I am my university’s Research Integrity Officer — I investigate data fraud for a living — and when I finally pulled the raw dataset from the 2021 grant study and ran the digit distribution analysis, I understood that my mentor had fabricated the statistics, and my name was the co-author on every paper that used them.

I am my university’s Research Integrity Officer — I investigate data fraud for a living — and when I finally pulled the raw dataset from the 2021 grant study and ran the digit distribution analysis, I understood that my mentor had fabricated the statistics, and my name was the co-author on every paper that used them.

My name is Nadine Ashby. I am my university’s Research Integrity Officer. I investigate data fraud. I have spent six years building a reputation for objectivity in this role — and Dennis Cullen has spent those same six years using my objectivity as a citation.

I’ll tell you what this job actually is before I tell you what it found.

Last spring, a graduate student in the chemistry department filed a self-report with my office. His thesis committee had flagged anomalous results in his third chapter, and rather than wait for a formal complaint, he walked into my office and sat down with his laptop open. He was twenty-four. He had driven three hours to attend this university on a scholarship that did not cover his housing. He pulled up his dataset and said: “I think I made a mistake.”

I ran the Benford’s Law analysis while he watched. Benford’s Law describes the expected frequency of leading digits in naturally occurring datasets — the number one appears as the first digit roughly thirty percent of the time in real-world measurements, with each subsequent digit appearing progressively less often. Fabricated data almost always fails this distribution because humans, when inventing numbers, unconsciously distribute digits more evenly. It is a forensic tool and it works.

The output flagged eleven entries. I looked at the entries. I looked at the student. I said: “Walk me through what happened with these eleven.”

He had transposed two decimal places on a subset of measurements during transcription from his lab notebook to his spreadsheet. He showed me the notebook. The original values were there, in pencil, uncorrected. I ran the corrected data. The flag cleared.

I wrote the finding the same afternoon: No evidence of intentional data fabrication. Recording error identified and corrected. Original data preserved in investigator’s notebook and verified against amended spreadsheet. Committee may proceed with defense review.

The student sat across from me while I printed the letter. His hands were in his lap. When I handed it to him he read it twice without looking up.

I told him: “One is human. One is a decision. This was human.”

He cried, briefly, with his face turned toward the window. I gave him time. Then I walked him to the door.

That is what this job is. You learn to read the difference between a mistake and a choice. The two leave different signatures in the data — and different signatures in the person sitting across from you.

I presented Benford’s Law to the university’s annual research compliance training three months later. Forty-two faculty, twenty-six graduate researchers, eight department administrators. I showed two slides: a naturally occurring digit distribution from a published atmospheric study, and a fabricated distribution from a retraction case I used as an anonymized example. The difference is visible to the naked eye once you know what you’re seeing. The natural distribution is jagged and irregular, tilted heavily toward low leading digits. The fabricated distribution is suspiciously even, almost flat across the digit range.

I showed both slides. I said: “Fabrication is almost always detectable in aggregate. Individual numbers can be invented. Patterns cannot.”

A junior faculty member in the third row raised her hand: “Can you tell if someone fabricated a dataset just from the published numbers?”

I said: “Most of the time, yes.”

I advanced to the next slide. The room was quiet in a specific way — the way rooms get quiet when people are recalibrating what they thought they understood.

I had a row of green hardcover field notebooks on my office credenza. One per study, labeled in black permanent marker on the spine. When students asked about data management — and they asked often, usually after an integrity training when they were suddenly anxious about their own records — I showed them the row and told them: “Paper can’t be reformatted. That’s why I still keep one.”

The 2021 Catchment Study notebook had been on that credenza for three years. Green hardcover, Rite in the Rain waterproof paper — the kind field researchers use because standard notebooks dissolve in rain. The spine read CATCHMENT STUDY 2021 — N. ASHBY in my handwriting. I reached past it every day to pull the binder for whatever active case was on my desk. I had looked at that spine hundreds of times. It meant: organized, complete, archived.

It meant nothing yet.

The surface crack came six months before everything else. Dr. Sonja Lindqvist, at the University of Michigan, sent me an email about a partial replication she was running of the 2021 study. She used different sampling sites in a different watershed. Her message was collegial, careful, the kind of note academics send when they want to flag something without accusing anyone of anything:

“Our effect sizes are coming in quite a bit lower than your 2021 baseline. Could be site variation, but thought I’d flag it.”

I replied: “Interesting — will follow up.”

I filed the email. I did not follow up. I told myself it was site variation. The 2021 study used catchment sites in a different region, different soil composition, different seasonal precipitation patterns. Replication at different sites routinely produces different effect sizes. This is expected. This is science.

That was six months ago.

And then there was the memory I carried without knowing it was a memory that mattered: Dennis, three years ago, appearing in my office doorway with a bottle of wine and a rare afternoon off. The NIH grant had just been awarded — the one built on the 2021 data, the one that brought the department $2.3 million over five years. He set the bottle on my desk and stood there looking as though he’d done something he was genuinely proud of.

He said: “Your field data is what made that application competitive. You’ve been the best thing to happen to this department’s credibility in fifteen years.”

He called me by my first name. He was generous and specific. He was the person I had done my post-doctoral fellowship under, who had trained me in the very precision he was now complimenting.

I believed him. I was not wrong to believe him. That’s the thing I keep returning to — not to excuse what I missed, but because it matters: I was not wrong to trust him then. The trust was built on real evidence. Six years of it.

I pulled the 2021 study dataset from the university repository in February — not because I was looking for anything. I was preparing a research methods seminar for eight second-year PhD students. I wanted to use a recently published departmental study as a worked example of a clean, compliant dataset. The 2021 catchment study was the obvious choice: federally funded, peer-reviewed, published in three reputable journals, co-authored by the person who runs the research integrity office. I would run the Benford’s Law analysis on it live, in front of the students, and show them what an unflagged result looks like.

I uploaded the dataset. I ran the analysis.

The output flagged.

The leading digit frequency was non-natural. The distribution did not match the expected Benford’s curve. I looked at the screen. I assumed a software error — this had happened once before, years ago, with a corrupted file import. I cleared the cache, re-uploaded the dataset directly from the journal’s supplementary data portal, and ran it again.

The flag came back.

Same result. Non-natural leading digit distribution across the primary measurement columns.

I did not call anyone. I sat for a moment and then I pulled the 2021 field notebook from my credenza.

I was not looking for evidence. I was looking for my site measurement protocol — the exact procedure I used to record values, the instrument calibration sequence, the sampling interval. Standard preparation for presenting research methods. I opened the notebook looking for site coordinates and procedure notes.

I found the data column first.

The notebook opens left-to-right, site measurements in the left column, condition notes in the right. October 15, 2021. Site 1. The measurement in my handwriting: 0.47. I looked at Published Table 1 on my screen. Site 1 mean: 0.11. I looked back at the notebook. October 16. Site 1. My measurement: 0.52. Published Table 1, Site 1: 0.11.

I looked at the next page. Site 2, October 15. My measurement: 0.81. Published: 0.12.

I set the notebook flat on the desk beside the screen and did not close either.

The field measurements showed variance from 0.34 to 1.87 across all sampling sites and dates. The published table showed variance of 0.09 to 0.14. The published numbers were not a statistical smoothing of my measurements. They were not a controlled adjustment, not an outlier exclusion, not a model correction. They were different numbers entirely. Every entry in the published table was a different value from what I had recorded. Every single one.

Here is what I remembered, with the notebook open in front of me.

I had stood in the marsh on site 4 in October 2021 wearing rubber boots that came to my hips. The water was cold enough that I could feel it through the neoprene after twenty minutes. It was raining — not heavily, but steadily — and I had the notebook zipped inside my vest and pulled it out with wet gloves to record measurements. Site 4 read anomalously high on day three: 1.63, where the other sites were running between 0.4 and 0.9. I wrote a note in the right column: High variance — check for local runoff. Site 4 may have contamination variable. Discussed with DFC. He said handle in analysis.

I recorded it exactly as I had measured it. I pressed my pen down in waterproof ink on waterproof paper in the rain because that is what field data is: what you actually measured, not what you expected to find.

At the end of that week I exported the data to an Excel file and sent it to Dennis with a message: “Variance is higher than expected on site 4 — suggest we discuss in analysis phase.”

He replied: “Got it — I’ll handle the statistical treatment.”

He handled it. The notebook page still had the field note where I’d written it. His initials, mine. The conversation I’d trusted was in my own handwriting, archived.

I remembered the paper submission meeting three years ago — Dennis bringing the final manuscript to my office for co-author review. I had read the methods section and the results carefully. Table 1 had shown clean, consistent correlations across all sampling sites. I read it twice, because Table 1 is always where you look hardest.

I had said: “The variance on site 4 came down significantly.”

He had said: “After we controlled for the contamination variable you flagged — yes. The outlier sites were excluded in the final model.”

He was calm. He was specific. He had used my own note — the contamination variable I had flagged in the field — as the logical explanation for why the values had changed. I had set my green pen down on the manuscript without marking anything. I had told myself: he is the statistician, I am the field researcher. The division of expertise is appropriate. He explained the correction in terms that referenced my own field observation.

I signed the co-author agreement. The pen moved the way it moves on a thousand documents. I was not wrong to sign it. I was given an explanation that referenced the evidence I had provided. I had no reason to pull the field notebook and compare values line by line. Nobody does that. Nobody assumes the statistician replaced the numbers rather than adjusted them.

I know this because I had been the investigator in a case four years earlier where the distinction between adjustment and replacement was exactly the line between defensible methodology and fabrication.

A doctoral student in the biology department had fabricated gel electrophoresis data in two published papers. He was twenty-eight. He was three months from his dissertation defense. He was on a student visa. When I received the complaint from his committee I spent four days reviewing his data files, comparing his published western blot images against the raw gel photographs. I held the two sets of printouts side by side under my desk lamp. The fabricated images were not crude — they were careful. He had been meticulous about it.

I wrote the finding: Evidence of intentional data fabrication. Recommend retraction of both papers and suspension of degree candidacy.

I held the cursor over the Submit button in the case management system for three seconds. I clicked it.

The doctoral student lost his academic career. I wrote the finding because the evidence required it and because the objectivity of my position is what gives the finding its authority — it cannot be a negotiation, it cannot be softened because the consequences are severe. I was objective because it was not my name on the paper.

I understood, sitting at my desk in February, that it was now my name on the paper.

I retrieved Dr. Lindqvist’s email from six months ago and read it again with the field notebook open beside me. “Our effect sizes are coming in quite a bit lower than your 2021 baseline.” I looked at the published Table 1. I looked at my own notebook values. Her replication results were consistent with my field notebook measurements — variance in the range I had recorded — and inconsistent with the published table. Someone else’s real data matched my real data. Neither of them matched what was published under my name.

I wrote the two variance ranges on a sticky note. My field measurements: 0.34–1.87. Published table: 0.09–0.14. I pressed the sticky note against the top of the data column in the notebook. The notebook I had kept for three years as evidence of completed work was now evidence of a discrepancy between what I had measured and what was published under my name. I had not written anything new in it. I was looking at what I wrote four years ago. The handwriting was mine. The published numbers were not. The spine was soft from the single field season it survived — the rain in October, the mud on site 4, the week I spent recording the variance he chose to erase.

My coffee sat on the desk until it was cold.

The third layer came when I opened the site visit slide deck Dennis had sent me the previous Thursday for review. I was listed on the agenda as co-presenter for the data methodology block. I had not yet looked at his slides. I opened the file. Slide 7: Data Validation and Quality Assurance. A table with two columns. Left column: process description. Right column: responsible party. Field data collection and quality assurance: Nadine Ashby, Research Integrity Officer, Department of Environmental Sciences.

He had listed me as the quality assurance lead for the 2021 study in a document he was preparing to present to NIH program officers. He was going to tell the federal funding agency that a Research Integrity Officer had verified the data. He had listed this without asking me. Without telling me. He had named my professional role as the validation instrument for data I had never seen in its published form until two weeks ago.

The site visit was in twenty-three days.

I closed the PDF. I opened the Benford’s Law analysis output and saved a copy to my personal encrypted drive. I photographed the field notebook’s first two data columns with my phone — the timestamp visible in each frame. I opened the ORI online complaint portal. I read the form instructions from beginning to end. I did not call Dennis.

I began typing at 8:43 PM.

Here is what I have to account for before I tell you what happened next: the pattern I tolerated and for how long.

Three years. That is the window. From the paper submission meeting — where I set down my pen without marking anything, where I accepted the contamination-variable explanation without checking the notebook — to the February afternoon when I sat with both documents open on my desk. Three years in which I signed the co-author agreement on the first paper and both subsequent papers that used the 2021 data as their baseline. Three years in which I received Dr. Lindqvist’s email and filed it under site variation without following up. Three years in which I attended faculty meetings where Dennis cited the catchment study findings and I did not cross-reference my memory of the field measurements against the published values, because I trusted the division of labor I had agreed to. I was the field researcher. He was the statistician. That is how it had always worked. I had no reason to audit my own published work.

I know what I saw and when I saw it. That is not the same as knowing what it was. The email from Lindqvist was a flag. I read it as site variation because site variation was plausible. Plausibility is not the same as truth. I chose plausibility because the alternative required me to question the person who had trained me, mentored me, co-authored with me for six years — and that was a question I was not ready to ask.

I am accounting for this now. Not in self-pity. In precision. This is what investigators do: we account for what we chose to see and what we chose not to.

The Monday morning after I began the complaint, I arrived at my office at 7:45 AM to find a new email from Dennis waiting. Subject: NIH Site Visit — Updated Agenda. He had added me as co-presenter for the forty-five-minute data methodology block. His note was collegial and specific, the way he always was: “The NIH program officers specifically asked about our QA processes — thought you’d want to walk them through it directly. You’ll be the most credible voice in the room for this.”

I sat with that email for a long time.

Filing during an active site visit window would look deliberately disruptive — that is what anyone watching would think. Waiting until after the site visit would mean I had co-presented data I knew was fabricated as validated research, which is precisely what his slide deck already said I had done. He had placed me at the center of the fraud documentation so completely that the only path forward that did not implicate me further was the one I had already taken the night before.

I was still listed on the agenda. The site visit was in eleven days.

Across the hall, through the glass wall of his office, Dennis was preparing his presentation. The windows of his office faced east and the morning light came in at an angle across the grant award plaques on his wall — three of them, NIH, NSF, a private foundation. A framed federal flag from the last review panel. He was on the phone, walking through the room layout with his department administrator, relaxed in the way he was always relaxed before a federal review. He had done this three times before. He knew how NIH site visits worked: the program officers ask about QA processes, they see a compliance credential in the room, they move on.

He reviewed his slide deck. The graphs were clean. The correlations held. The narrative was strong. On Slide 7, Nadine Ashby’s name and title. He looked across the hall through the glass wall at her office. She appeared to be working on a document.

Good, he thought.

He told his administrator to include my bio in the site visit binder as Data Quality Assurance Lead for the 2021 and 2022 Catchment Studies. He named my role in the fraud documentation without asking me. He did not consider that this required my consent. He was thinking about the second tranche of the $2.3 million that released after site visit approval. He had presented to NIH panels three times. He knew what happened when someone with a compliance credential was in the room.

I submitted the ORI complaint at 6:47 AM, twelve days before the site visit.

I attached four files: the Benford’s Law analysis output with the non-natural leading digit flag clearly marked; the field notebook photograph comparison showing my handwritten measurements beside the published Table 1 values; a copy of the published table itself; and Dr. Lindqvist’s replication email. I clicked Submit. The portal returned a confirmation screen with a case number.

I wrote the case number in my paper notebook — the new one I kept on my desk, not the 2021 field notebook — in blue ink. I closed the laptop. I opened a new document and began writing the methodology section I would actually present if the site visit proceeded: my real field methods, my real measurement protocol, the real variance ranges I had recorded at every sampling site in October 2021.

I did not know yet who would be in the room when I walked in.

I did not know whether the ORI would move before the site visit or after. I did not know whether I would be asked to step down from the agenda or remain on it. I did not know whether the university administration had been notified. The automated acknowledgment arrived at 6:49 AM with my case number and a processing timeline of five to fifteen business days.

I was still scheduled to co-present in twelve days.

I kept writing.

The NIH site visit was scheduled for 9:00 AM on a Tuesday in late February. The university’s main conference room — mahogany table, twelve chairs, a ceiling-mounted projector, windows that faced the campus quad. I arrived at 8:40 AM with the methodology materials folder and the 2021 field notebook in my bag. Dennis was already at the front, laptop connected, slide deck running on the screen. He was arranging handout packets at each place setting. He looked up when I came in and nodded — the gesture of a person managing logistics who is not concerned.

I sat to his left. I opened the folder in front of me.

The NIH program officers arrived in a group at 8:52 AM — three of them, two men and a woman, institutional lanyards, rolling briefcases. The university provost took the seat at the far end of the table. Six department faculty filed in. Dr. Helen Park, the department’s associate chair, sat directly across from me with a copy of the printed site visit agenda.

At 8:58 AM, a fourth person entered the room who was not on the agenda.

She introduced herself before taking her seat: Dr. Constance Fisk, ORI Division of Investigative Oversight. She sat at the mid-point of the table, between the NIH program officers and the university faculty. She placed a leather portfolio in front of her and opened it to a blank page. She did not explain why she was there. She did not need to.

Dennis looked at her. His hands, which had been flat on the table in front of his laptop, went still.

He said: “We weren’t notified of a concurrent ORI inquiry. This is irregular.”

Dr. Fisk said: “Federal research integrity inquiries don’t require advance notice to the subject of the inquiry.”

He looked at me. The room was quiet in the way rooms get quiet when the surface of things has shifted and everyone is waiting to understand what is underneath.

He said, quietly: “What did you do?”

I said, not quietly: “I submitted a data integrity complaint twelve days ago. I’m the Research Integrity Officer. It’s my job.”

The lead NIH program officer set down his pen.

Dennis straightened in his chair. He said: “The statistical analysis was validated by an external biostatistician before submission.”

I said: “The digit distribution in the published dataset doesn’t follow Benford’s Law. My field notebook from the 2021 study shows variance ranging from 0.34 to 1.87. Table 1 of the paper you submitted to NIH shows 0.09 to 0.14.”

He said: “Field variance is controlled in the analysis phase. That’s standard practice.”

I opened my bag. I took out the 2021 field notebook. I set it open on the conference table, in the center of the table, spine flat, the first data column facing upward. I said:

“My handwriting. October 2021. Site 4, day three. Variance: 1.63. You were not in the field. I was.”

The lead NIH program officer reached across the table and took the notebook. He opened it to the first data column. He did not look up at Dennis for the next two minutes. His index finger moved down the column of handwritten measurements, paused at the site 4 entries, moved to the corresponding published table values on the printout in front of him. He turned the page.

At the far end of the table, the provost closed the site visit agenda folder. He set it face-down. He picked up his phone from the table and held it. He did not put it down.

Dr. Helen Park pushed her chair back from the table by four inches. She looked at her copy of the published paper — the one that had been in the faculty handout packet Dennis had distributed. She looked at the notebook in the NIH officer’s hands. She did not look at Dennis again.

The Benford’s Law analysis output was in the Benford’s Law print. The lead program officer held the notebook in one hand and the analysis printout in the other. He asked Dr. Fisk: “Is this document associated with the active inquiry?” She said yes.

I said: “The Benford’s Law analysis flags non-natural digit distribution in the published dataset — the leading digit frequency does not match naturally occurring data — and the field notebook sitting on this table shows the values I actually measured, which are not the values that appear in Table 1 of the paper that carries my name.”

Dennis was quiet for a moment. Then he said: “The statistical analysis was directionally correct. The contamination variable on site 4 warranted exclusion. The overall trends are real.”

Dr. Fisk wrote something in his portfolio without looking up.

Dennis gathered his presentation materials. He straightened the edge of his folder against the table — once, carefully, as though the alignment of the papers required his attention. He picked up his laptop. He said: “I built this department’s NIH portfolio from nothing. Everything in those papers is directionally correct.”

He left the room without making eye contact with me. Dr. Fisk noted the time of his departure in her records: 9:47 AM.

The NIH program officers placed the active $2.3 million grant under funding hold pending the outcome of the ORI inquiry. The provost called the university’s legal counsel before the remaining attendees had left the room. Dr. Park gathered the distributed handout copies of the published paper and stacked them face-down on the corner of the table. She did not take hers with her when she left.

The afternoon light had gone flat in my office by the time I returned from the provost’s suite. The building’s HVAC hummed evenly. The air smelled of old paper and the coffee I had left on my desk that morning, now cold and completely still.

I carried the field notebook back with me from the conference room. I had carried it in both hands, holding it across my chest, through the corridor and up the stairs. When I walked into my office, I set it on my desk, not the credenza. It was no longer an archived record. It was evidence. A copy was with the ORI investigator. A copy was in the provost’s administrative file. But this copy, the physical object, I kept.

I opened it to the first data entry: October 15, 2021. Site 1. Temperature: 8 degrees Celsius. Sky: overcast. Measurement: 0.47. I looked at the page. My handwriting was careful and small. I traced the edge of the paper with my thumb. The spine was soft from the rain it had survived. I read the page from top to bottom. Every number I wrote was still there. Nobody had touched them. That was the one thing that had not happened to this notebook. The data was exactly what I had recorded in the cold mud of the wetland. It had always been exactly what I recorded. He had tried to erase it, but he had only ignored it. The numbers were still there. That was the thing I would keep.

The ORI finding will take six to eight months. The preliminary notice stated there is no evidence I was aware of or participated in the fabrication. It clears me. The retraction notices for the three published papers will be filed by the journals in the coming weeks.

The retraction notices will carry all co-author names. In the academic publication databases, Nadine Ashby’s name will permanently appear in the retraction notices alongside Dennis Cullen’s. The ORI finding is a separate document. It is not linked directly from the database search results. The retraction notice does not say “Nadine Ashby was innocent.” It says: retracted due to data fabrication. It lists the authors. My name is one of them. That cannot be corrected.

I sat at my desk and looked at the credenza. The row of green notebooks stood exactly as they had that morning.

I opened my bottom drawer. I took out a new green hardcover field notebook — Rite in the Rain, identical format, completely blank. I set it on the desk next to the 2021 log. I opened it to the first page. I took my pen and I wrote the date in the top right corner. I wrote the name of the new compliance audit I was beginning tomorrow. I wrote: Day 1.

I set my pen in the gutter of the spine. The blank lines waited.

Dennis thought the field researcher and the compliance officer were two different jobs. He forgot that I brought the same notebook to both. He forgot that I have been writing down what I actually measured since 2004 — and paper doesn’t reformat itself to fit anyone’s analysis.