuelled and checked. It had seats. The question that had to be answered inside forty-one days was whether Marisol Ferrán should be in one of them.

If she went: eight months of transit, arrival at a place with hospitals, and a definitive answer to a question that at present had a thirty-four per cent estimate attached to it, or a twenty-four, or a forty-four, or something else. If the syndrome was developing, she would be under proper care before it mattered. The station would lose its agronomist and closed-loop specialist for the remainder of the rotation, and the braid would be run by a system that could run it and by nobody who could argue with it.

If she stayed: twenty-six months. Treatment was possible on station — the drug was in the formulary, because somebody sensible had put it there — but the treatment was not benign, the dosing in partial gravity was one of the eleven pages that admitted the measurements did not exist, and starting it on a thirty-four per cent estimate meant treating two people wrongly for every one treated rightly, in a place with no laboratory to tell you which you had done.

Both branches were outside the model's competence. That was not a failure of the model. Nobody had ever built a model for this, because the population it would need is eleven people, spread across twenty years, at three stations, and that is not a population. It is a list.

11. The vehicle nobody had asked about

She spent the Sunday on the transit vehicle's environmental specification because she wanted to know what she would be sending Ferrán into.

The vehicle had its own physiological modelling package. Of course it did. It was a different procurement, run by a different agency, validated against a different environmental profile — the transit profile, which is not like a surface station: lower humidity, higher carbon dioxide setpoint, different particulate spectrum, no growth module at all, and a radiation exposure that is not comparable in any direction.

The vehicle's package was competently built and its dossier had its own Section 4.2.

She read it twice and then sat back and understood that she was looking at the same defect one level up.

Nowhere in either document — station or vehicle — was there a statement about what happens to a fitted individual model when the individual moves from one envelope to the other. Not a warning, not a caveat, not a note saying the question was open. The station document was conditional on the station profile. The vehicle document was conditional on the vehicle profile. A person going home crosses from one to the other in an afternoon, carrying a model that had been fitted in the first and would be read in the second, and both documents were silent because the transition was in neither of their scopes.

Each programme had drawn a boundary and stayed honestly inside it. The gap between the boundaries was where the actual human being went.

She wrote that down twice, once in the log and once on the back of a printed rota, because she had learned at nineteen that she did not trust herself to remember the shape of a thing she had only thought.

12. What she asked Ferrán

She asked her on the Wednesday in the growth module, standing up, which was deliberate, because sitting down in an office makes a conversation into a consultation and she wanted this one to stay a conversation for as long as it could.

She told her the number. She told her the envelope was invalidated and what that meant. She told her that she did not know the direction of the error and that Toulouse did not either. She did not soften it, and she was aware while she was not softening it that she had been trained by a man in a corridor who had refused to soften something to her.

Ferrán listened and then said the first thing that any competent person in her position would say, which was: is it the rebalance.

Nadia said that the rebalance had changed the load, that the load had moved the control band, and that the control band was what the model was fitted against. She said that this was a chain of correct decisions and that no part of it was a mistake.

Ferrán said: I made the air different for everyone to fix one person's folate.

Nadia said: yes.

There was a pause of about four seconds which she would think about for a long time afterwards, because she had expected guilt and what she got was an agronomist.

Ferrán said: then it is in the others too.

That is the second question, arriving from the wrong direction and from the wrong person. Nadia had been treating a case. Ferrán, who could not read a physiological model and did not need to, had gone straight to the property of the system: if the envelope moved, it moved for twelve people, and eleven of them had estimates that were being computed at that moment against a profile that no longer existed, and all eleven of those estimates looked exactly as trustworthy on the display as they had in February.

Nadia had known this. She had known it in the sense that it followed from what she had written on the Thursday, and she had not looked, because she had a case in front of her and a window in forty-one days and attention is not infinite.

She went and looked. Two others had flags that she had marked observed. Neither was Ferrán's picture. Both were now uninterpretable in the same specific way.

It did not change the decision she had to make. It changed the size of the thing she was standing in, from a case to a station.

13. The decision

She kept her.

The reasoning, as she wrote it, in the order she wrote it:

The transit envelope was less characterised than the station envelope, not more. She could state that the station model had been moved and by roughly what. She could state nothing at all about what a fitted individual model does across a profile transition, because no document addressed it. Sending Ferrán home would have replaced a known invalidation with an unknown one, and would have done it in a vehicle where the physician on board was a package.

Eight months of transit is not a corridor to a hospital. It is eight months of the same problem somewhere with fewer people.

She could observe here. Not revalidate — she had twelve people and Toulouse was right that the exercise would be decorative — but observe, daily, in a room, with her own eyes, the specific person, against her own recorded baseline from before March. That is a worse instrument than a validated model and a better one than a model that has been moved, and knowing which of those two sentences applies is the whole of the job.

She did not start the drug. She wrote the criteria that would make her start it, in advance, with numbers in them, and she wrote them before she was frightened, because criteria written while frightened are not criteria.

She sent the reasoning to Toulouse, and she sent it to the environmental group at Noordwijk, and she sent it to Ferrán, which is not standard and which she did anyway.

And she asked Ferrán to hold the braid where it was. Not to revert it — reverting would move the envelope a third time and Kaur's folate was a real finding and the intervention had been right. To hold it. Sixty days of a stable profile is worth more than a return to a profile that was only ever the reference by accident of being first.

That is the whole decision. It reads as small on the page. It took her four days and she did not sleep properly for two of them and she has never been certain about it since.

14. What she cannot know

The window opened and stayed open for nineteen days and the vehicle went with eleven seats empty and one crew member aboard whose rotation had ended in the ordinary way.

That is where a story wants an outcome, and the honest position is that the outcome and the decision are not connected in the way a reader needs them to be.

Ferrán's inflammatory picture flattened over the following four months and then declined. The syndrome did not develop, or had never been developing, and there is no test that distinguishes those two sentences retrospectively. Nadia did not start the drug. If the estimate had been correct and she had been lucky, she was wrong and got away with it. If the estimate had been inflated by an envelope she could not characterise, she was right for reasons she could not demonstrate at the time.

She will not know. Not ever, and not because the record is poor. Because n equals one, on a question that has no control arm, in a place where the same conditions will not recur.

This is the part that does not appear in the guidance documents, and she thinks now that it should. The literature on decision-making under uncertainty is enormous and almost all of it assumes that the decision is eventually scored. Twenty-six months is long enough that the thing you decided has become the situation you are in, and there is no version of the world available in which you decided otherwise, to compare it to.

What she did afterwards was what people do. She looked for a way to feel that the decision had a property other than its outcome. She found one, and she distrusts it, and it is this: the decision was made against a stated understanding of what was and was not known, and that understanding was written down before the answer arrived, and it can be read by someone else and disagreed with.

That is not the same as being right. It is the only thing available that survives the interval.

15. The memo

She wrote it in the fourteenth month and it was eight pages.

It did not describe Ferrán. It described the seam: that a physiological model is conditional on an environmental profile, that the profile is under the control of a different system with different owners and a different log, that an individualised nutritional programme is by construction a mechanism for altering the shared environment on behalf of one person, and that at present nothing anywhere in the architecture connects a change in one to a revalidation flag in the other.

It proposed the fix, which was not interesting and was three lines of logic and one shared identifier: the environmental profile gets a version, the physiological dossier declares which version it was fitted against, and any adjustment to the first raises a flag against every model that names it. A schoolchild could specify it. It had not been specified because it belonged to neither programme.

It proposed the second fix, at mission level, which was harder: that a fitted individual model must declare its behaviour across an envelope transition, or state in terms that the behaviour is unknown, so that a person deciding whether to put someone on a return vehicle is told that the question is open rather than left to discover the silence herself on a Sunday.

She recommended that the phrase within envelope appear on the display beside every estimate, and that the display be capable of saying envelope invalidated — magnitude and direction of error unknown, which is an ugly string that will alarm people and is the truth.

Then she came to the part where a memo names its recipient.

The next crew was not selected. Selection would happen after the window after this one; the people who would read this on the surface of Mars were, at the moment she was writing, doing other jobs and had not applied. The people at Toulouse and Noordwijk who could implement it would read it in sixteen and a half minutes and would take between one and four years, if it happened at all, because a fix that belongs to neither programme is a fix that has no budget line.

She addressed it to the record.

She was aware of the shape of what she was doing. She had spent her professional life on the consequence of a document that eighty-one people had not opened, and here she was writing one, at distance, for an audience that did not exist yet, knowing exactly the odds.

She wrote it anyway, and not out of hope. A record is not a message. A message needs a reader inside a human interval. A record is a thing that is true and stays put, and the only property it needs is that when someone finally does have the question, the answer is already there and is dated earlier than they are.

16. The interval

Twenty-six months later the geometry came round again and a vehicle left and another arrived, and the woman who came off it with Medical Systems on her roster was thirty-three and had read the memo, because it had been put in the handover pack by an administrator at Toulouse who had never met either of them and who had decided, for reasons that are not recorded anywhere, that it belonged there.

The three lines of logic had not been implemented. They are, as of the writing, still not implemented, and the reason given each time is that the interface between the environmental programme and the physiological programme is under review.

The display still says thirty-four per cent, plus or minus a number that describes a world that may or may not be the one outside the wall.

Nadia went home on the vehicle after that one, and the transit took eight months, and she spent a good deal of it thinking about the fact that she had crossed the envelope transition she had written eight pages about, carrying her own fitted model, which was being computed the whole way against a profile it had never been validated in, and reading out numbers that looked exactly like numbers.

She is not going back. The child in the cardboard helmet would find that difficult to accept and would be wrong about why.

What she would say to that child, if the interval ran the other way, is not read the small print, because she read all of it and it was correct and it did not help. It is not ask a second question, which she learned on the Moon and which is not a method. And it is not write it down, which sounds like wisdom and is only the residue of having no better option.

It is something less usable than any of those, and she has stopped trying to phrase it well:

Every system that keeps a person alive at that distance is assembled out of parts that are each correct, documented and owned. The harm does not come from the parts. It comes from the places between them, which are complete in nobody's document, and which have no owner because ownership is defined per part.

Someone has to stand in the gap and know that they are standing in it.

There is no procedure for that, and the twenty-six month interval means you will not find out whether you did it well. You will only find out what happened, which is a different thing, and which people will mistake for the same thing for as long as there are people.

Fiction. The Sentinel Arc, Book Two, concludes here.

The technical substance — environmental envelope conditionality in fitted physiological models, the absence of measured pharmacokinetics in partial gravity, the difference between an estimate that carries an uncertainty and an estimate whose uncertainty has itself been invalidated, and the governance gap between separately validated subsystems — is drawn from the published literature set out in The Waiting Problem (WINNIIO, 2026), where the sources are given. The individualised agriculture programme described here is a plausible extrapolation of current closed-loop life support research and is not a description of any existing system. Mars in this story is a place where two correct documents leave a person standing in the space between them, which is the only way it resembles anything we have actually built.