Ground Truth

The Sentinel Arc, Book Two — Part One: Earth

by N.W.

1. Solna, 2049

The room was booked for two hours and they were done in forty minutes.

There were nine of them. Two had flown in, three had come by tram, and four were on screens from three time zones, one of whom had a cat that walked across the camera twice. It was the fourth item on an agenda of eleven. The first three had been about procurement.

The question was where to put the line.

Ekwueme, who chaired, had been chairing for six years and had learned that the way to finish a meeting was to say the number first and let people argue down from it. "The working group proposes eight per cent. Over the mission envelope. Above that, no rotation assignment."

"Eight of what," said the man from Toulouse.

"Probability of a clinically significant recurrent event. Any system. Over two hundred and eighty days."

"That's very tight."

"It's meant to be tight."

The man from Toulouse looked at his hands. He was called Bergqvist and he had spent nine years in flight medicine before he stopped, and the reason he stopped was in a file that four people in the room had read and none had mentioned.

"I think it's too high," he said.

Somebody's chair creaked.

"You want it lower," Ekwueme said.

"No." Bergqvist took his glasses off, which was a thing he did when he was about to be difficult and knew it. "I want it to not be a number."

Halvorsen, at the far end, put down her pen.

"Explain," Ekwueme said, and to her credit she meant it.

"A model gives you an estimate and a confidence interval," Bergqvist said. "And it also gives you — if you ask it, and nobody asks it — a statement about which of its own assumptions the estimate is standing on. Sometimes the estimate is a measurement wearing a probability. Sometimes it's a guess with thirteen years of good behaviour stacked on top of it. Those are not the same object and eight per cent cannot tell them apart."

"So what would you have us write."

"A threshold that moves. Eight, when the estimate is well-supported. Twelve, fifteen, when it isn't, because when it isn't we don't actually know that we're excluding anyone."

"And who decides which it is."

Bergqvist opened his mouth and did not have an answer that would fit in the remaining ninety minutes of an eleven-item agenda.

"That's the problem," Ekwueme said, not unkindly.

"Yes."

"Because if the threshold moves, someone has to move it, and that someone is a person at four in the morning with a queue, and what they will do is pick the number that gets them home." She looked round the room. "A fixed threshold is not better medicine. It is better governance. It means nobody has to be brave at four in the morning."

The woman from Cologne, who had said nothing for eleven minutes, said: "How many people are we talking about? In the band you're describing."

Bergqvist did not know. That was the honest answer and he gave it. "A handful a year. Perhaps fewer."

"Out of how many?"

"Two hundred, two-fifty."

"So two per cent."

"Perhaps."

"And for those two per cent we would ask a duty officer to make a judgement they are not qualified to make, at a moment when they are tired, in order to be more accurate about people we cannot currently be accurate about at all." She was not hostile. She was doing the arithmetic out loud, which is what she had been asked to the meeting to do. "I would take the fixed line."

Somebody said: "Agreed."

Somebody else said: "Can we come back to this at the review?"

"We can," said Ekwueme. "It's scheduled for 2052."

And that was the moment. Not the vote. The moment was a reasonable question asked by a competent person, answered honestly by a man who did not have the number he needed, followed by a scheduled review that would arrive three years later with eleven items on its agenda and no new information about item four.

They voted. Eight to one. Bergqvist asked for his dissent to be minuted and it was, in a single line, on page fourteen of a document that would be reviewed twice in the following six years and would survive both reviews unchanged.

Then they broke for coffee and did item five, which was about procurement.

2. What a signature is

Halvorsen signed her first determination in 2050 and her eighty-first in 2061.

She was good at it. She was better at it than the people who trained her, and she was better at it than the two who came after, and the reason was that she read the whole file. Everyone said they read the whole file. She actually did — the intake notes, the parent's account, the pages of longitudinal data that the summary compressed into four lines. She could tell you the name of every one of the eighty-one and most of the reasons.

What she never did, in eleven years, was ask what the estimate was weakest at.

It was not laziness. The number arrived with a confidence interval and she read the interval. She understood what a credible interval was. She would have told you, correctly, that the interval was the model's statement of its own uncertainty, and she would have been describing a different quantity from the one that mattered, and nobody in the building would have corrected her because nobody in the building had noticed either.

The credibility file existed. It was generated automatically with every fit. It contained, among two dozen diagnostics nobody read, a row called sensitivity to initialisation prior, and the row had a number in it, and the number was almost always somewhere between nought point four and one point one percentage points, which is to say: nothing.

Except in one envelope. The one nobody could check.

The file was not withheld. It was not classified. It was one link away from the determination, in a schedule that a signatory could open by clicking on it, and in eleven years Halvorsen clicked on it never, because the number on the front of the document had a word attached to it and the word was cleared or not cleared, and the file behind the link had no word attached to it at all.

That is what a signature is. Not a decision. A decision that somebody has stopped looking behind.

3. Gothenburg, 2042

The child was four years and two months old and had been in the unit for three days when the consultant asked for a model.

His name was Lindqvist and he had been doing this for nineteen years and he asked because he did not know what was killing her. They had cultured her twice and got nothing back. The second nothing had arrived that morning at 04:10, printed, because the unit still printed. The antibiotic she was on was the third one and it was the one they used when they had run out of ideas about what they were treating, and Lindqvist had a private name for that antibiotic which he had never said out loud to a parent.

He stood at the end of the bed for a while. The machine breathed for her and made a sound like a man sighing in another room.

What he was weighing was not whether the model would help. It was that asking for one meant admitting, in a system that generated a record of everything, that he did not know. Nineteen years is long enough to learn that this is the part people avoid, and long enough to have avoided it himself, twice, in his thirties, in ways he could still describe.

He asked for the model at 05:20.

The mother signed the form in the corridor. She did not read it. She would have signed a confession. A woman she never saw again put a hand on her shoulder for about two seconds and then went to do something else, and that two seconds is the thing she remembers most clearly from six days, more clearly than the doctors, more clearly than the room.

She would think about the signature for thirteen years. Not with regret — with something she could not name, which was closer to the feeling of having been handed a stranger's child to hold.

The fit ran overnight. It took everything they had — the genome, four days of continuous telemetry, the transcriptome from the Tuesday sample, the two negative cultures which are themselves information — and it built a representation of a small body's inflammatory response, and then it did the thing that made it useful, which was to say what that body would do next under each of four treatment paths.

It also said, because it was asked, what the body would do under none.

Sixty-one per cent.

The consultant read it at 06:40 on day four and changed the antibiotic and the fluid strategy and, on the model's account of her renal trajectory, started support eleven hours earlier than the guideline would have. She was on it for nine days. She lived.

Nobody in that room did anything wrong. The model was well-built, honestly reported, and correct in every particular that could be checked within the week. The number sixty-one was not a mistake; it was an accurate statement of what could be known on the morning of day four with two negative cultures and a child who was still getting worse.

But it was the first thing the model ever knew about her, and everything the model learned afterwards, it learned by adjusting away from it.

Thirteen years. Forty-seven refits. Blood, sleep, growth, load, two ordinary chest infections and one broken wrist, every one of them observed, every one of them checked, every one of them moving the parameters a little further from the shape that a dying four-year-old had required. By 2055 the initialisation was invisible in every envelope anyone had ever measured her in.

There was one envelope nobody had measured her in.

4. What the letter said

It came on a Tuesday in June, to her and to her parents and to the Agency, simultaneously, which was the law.

RECURRENT DYSREGULATION UNDER SUSTAINED HYPOGRAVITY: 22.4% (95% CrI 14.1–33.8). THRESHOLD: 8%. DETERMINATION: NOT CLEARED.

Nadia Okonkwo-Lind read it in her room with the window open and the sound of the number four tram coming up from the street every seven minutes.

She was seventeen. She had wanted this since she was nine and had understood what an orbit was, which is that the same thing comes back, and had found that consoling in a way she could not explain to anyone including herself.

She did not know that the threshold had been set in forty minutes.

She did not know that a man called Bergqvist had said it should not be a number, and had been right, and had not been able to say why in the ninety minutes available.

She did not know that a woman called Halvorsen, who was by then teaching and who Nadia would meet in a corridor eleven days later and would decide she liked, had signed eighty-one of these and had never opened the file that would have told her the estimate was standing on a guess.

She did not know that on the fourth day of her life a machine had been told she would probably die, and had believed it, and had spent thirteen years being talked out of it by her heart and her lungs and her ordinary unremarkable Tuesdays, and had come almost all the way back.

Almost.

She knew none of that. What she knew was that a number said no.

She read the letter four times, and then — because she was the kind of person who read the small print, which is the only piece of luck in this entire account — she read the provenance block underneath it, which nobody had written for her and which nobody expected her to read.

Model class: mechanistic, host-response, individual-fitted. Fitted 2042-03-11. Refitted 47 times.

Note: the individual has no post-septic exposure to hypogravity. The estimate is derived from the fitted individual model under simulated conditions, not from observed response.

She sat with the last sentence for a long time.

She did not know any of that yet. What she did was read the small print, which is the only piece of luck in this entire account.

5. The corridor, thirteen years on

Her mother read it standing at the kitchen counter with her back to the room, and then went and sat in the car for twenty minutes without starting it.

What she was thinking about was the corridor. Not the six days, not the machine, not the consultant with the kind face who had explained the renal support twice because she had not taken it in the first time. The corridor, and a clipboard, and a form she had signed the way you sign for a parcel.

She had asked once, years later, at a follow-up appointment, what exactly it was she had agreed to. The registrar had been perfectly pleasant and had explained that it was consent to construct an individual physiological model for the purposes of treatment optimisation, and that it had almost certainly saved her daughter's life, and that yes, it was retained, and yes, it was kept current, and that this was standard and was a good thing.

Everything he said was true.

Nobody had ever told her — because nobody had needed to, because nobody had thought about it — that the thing she signed for in a corridor at half past five in the morning would still be answering questions about her daughter when her daughter was seventeen. That it would be asked whether the girl could leave the planet. That the answer would be shaped, a little, by the worst morning of her life.

She sat in the car and worked out that she would have signed it anyway. Of course she would. That was not the thing that was bothering her and it took her most of those twenty minutes to find what was, which was this: she had made the most consequential decision of her daughter's life in a state where she was not capable of making decisions, and the system had known that, and had asked her anyway, and had been right to.

There is no better arrangement available. She has looked.

When she came back in, Nadia was at the kitchen table with the letter in front of her.

"I want to see what it's least sure about," Nadia said.

Her mother sat down opposite her.

"Then that's what we'll do," she said.

6. The number, moving

Consider what the number did, in the six years between the meeting and the letter.

It went into a schedule. The schedule went into a system. The system produced determinations, and each determination had a word on the front and a file behind a link.

Two hundred and forty-one people were assessed against it. Two hundred and six were cleared. Thirty-five were not.

Of the thirty-five, eleven were not cleared by a wide margin and would have been excluded by any threshold anyone in that room would have proposed. Those eleven are not this story.

Nineteen were not cleared by a margin that a differently-drawn line would not have changed either. They are not this story either.

Five sat inside the band that Bergqvist had been trying to describe and could not — where the point estimate cleared the threshold in one direction and the credibility of the estimate would have carried it back the other way, if anyone had computed the arithmetic and if the arithmetic had been allowed to matter.

The first was a man called Idris who had wanted it since he was eleven and who, on being told, went home and did not come out of his room for two days and then came out and enrolled in anaesthetics. He was good at it. He is still good at it. He works in a paediatric theatre in Manchester and he is the person you would want if your child needed putting to sleep, and he thinks about the letter perhaps once a year, usually in the autumn, and the thought is not bitter, it is just a room he walks past. Whether the world is better for having him in that theatre instead of in orbit is a question with no method behind it, and anyone who answers it confidently is guessing.

The second appealed. She lost. She appealed again on new grounds, and lost, and the second loss took fourteen months and cost her most of a relationship, and at the end of it a clerk explained to her kindly that the body could not overturn a determination, only establish whether it had been properly made, and that hers had been. She stopped after that. She was right to stop. The system had told her the truth about itself and she had heard it.

The third was told by a flight surgeon in Houston, informally, in a corridor, off any record, that if she waited three years and put on some weight she would probably pass. She did wait three years. She did not put on weight. She passed. She flew twice, and did well, and never learned why, and to this day would tell you that the second assessment was fairer than the first because she was fitter for it — which is not what happened. What happened is that three more years of ordinary living had pushed a thirteen-year-old prior a little further from the shape a dying child had required. She thinks it was her. It was arithmetic. She is not wrong to be proud.

The fourth died in 2053, at thirty-one, of a road accident on a wet Tuesday outside Malmö, and the file closed, and nobody in the system ever noticed that it had been one of the five, because the system had no way to notice that and no reason to look.

And the fifth was seventeen and read the small print.

None of the nine people in the room in 2049 could have named a single one of them. That is not a failure of imagination. It is what a threshold is for — to be applied without the applier having to hold the applied-to in mind. Ekwueme was right about four in the morning. She was right, and being right did not make the band go away; it moved the band somewhere nobody was looking.

7. Ström, before

The man who would represent the model at the hearing spent the preceding week doing his job well.

Ström was fifty-two. He had built the disclosure schedule in 2047 and had been maintaining it since, and it was a good schedule — it was in fact one of the better artefacts in the building, and two other agencies had copied it. It listed, for every determination, what was to be released to the individual as of right.

The credibility file was not on it.

This was not an omission in the sense of something forgotten. It had been considered. There is a minute, from 2047, in which the question was raised and answered: the credibility file contains diagnostic material whose interpretation requires modelling expertise, and releasing it as of right would produce a class of appeals founded on misreadings of technical diagnostics by people without the training to read them.

That reasoning is correct. Ström had written it and he still believed it and it was still correct at the moment Nadia put page twenty-nine on the table.

What it did not account for is that a diagnostic which is dangerous in the hands of a hundred confused readers can be decisive in the hands of one who reads it properly, and that a schedule which optimises for the hundred has silently chosen against the one.

He would say later, to a colleague, that the thing which undid him was not the argument. It was that she had the page number. Anyone can say your model might be biased. She said page twenty-nine, row four, nine point eight, and the difference between those two sentences is the difference between a complaint and a finding, and he had spent five years building the schedule that made the second one almost impossible to say.

8. The corridor

"You're doing the wrong argument," Halvorsen said.

They were in a corridor in Solna with rain on the window and she was holding a paper cup and had bought a second one, which she now handed over without looking.

"It's a good argument."

"It's an excellent argument and it will lose. You're arguing about who decided. Ström will tell you the committee decided, and he'll be right, and you'll have spent your afternoon on it." She drank. "Ask instead what it decided from."

Nadia said nothing.

"You read the provenance block," Halvorsen said. It was not a question. "Almost nobody does. Say the third thing. You've had it since the kitchen."

"Those three thousand children were fitted from what happened to them," Nadia said slowly. "But the model of me was fitted on day four. When they still thought I was dying."

Halvorsen crushed the cup.

There was a silence of a particular kind, which is the silence of a person who has just been told something about eleven years of their own work.

"Eighty-one," Halvorsen said.

"What?"

"Nothing." She dropped the cup in the bin. "Go and say it."

9. Eleven days

Her father did not speak to her about it for two days. On the third he came into her room and sat on the end of the bed, which he had not done since she was small.

"Twenty-two per cent," he said.

"I know what it says."

"Do you know what it means."

"It means that if you ran a hundred of me—"

"It means that if it happens, you are eleven light-minutes from a hospital and there isn't one." He was not a man who raised his voice; she had heard him do it twice, both times at a car. "It means the same eleven hours, except eleven hours is the good version. The good version is the one where they know immediately and can do nothing."

"Yes."

"You don't."

"I'm the one who was there."

"You were four."

"I was there," she said.

He looked at his hands for a while.

"When you were in that room," he said, "I sat outside it for six days and watched a screen that told me what your kidneys were doing. And on the fourth day I understood that the screen knew you better than I did. That it could say what you were going to do before you did it, and I couldn't." He stopped. "I decided that was all right, because it kept you. I don't know how to be against it now."

"I'm not asking you to be against it."

"Then what."

"Make it prove it," she said. "It's never been wrong about me. Not once. But it's never been asked anything it couldn't check."

She spent most of those eleven days reading her own file, which is a stranger experience than anyone who has not done it can be told.

It went back to the fourth day. Everything did. There was a parameter set from 2042 and then forty-seven revisions of it, and each revision had a date and a reason, and the reasons were the ordinary furniture of a life: routine bloods, growth percentile update, febrile episode, resolved, fracture, left radius. She had been eleven when she broke her wrist. She remembered the cast and the smell of it and a boy called Anton who had drawn something obscene on it. The file recorded that the fracture had shifted her inflammatory response parameters by a decimal in the third place.

Both of those things were true. Only one of them was in the file.

She kept coming back to the first entry. Not the number — she could look at the number now without much happening. What she could not get past was that the model had been asked, on the fourth day, what she would do, and it had answered honestly with what could be known, and what could be known was that she was probably dying. And then everything it learned afterwards, it learned as a correction to that.

She tried to work out whether that meant it did not know her.

It knew her. That was the trouble. It knew her better than her mother did, in the specific and narrow way that mattered — it could say what her body would do on Thursday. But the shape of what it knew had been struck in a room she could not remember, from a set of facts that had turned out to be wrong about the only thing they were asked, and no amount of Thursdays had melted that shape all the way down.

You were built out of a mistake about me, she thought, and then you spent thirteen years being right.

She could not decide whether that made it more trustworthy or less. She still cannot. It is the sort of question that does not have an answer so much as a temperature, and hers changes with the weather.

10. What it was standing on

The credibility file was forty-one pages. Nadia read all of it twice, understood perhaps a third, and found on page twenty-nine a table with a row in it.

Sensitivity to initialisation prior, hypogravity envelope: Δ = 9.8 percentage points. Sensitivity, all other envelopes: Δ = 0.4 to 1.1.

It took her most of an evening to be sure of what she was reading, and when she was sure of it she found that she was not angry, which surprised her, and that what she felt instead was something closer to recognition.

The model was not wrong. That was the thing she had got wrong for eleven days. The model was extraordinary. It had been right about her every time it could be checked — thirteen years, forty-seven refits, blood and sleep and load and two ordinary infections, and in every one of those envelopes the day-four guess had washed out to within a percentage point of nothing.

Except in the one envelope where nothing could be checked. There, and only there, the thing a machine had been told about a dying four-year-old still moved the answer by nearly ten points.

Twenty-two point four, minus nine point eight, was twelve point six.

Twelve point six was still not eight.

She wrote both numbers on the inside cover of a notebook, underneath a word she had learned to spell when she was six, and looked at them for a long time.

11. The hearing, which is not the point

It took two days and there were eleven people in the room and it is the least interesting part of this account, so here is what happened in it.

Ström, who represented the model and wore a cardigan and had a printed document in front of him, said the number was computed and the threshold was chosen and both were correct. He was right about both.

Nadia said the number was computed from a counterfactual — a simulated exposure, for a body that had never been in any of those conditions — and that the model's own file said this was the one place its origin still showed, and that the number had gone into one document and the doubt into another, and only one of them had a word on the front.

Ström was quiet for about fifteen seconds, which in a room of eleven people is a long time.

Then he said: "She's right, and I'd like it minuted, because if I don't say it now I'll have to write it later." He took his glasses off, which he did not know he had in common with a man in Toulouse. "The determination coupled a point estimate to a threshold without carrying the estimate's own statement of where it was weakest. That's a reporting failure and it's mine. I sign the disclosure schedule."

They did not clear her.

They vacated the determination, which is a different thing — it meant it had not been properly made. They ordered a new one inside a hundred and twenty days, carrying the credibility figures on the face of the document. And they undertook to review whether a point estimate with a sensitivity of that magnitude could be compared to a fixed threshold at all.

Six years, four months and eleven days after a man in Toulouse had said I want it to not be a number, and had not been able to say why.

12. What it cost, and to whom

Bergqvist was seventy-one and had been retired for four years when he read the notice.

It was two paragraphs in a bulletin he still received because nobody had taken him off the list. Determination vacated. Credibility disclosure to be carried on the face of the record. Threshold under review. No names. There are never names.

He read it twice and put the tablet down and looked out of the window for a while at a street in Toulouse where a man was doing something to a scooter.

He had been right. He had been right in 2049 with eight people looking at him and a cat walking across a camera, and he had known he was right, and the thing he had not been able to do was say it in a way that could be written down. That was the whole of his contribution: a single line on page fourteen, correct and useless.

He thought about the three years after, when he had gone on signing, because the alternative to a threshold he thought was wrong was no threshold at all, and no threshold at all is worse. He thought about what he could have done instead of dissenting, which was to go and compute the sensitivity himself and bring it back with a number on it, and how he had not done that, and how the reason he had not done that was that item five was about procurement and then it was 2050 and then it was 2053.

Then he made coffee.

That is what it costs. Not a career, usually. Just the twenty years between being right and being able to prove it, and the knowledge that somebody spent those twenty years inside the gap.

He had told the story once, in 2051, at a conference in Ghent, in the eleven minutes allotted to him after lunch. He had put the argument up in three slides and it was a good argument and forty people had listened politely and two had asked questions afterwards, both of them about something else. He understood, walking back to the hotel, what had gone wrong: he had presented a concern. Nobody schedules a review because of a concern. They schedule a review because of a number, and he had never gone and got the number, because getting it would have meant asking the modelling group for six weeks of their time on a hypothetical, and he was not sure enough to spend somebody else's six weeks.

That is the honest account of it. Not cowardice. Proportion — applied wrongly, and reasonably, and for nine years.

The thing he would say now, if anyone asked, is that he had confused raising it with doing something about it, and that the difference between those two is roughly one competent afternoon with the diagnostic output, which he had every year for nine years and did not take.

Halvorsen was teaching by 2062, and told the story once to a room of forty-two students who did not understand why she stopped in the middle of it.

She said: eighty-one. She said: I read every file, every page, and I could tell you their names. She said: I never once opened the schedule behind the link, and I want you to understand that this is not a story about a lazy person, because if it is a story about a lazy person then none of you will check, since none of you are lazy.

Somebody asked what the file would have told her.

"Almost always nothing," she said. "That's the part that makes it hard. Seventy-nine times out of eighty-one it would have said nought point six, and I'd have closed it and felt clever for looking."

"So it wouldn't have mattered."

"It would have mattered twice." She let that sit. "I don't know which two. That's the whole problem with this job and I've been doing it for twelve years and I still don't have anything better to tell you than: open it anyway."

A student at the back asked what happened to the girl.

"She's twenty. She's not cleared. She'll be assessed again in eighteen months against a document that carries its own doubt on the front of it, and she'll probably still not be cleared, because twelve point six is not eight."

"Then what was the point?"

Halvorsen thought about it long enough that the room went quiet.

"The point," she said, "is that the next eighty-one will be told what the number is standing on. That's all. That's the whole of it. It does not get her to the moon."

She did not say the other thing, which was that she had gone back through her own eighty-one that spring, one file at a time, on her own hours, and had found the two.

She has not contacted either of them. She has thought about it for eleven months. One of them is a paediatric anaesthetist in Manchester who by every account is happy, and she cannot work out whether telling him would be an act of honesty or of self-relief, and she suspects that the fact she cannot tell the difference is itself the answer.

13. Fourteen days

Ekwueme said the thing at the very end, after the record had closed, while she was gathering her papers and not looking up.

"The envelope you were assessed against is two hundred and eighty days with no evacuation pathway. That's the hardest thing we currently fly."

"I know."

"It's not the only thing we fly." She squared the pages. "The surface stations are eleven light-seconds from here. There's a cycler. There's a return window every fourteen days. Nobody in this room assessed you against any of that."

"Why not?"

"Because nobody asked us to," Ekwueme said. "You asked to be an astronaut. That word does a great deal of damage in this building."

14. The window

That night she sat at the window with the blind up and the tablet dark on the sill beside her.

Somewhere above the cloud the transfer window had opened and closed without her. It would come again in nine days. It always came again — that was the entire point of an orbit, and she had known it since she was nine, and had never until tonight found it anything other than consoling.

She thought about the fourth day. She could not remember it. She had never been able to remember it. But she knew now what had happened in it: that in a room she had never seen, a machine had been handed almost nothing and asked what she would do, and had said she will probably die, and had been built around that answer.

And then she had gone and been alive for thirteen years, every ordinary day, in front of it. And every day it had moved a little. Forty-seven times. Never all the way.

She put her hand flat against the cold glass.

Fourteen days, she thought. Eleven light-seconds.

Below in the street the number four tram came up out of the dark, and stopped, and let nobody off, and went on.

Part Two: The Fourteen-Day Window — the Moon.

Fiction. No person, committee or proceeding here corresponds to a real one. The clinical and modelling substance — post-septic immune reconstitution, mechanistic host-response models fitted to an individual, sensitivity to an initialisation prior, and the difference between an observed response and a simulated one — is drawn from the published literature set out in The Waiting Problem (WINNIIO, 2026), where every source is given. The failure the story turns on is real and ordinary: a point estimate reported without the credibility assessment that qualifies it, by people who were each doing their job correctly.