The Handover
The last email Ilse Brandt ever sent me arrived at 4:17 in the morning, Chile time, on a Tuesday. Subject line: f_blend. Body: one sentence. Do not touch until I say.
She died eleven days later, on a Thursday, of a stroke in her kitchen in Pasadena with a burnt saucepan on the stove and a pencil behind her ear. I found out from the institute administrator, who called me at 6:04 a.m. because I was still listed in her contacts as primary, pipeline. Fifteen years I had been primary, pipeline. I was also the person who ordered her tea, watered the plant that was slowly dying in her office, and knew that the stove timer in her kitchen ran three minutes fast.
My name is Elena Soto. I run the Cepheid photometry pipeline at the Institute for Astronomical Calibration, which is a polite name for a building in La Serena with very good seeing and a very old data archive. I am forty-six years old. For fifteen years my job has been to turn pixels into magnitudes, magnitudes into distances, and distances into the one number the rest of astronomy argues about: the Hubble constant, the current expansion rate of the universe. Lately it has been 73 or 68, depending on who you ask, and the gap between those two numbers is the largest unsolved problem in my field. It is also, when you get down to the actual labor of it, a gap of about a tenth of a magnitude in the absolute brightness of a kind of pulsing star called a Cepheid variable.
A tenth of a magnitude. You can lose that in a bad night’s transparency.
The appointment committee to replace Ilse would meet in ten days. I had wanted that chair since I was thirty-one years old and she told me I was “adequate with a photometric aperture, finally.” I want to say I wanted it for the science. I wanted it partly for the science. I also wanted the parking space and the line on the grant forms and the thing where people stop calling you the technician.
So: the week.
I did not open her last directory for three days. There was a funeral to arrange. Ilse had a sister in Tucson who collects teacups and a lawyer in Pasadena who collects nothing and was therefore the executor. There was a will that left her books to the observatory library and her clothes to nobody, since there was nobody, and her unpublished work to “the institute, in trust, per Soto’s judgment.”
Per Soto’s judgment. She wrote that in March, eight months before the stroke. It did not make me feel chosen. It made me feel like a drawer she had labeled.
I flew up for the funeral on the Friday. Ilse’s apartment smelled like Lapsang souchong and old hardbacks. Her sister, Dolores, had already packed the kitchen into boxes that said KITCHEN in black marker, twice, in case the label lost its nerve. On the dining table, where Ilse used to rederive the distance modulus by hand when the pipeline annoyed her, there was a legal pad with the funeral program notes in Dolores’s round hand, and under that, in Ilse’s pencil, what turned out to be the seating order. Row one: the sister. Row two: the director of the institute, and next to him, empty, a chair with a program already on it. Ilse had written E in the margin.
Between the funeral and the reception I sat on her bathroom floor for eleven minutes, which I know because of the timer on her phone, and did not cry, which I also know because of the phone. What I did was read her last paper notebook, the black one with the elastic band, because Dolores asked me to go through the office and “see if anything in there is the kind of thing people fight over.”
The last eleven pages were re-fit notes. Pencil, small, the letters leaning right, impatient as always. Most of it was arithmetic. One line, on the second-to-last page, had a box around it:
The anchor was never the data. The anchor was the agreement.
Underneath, smaller: Tell Elena. Or don’t.
There were no pages after that. The elastic band left a dent in the paper where it had been holding the book closed for a long time, so I know she did not write in it again.
On Monday, back in La Serena, I opened the directory.
Her last work was a re-fit. The survey data she had used was from a deep field program called ANDES-Deep, which is not ours and does not release its images to the public for eleven weeks after observation. Ilse had had access as a referee of their calibration paper, which is a privilege you are not supposed to use for your own reduction, and she had used it for two years. The folder was dated 2026-03-11 and named, with her usual wit, final_final_v9. There were 412 subdirectories, one per Cepheid in the Large Magellanic Cloud, and a driver script in Python that I had written most of.
The pipeline works like this, and I am going to keep it short because Ilse would have hit me with a ruler. Cepheids pulse, and their pulsation period is locked to their true brightness: long period, bright star. That relation, discovered by a woman named Henrietta Swan Leavitt in 1912, is the rung on the cosmic distance ladder that everything else stands on. To use it you need to fix the zero-point: for a period of ten days, exactly how bright is a Cepheid, in absolute terms? Fix that and you can measure to any galaxy that has one. Use those distances to calibrate Type Ia supernovae, and you can measure the expansion of the universe all the way out.
The zero-point is set by Cepheids whose distance we know by some other means. In our case, the Large Magellanic Cloud, where there are eclipsing binary stars that give you a geometric distance, and where there are a lot of Cepheids, close together, in a crowded field.
Crowded. That is the whole story, and I will come back to it.
Ilse’s anchor paper, the one that half the field has built on for a decade, put the zero-point at M(10d) = −3.28. The rival camp, working with different anchors and different crowding corrections, got −3.15. That 0.13 difference is the entire Hubble tension. Hers gives H₀ = 73.1. Theirs, with their zero-point, gives H₀ = 68.7. Two measurements of the same universe, disagreeing by more than their combined error bars, which is another way of saying that somebody, somewhere, has made a mistake of about one tenth of a magnitude, and nobody can prove it is not themselves.
I ran her driver script Monday afternoon. It was designed to run overnight. The residuals, when it finished at 3:40 a.m., were wrong.
That is the wrong word. They did not fail. The pipeline has a dozen places it can scream: NaN in the photometry, a star list that does not match the image header, a fit that does not converge. None of that happened. The fit converged. The χ² was, on the whole, excellent. What was wrong was smaller and worse. The residuals, plotted against the pulsation period, had a shape. A long, shallow, coherent bend, about two hundredths of a magnitude deep, running from short periods to long ones. If you are the kind of person who has looked at ten thousand of these plots, you do not see a bend. You feel your stomach drop: a plane’s engines changing note.
A coherent residual against period is the signature of a period-luminosity relation that has the wrong slope, or of a photometric bias that grows with something correlated to period. Cepheids in the LMC: the long-period ones are the bright ones. The bright ones are the ones you can see in the crowded inner fields, blended with their neighbors. Blending makes a star look brighter than it is. Set your zero-point on stars that are secretly too bright and every distance you build on it comes out too long. The supernovae at the far end then look faint for where you have put them, and H₀ lands high.
Ilse knew this. Ilse taught me this. The question her residuals were asking, politely, at 3:40 in the morning, was whether her own anchor had a blending bias in it.
Her anchor paper had a parameter in it called f_blend: the fraction of flux in a Cepheid’s aperture that belongs to something that is not the Cepheid. It is a small, hateful, essential number. You estimate it by artificial-star tests: you drop fake stars into the real images and see how much the neighbors pile on. In Ilse’s published anchor, f_blend was fixed at 0.00, pinned with a prior of width 0.01, on the grounds that her photometry had been done on a deblended catalog and the correction was therefore already applied and reapplying it would be double-counting.
That reasoning is not crazy. It is the reasoning of a person who wants to be right, which is different.
I stared at the bend until the sky outside went the color of a refrigerator. Then I did the only thing I know how to do. I let f_blend go free, with a wide prior, and I ran it again.
It took eleven hours. I ate a sandwich at my desk and watched the Markov chain walk, pot-watching, except the pot is the Hubble constant. In the last third of the run, the chain found a place to sit and stayed there. f_blend = 0.086 ± 0.014. The bend in the residuals went flat. The zero-point came out at −3.15, give or take two hundredths, which is the rival camp’s number. Give or take two hundredths, which is also, when you propagate it, H₀ = 68.4 and a flat ΛCDM age of 13.81 billion years, up from the 13.29 in every slide deck this institute has presented since 2019. Half a billion years older. The universe gets half a billion years older, and the poster in the hall with Ilse’s face on it stays where it is until somebody moves it.
I told nobody. I want that on the record with myself. Tuesday I went in, ran the office, ordered new filter glass for the acquisition camera, ate lunch with the night assistants, went home, lay on my bed with the light off, and thought about the parking space.
Wednesday, the version history.
Git is a boring piece of software and I am not going to make it interesting. What matters is that Ilse’s repo has timestamps, and that she kept her history.
f_blend appears in her config files 148 times across two years. The pattern: it is set free, with a wide prior, in the first commit after each data release. Then, over the following weeks, in three or four careful edits, the prior narrows. Wide prior becomes a Gaussian of width 0.05, centered on a small number, usually around 0.08 or 0.09. Then that becomes a hard pin at 0.00, with a commit message that says deblended catalog, no double-count or per Colloquium 03/24 or once, only once, just the word agreed.
The pin dates line up. I checked them against the institute calendar, because that is the kind of thing I am the kind of person who does. March of each of the last two years: the external review. October: the funder’s progress report. Once, a September, which took me an hour to place, and then I found it: the year the rival camp published their competing anchor, and the week Ilse was invited to respond in the newsletter.
She would free the parameter, run the honest fit, watch the chain walk to 0.08, and then, before anybody else could see it, she would pin it back. Every time. She never once committed a message that said this is wrong. She committed agreed.
I sat in the empty office on Wednesday afternoon with the log scrolling past and understood something I should have caught at the funeral: Ilse Brandt had spent two years in a private argument with a piece of software, lost it three times on purpose, and never told her technician, the only person in the building who knew what a config file was for.
Thursday, Tomás.
Tomás Rivera is twenty-nine and finishing a thesis on Cepheid crowding in the inner bar of the LMC. He is very good and very anxious. He can already see the shape of his whole career, and it is narrower than he hoped. He came into my office with a printout and the expression of a man about to ask a doctor a question he already knows the answer to.
“Can I ask you something about the anchor?”
“No,” I said.
“Elena.”
“Ask.”
“If someone re-ran the anchor with f_blend free, what would happen to my thesis?”
I looked at him for a while. Behind him, through the window, the mountain was doing its late-afternoon thing where the light goes sideways and every telescope dome in the valley turns the color of a peach.
“Depends on who ran it,” I said.
“Would it change the numbers?”
“Tomás. You did your artificial-star tests. You got f_blend around 0.08 for the inner bar. Your own thesis undercuts the anchor by about the same amount my residuals do. You know this. That is why you are in here.”
He did not deny it. He sat down without being asked, which he only does when he is past fear and into the part where he wants to be told what to do.
“My committee is chaired by Aróstegui,” he said. “Who was on Ilse’s original paper. And on the review panel. And on the funder’s committee. If I defend a thesis that says the anchor is biased by nine percent of the aperture flux, I will need a job afterward, and the jobs are held by people who signed the anchor.”
“What do you want me to do?”
“Tell me whether it’s true.”
“It’s true.”
He put his hands flat on his knees and looked at them for a while. I thought of a config file with a commit message that just said agreed.
“Then defend it,” I said. “And cite Ilse’s own 2024 colloquium, where she said, and I am quoting from the slides, ‘the deblending question is not settled and anyone who tells you otherwise is selling something.’ I have the file. I’ll send you the file.”
He took the file. He also, before he left, asked me how I was sleeping, which is the kind of question night assistants ask and astronomers usually do not, and I said fine, and he said you don’t look fine, and I said I know.
After he left I went into the archive and found her colloquium slides. She had said that. Slide 14, in her dry little font. Underneath it, in the notes field, in the version of the file last modified at 2:11 in the morning: Tell them the truth or tell them nothing. Do not tell them a third thing.
That is when I started to believe she had left this for me on purpose, not just left it lying around like a coat. People leave coats. People do not write per Soto’s judgment in a will in March and mean nothing by it.
Friday, the note.
Her office had already been reassigned on paper, though not in fact, because nobody was going to sit in Ilse’s chair before the committee met and that is the whole problem in one sentence. Her notebooks were in a banker’s box on the floor. I had gone through the black one. There were four others. The fourth was a lab notebook, the graph-paper kind, the kind she had not used since the 1990s, and it was almost empty, and the last entry was dated four days before the stroke.
The entry was not long. She had drawn, by hand, a little diagram of an aperture and a neighbor star and the fraction of light that lands where it should not. Then this:
Re-ran anchor free with the 2026 deep frames. Chain sits at 0.087. Of course it does. I have been pinning this shut for two years because the building needs the number, and because I was the one who gave it to them, and a zero-point is also a reputation. Anyone will tell you that. Nobody writes it down.
The re-fit is in final_final_v9, which Elena will laugh at. If Elena is reading this: the honest run is the wide prior on f_blend. The other run, the one with my annotation as the prior, gives the old number back to four decimal places, which should tell you exactly how much of my life’s work was the prior. Do what you want. I have run out of ways to not decide.
I did the second thing before the first thing, because I am a technician and I do not trust a claim until I have reproduced it.
I set up the fit twice. In run A, f_blend free, wide prior: the chain walked to 0.086 ± 0.014, the zero-point landed at −3.15, H₀ = 68.4, the residuals flat as a ruled line. In run B, I took Ilse’s published prior, the one with the pin at 0.00 and the width of 0.01, and I ran the same 412 stars, the same images, the same software, her annotation in place of my honesty. The chain did not need to walk anywhere. It sat down on −3.28 and stayed, and the bend came back into the residuals, and the H₀ came out 73.1, and the age of the universe came out 13.29, and if you put the two runs side by side and did not know anything about astronomy you would say one of them looked like a measurement and one of them looked like a decision.
That is what she meant by how much of my life’s work was the prior. All of it, is the answer. All of the part that mattered, and none of the part that was real, and those are not the same sentence, and I sat with them both until the cleaning crew came through at ten and asked if I was staying late, which I was.
Saturday, the phone call.
Aróstegui called me at home, which he had no business doing, on a landline I had never given him, which meant he had gotten it from the institute directory that I maintain, which is its own small betrayal I am still deciding about.
“Elena. I’ll be direct. The committee meets Thursday. Your name is on the short list, and you know why you are on it.”
“Do I.”
“You’ve been running that pipeline longer than anyone. The field needs continuity right now. Ilse’s anchor is the standard. A technician who has spent fifteen years validating it, stepping into her chair, is a message of stability.”
“A message,” I said.
“To the funders. To SH0ES, to the review panel, to Aronson and Raghunathan’s people who keep writing letters. Stability. The tension is settled. We don’t need it reopened in an election year.”
“There is a re-fit,” I said. “Her last one. She ran it herself.”
A pause. It was a long enough pause that I could hear a television in his room, something with a crowd in it.
“Then it’s good that it’s in good hands,” he said. “Ilse’s last work, under Ilse’s supervision, so to speak. Filed with the estate. Nobody’s asking you to do anything but the sensible thing.”
“Sensible,” I said.
“Elena. You’ve wanted that chair for fifteen years. Take the chair.”
He hung up. I stood in my kitchen in La Serena with the phone in my hand, listening to a man itemize the whole transaction out loud as if it were not insane: the chair for the drawer, my career for his stability, fifteen years of devotion cashed in at a rate he thought I would take. And the worst part, the part I still think about, is that at 8:40 on a Saturday night, standing in my own kitchen, I did the arithmetic. I did not stop myself from doing it. I stood there and I weighed a chair against half a billion years.
Then I thought about Tomás putting his hands on his knees and asking anyway.
Sunday night, the release.
I did not decide in a burst. I ran the thing and looked at what came out. That is how I decide.
I wrote the paper. It is not a long paper. It says: we re-fit the anchor with the blending term free on ANDES-Deep imaging; the chain converges at f_blend = 0.086 ± 0.014; the zero-point moves to −3.15 ± 0.02; the derived H₀ from our own calibrated sample moves from 73.1 to 68.4; the residuals flatten. It says, in the discussion, that the pin on f_blend was originally justified by the deblended catalog, and that the artificial-star tests, applied to the same catalog, do not support the pin, and that this was known in substance in the author’s own 2024 colloquium.
I listed the author as Brandt, I., first, and Soto, E., second. I put her ORCID in first, mine in second, and the software citation in third, and the data release DOI fourth. I wrote the abstract as she would have: without adjectives. In the acknowledgments I wrote that the re-fit was computed on the author’s equipment in the weeks before her death, and that the free-parameter run was her final reduction, and that I am releasing it at her written instruction.
That last sentence is true. I have the notebook, the page, the box drawn around the sentence in pencil, and I have not shown any of it to anybody.
I did not email Aróstegui or the review panel. I did not call Dolores to ask whether a legal pad counts as an instruction. Dolores would have said yes in a heartbeat, and then I would have had to admit that I had gone looking for permission, which I did not want. Ilse put my name in a will in March and told me not to touch anything, and eight months later wrote do what you want on the last page of a lab notebook. Those are not the same sentence. I have not been able to make them into one.
I submitted at 3:11 a.m. Monday, from the control room, because the pipeline was running and I was watching it, because that is what I do. The server returned a timestamp. I copied it into a text file and put the text file in the drawer of my desk, the one that sticks, the one she always said I should oil, and I did not oil it, because some things are load-bearing.
Then I went out onto the catwalk, empty and very cold, and stood there until the sky began to fail: black thinning to navy, then to a pale washed blue with the dome lights still burning in it. The dome next door was opening for the evening. Somebody in there was going to find something true and sign it, and that would be a good morning for them. Thursday there would be a meeting. My name would not be on the chair. My name would be on a preprint with hers, second, which is where the technician goes.
I went back inside. The nightly run had finished while I was out. I looked at the log, as I always do, because you do not trust a run you did not watch. All forty-one exposures. Photometry clean. Zero-point stable.
I left the drawer open and went to bed.