The Sequence Was Not Withdrawn. It Was Replaced, By Its Own Second Author.
The record everyone calls deleted is still at its original address, served fourteen times in five years. Both versions are there. The second author changed one field: a collection date. The accession number records who deposited, not when.
HINXTON, United Kingdom — Version 1 of accession GQ-208441 has sat at the same web address since 17 February 2021. In the five years to June 2026 that address was served fourteen times.
The record was not withdrawn. Its status field reads replaced.
Version 2 was filed on 26 February 2021, nine days before the outbreak was announced, by the second person named on the submission, and it changes one field.
Three states, written in 1994
The archive's version model was specified in 1994 and has not been altered since.
A record can be live. It can be suppressed, which takes it out of search and reserves its accession number for ever, so that the number can never be reissued to anything else. Or it can be replaced, which mints a version 2 and leaves version 1 exactly where it was — retrievable, at its original address, with a banner across the top of it linking to the newer file.
There is no state called withdrawn. There has never been one.
We asked for the submission audit trail behind GQ-208441 and obtained it, with the depositing institute's written consent, under a curator research agreement: forty-one rows, timestamped to the second, submitter contact fields redacted.
Row 29 is the whole affair. On 26 February 2021 at 11:47 UTC the collection date was edited from 2020-11-03 to 2020-03-11.
A day and a month, transposed. Eight months of apparent difference produced by two digits changing places on a form that does not tell you which order it wants them in.
The number is a filing cabinet, not a clock
The second half of the claim is arithmetic, and arithmetic is the half you can check.
GQ-208441 carries a lower number than sequences deposited three weeks after it, which begin at GQ-441207. Read as a queue, that puts the disputed record earlier in the line than its own paperwork admits.
It is not a queue. Accession ranges are pre-assigned to institutions in blocks of 5,000, and the allocations are published quarterly in a table we can find no previous story citing. GQ-205001 to GQ-210000 went to the depositing institute in 2016. GQ-440001 to GQ-445000 went to a different laboratory, in another country, in 2020.
Each institution then works through its own block at its own speed. A number tells you which cabinet a record came out of. It tells you nothing about when the drawer was opened.
Dr. Emrys Nakato-Whyte, commissioned by this newspaper and not told the subject, tested it across 40,000 records. Within a single block, number predicts submission order almost perfectly, at 0.94. Across blocks it falls to 0.31.
Our own accessions, used as a control
This newspaper has deposited sequence data once, for the eleven-site watershed survey published as PX-2029. Before writing any of this we went back and looked at what our own numbers encode.
They encode our block, and nothing else. We submitted in two batches nine weeks apart, and the second batch carries numbers below the first, because a colleague had reserved a range in advance for a survey that was then abandoned.
Nothing about our accessions runs in the order we did anything. Ours are the numbers of a small institution using its allocation untidily, which is exactly what the disputed institute's numbers are.
The eleven days
The strangest line in the audit trail is not the version change.
The public submission date on GQ-208441 is 17 February 2021. The archive first received the file on 6 February at 04:12 UTC. For eleven days it sat complete and validated inside a system that showed it to nobody.
That is not a concealment mechanism, and it is not unusual. Any submitter can set a hold; records stay private until released, most often to line up with a journal's embargo. Across 2,300 submissions to this archive that year, the median hold was eleven days.
What we could not make interesting
Nine weeks of work, and the answer is a version-control artefact in a system older than most of the people arguing about it.
Dr. Ngozi Backstrom-Iwu's reply is at the foot of this page and it is better than our story. She is not a second author, she says: an archive submission has submitters, and she was listed second because the form sorts alphabetically. She thinks we borrowed the phrase because it made a better headline, and we did.
She also points out that eleven outlets wrote about her correction across five years and none of them wrote to her, and that the only thing separating this newspaper from those eleven is one email sent to an address printed on a public paper.
The desk has no answer to that. She is describing the minimum, and she is right that we should not be congratulated for meeting it.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We did not ask the archive whether a record had been deleted; we asked for the audit trail behind one accession number, which is a smaller and more answerable question, and obtained it with the depositing institute's written consent. We then recomputed the block allocation table for fifteen years, commissioned a bioinformatician who was not told the subject to test whether accession order predicts submission order at all, and finally went back through our own deposited sequences from an earlier story to see what our own numbers encode.
Who we spoke to
- Submission audit trail, accession GQ-208441, Forty-one rows, timestamped to the second, from receipt to current version. Obtained under a curator research agreement, with the depositing institute's written consent; released as a signed extract with submitter contact fields redacted June 2026 The institute consented before it knew what our conclusion would be, and asked for no sight of the draft.
- Dr. Sunniva Aalto-Marchetti, Data curator; nineteen years at the archive. Interviewed by video call three times and once in person; walked us through the status model on her own screen, with the ticket queue open April to July 2026 Has processed, by her own count, about 6,000 replacements. Says she has never processed a withdrawal because the system cannot do one.
- Accession range allocation table, 2011 to 2025, Which blocks of numbers were pre-assigned to which institutions, and when. Published quarterly by the archive at a public address; downloaded and recomputed. We can find no previous story citing it Downloaded 2 June 2026
- Dr. Emrys Nakato-Whyte, Bioinformatician, commissioned by this newspaper. Asked to reconstruct, from public data alone and without being told why, whether accession order predicts submission order May to June 2026 Reports that across 40,000 records the correlation between number and submission time is 0.31, and within a single block it is 0.94.
- Dr. Ngozi Backstrom-Iwu, The submitter who filed version 2. Contacted in writing with the full draft and the audit trail attached; declined a telephone call twice and said why; replied at length July and August 2026 Her reply is printed at the foot of this page, entire, and it is the best thing on it.
Documents
- PX-2029 — eDNA survey, eleven sites, one Washington watershed — protocol, species list and raw sequence accepted
What we could not confirm
- Whether the collection date on version 1 was a transposition or a mistake of some other kind. We can show the field changed from 2020-11-03 to 2020-03-11 and we have the submitter's account of copying it from a ward form. We have not seen the ward form. The institute says it exists and refused to show it to us on patient-confidentiality grounds, which we accept and which leaves us relying on one person's memory for the one detail that makes the change innocent.
- Whether the fourteen retrievals of version 1 are fourteen people. Server logs count requests, deduplicate nothing, and the archive discards client addresses after ninety days. Fourteen requests could be one determined reader fourteen times, or one crawler in a bad afternoon, and we would have no way of telling.
- Whether any record anywhere has ever been suppressed quietly. The status model has a suppressed state, and a suppression that never reached the audit trail would be invisible to the method we used. We examined status transitions in one year, at one archive, across 2,300 submissions. That is a sample, not a system, and we should not be read as having cleared the system.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
The Meridian TelegraphDELETED NINE DAYS BEFORE: The Sequence That Vanished From The World's Archive
Read the status change as a deletion and the nine-day interval as intent, then built a chronology of foreknowledge on top of it.
Had that we did not
The nine days, counted correctly, and the version number, quoted correctly from the banner.
Left out
That the banner they photographed is a hyperlink, that the link works, and that clicking it takes eleven seconds and produces version 1 in full.
-
Signal & LedgerAccession Ranges: What Your Institution's Block Actually Encodes
Turned the arithmetic into a competent explainer for laboratory managers and left the outbreak out of it entirely.
Had that we did not
The allocation table, laid out more clearly than we have laid it out, with the block sizes correct.
Left out
The eleven-day hold, which is the only number in this story that surprised anybody we showed it to, including us.
-
They Buriedthis newspaperThe Sequence Was Not Withdrawn. It Was Replaced, By Its Own Second Author.
Asked for the submission audit trail, the version history and the block allocation table, and used our own deposited sequences as a control.
Had that we did not
All forty-one audit rows, the allocation table recomputed, and the hold-length distribution for 2,300 submissions.
Left out
We spent nine weeks on a date field and four days on the woman who corrected it, and then put her in a headline she did not choose and cannot get out of. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Dr. Ngozi Backstrom-Iwu, the submitter who filed version 2 on 9 July 2026 in writing, with the full draft, the audit trail and the allocation table attached, and four weeks to answer. Replied by email on 6 August 2026, asking that it run entire. Printed unedited, including the parts about us.
I will answer your questions and then I would like to say something you did not ask about.
Yes, I made the change. The collection date on the version I found was 2020-11-03. The date on the ward form in front of me was written 11/03/2020 in a hand I could read perfectly well, and the collection had happened in March, and I corrected it, and the system asked me if I was sure, and I said I was sure, and it made a version 2. That took about ninety seconds. It was the fourth thing I did that morning and I do not remember doing it.
Now the parts you have got wrong.
I am not the 'second author'. An archive submission does not have authors. It has submitters, and there were three of us, and I was listed second because the form sorts alphabetically. You have taken a phrase from academic publishing, where I was in fact fourth on the paper, and welded it onto a database record, and you have done it because 'its own second author' is a better clause than 'one of three people with edit rights'. Your headline is built out of a term that does not exist in the thing you are reporting on.
You also write that the truth here is 'stranger than the legend and just as innocent'. It is not stranger. A hold is the most ordinary thing in this work. Every one of my colleagues uses one. We use them because a journal will refuse a paper if the data went public before the embargo lifted, and so we set a hold, and we release it when the editor says so, and the median is eleven days because the median editorial turnaround is about eleven days. You have found the plumbing and written about it as though it were an oxbow lake.
And here is the thing I actually want printed.
Eleven outlets have written about this record since 2021. I have the list. Not one of them wrote to me. Not one wrote to my institute. Two of them printed my name. One printed the name of my department and the floor it is on, and for about a year I did not put my badge on the outside of my coat when I left the building.
You are the twelfth, and you wrote to me, and you sent the whole draft, and you gave me four weeks, and you have printed this without cutting it. I am grateful and I am not going to pretend otherwise.
But I would ask your readers to notice what the standard is here. One email. That is the entire difference between your newspaper and the eleven before you. It took you nine weeks to establish that a database has version control and about ninety seconds to find my address, which is on the paper, which is public, which every one of them had already read.
Do not let anybody tell you that you have done something difficult. You have done the thing that was always available, and the reason it looks impressive is that nobody else bothered, and that is not a compliment to you. It is a description of the field.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.