Roisha Maharjan: Learning to listen to the data

Roisha Maharjan sits on a rock on the right of the frame. Behind her is a sunset over a large body of water. — Roisha Maharjan is a graduate student in Virginia Tech's Charles E. Via Jr. Department of Civil and Environmental Engineering. Photo courtesy of Roisha Maharjan.

This piece was written in the fall of 2025 by GRAD 5144 (Communicating Science) student Roisha Maharjan as part of an assignment to write a personal narrative about her research.

The night the plots finally made sense, the lab smelled like burnt coffee and warm plastic. The HVAC hummed while a loose window clicked in the wind. On my screen, points huddled below the 45-degree line like they were actively refusing to cooperate.

In my work to assess the potential for soil movement in earthquake situations, I’d run hundreds of site response cases on gravelly soil profiles. The code worked; the results didn’t. Every plot I generated felt slightly out of tune with what we knew from observations: how the data should behave, how the trends should bend. This mismatch plot planted doubt in my mind about the credibility of my own code.

I wondered whether I had made a quiet mistake somewhere, something small enough to hide but large enough to distort science, so I retraced everything. Line by line. Variable by variable. Then I compared my outputs with those from my professor’s code.

That was the turning point. The issue wasn’t the code at all. It was the results themselves. What I wanted was simple: proof I belonged in this work and a figure I could show my advisor without apologizing.

I messaged a friend: “I think my gamma values are lying to me.”

He wrote back: “Plot the simplest thing.”

So I did: one motion, one layer, one stress-strain loop. On the screen, the loop opened like a sideways almond, widened with earthquake shaking, then softened. For the first time all week, “effective stress” felt like something you could see, not just calculate. I rewrote the caption in plain English before touching the code: “This loop shows where the soil stops acting stiff and grains begin to slip.” It wasn’t perfect, but it told me what to do next.

I swapped γmax for γc on the x-axis – quiet variable, big consequence, and the comparison climbed closer to the diagonal. Not a miracle, but a step in the right direction. Then I noticed small truths I’d missed: A “problem” profile had a shallow water table mis-tagged in metadata, two motions with the same magnitude had different frequency content, and the soil cared. I fixed the tags, labeled the axes, and added a loud, honest error message to the pipeline so future me wouldn’t chase ghosts at midnight.

Roisha stands in front of two poster boards with research taped on. She stands on the right and wears a blazer. — Roisha Maharjan presents her work on soil movement during earthquakes. Photo courtesy of Roisha Maharjan.

I was told from the beginning that these kinds of analyses were never really designed for gravel, that they were built and tested in sand. But I ran them anyway, almost stubbornly, and then sat there staring at results that made no sense. The predicted and observed behaviors refused to line up. The ground motions twisted and reacted strangely in the gravelly profiles, as if the soil was constantly shifting personalities depending on how much sand it carried. The bias grew larger and larger, and I kept thinking: What am I doing wrong?

For a while, I again questioned everything: my code, my assumptions, even the credibility of the data itself. But then it struck me: Nothing about the soil had changed. Gravel will always be gravel. I was the one who had to change. I had been trying to force the data into the story I wanted to tell.

So I stopped. I stepped back, listened, and let the data speak in its own voice.

I made a sticky note: “One profile, one motion, one loop. Then scale,” and sent my advisor a short update that began with the English and left the equations to the appendix. He replied in the morning with a thumbs up and “Let’s talk about that mis-tagged profile tomorrow.”

I walked home under a thin, indecisive dawn and felt the week’s static quiet down. The plot still wasn’t perfect, but it finally pointed somewhere honest. And that’s when I understood: Wrapping up a research project or finishing a degree program isn’t about one heroic graph. It’s about weaving together a series of small, deliberate choices, choices you can say out loud, without flinching, choices where you can trust yourself.

That night, I finally learned how to make them.