Funny how he can recall my coffee order down to the garnish but couldn’t manage to explain why he flinched when I tried to kiss him.
“How did you—” I start, then stop.
He’s already halfway to his workstation, but he glances back. “How did I what?”
“The cinnamon.”
Something flickers across his face. “I have a good memory.”
He says it like it’s nothing. Like he hasn’t just proven that somewhere in that over-packed brain, there’s a file labeled AUDREY: COFFEE PREFERENCES. A file he’s maintained for god knows how long. Meanwhile, I’ve spent three months trying to debug our entire history, and I still can’t locate the error.
He has the source code. I’m stuck reverse-engineering from the crash report.
The imbalance is infuriating.
I force my face neutral and focus on the screen, even as my hands cradle the coffee. It’s warm and real and the closest thing to comfort I’ve let myself have in months.
“Did you have a chance to look at the models?” he asks after a minute, voice low.
“I did. The hybrid model looks solid.”
He straightens slightly—that’s his version of preening, I remember, that subtle shift when someone praises his work. “You think so?”
“I think it needs more testing before I’m convinced. But it’s a strong starting point.” I turn back to my screen, pulling up his documentation. “Walk me through your methodology for the stability metrics. I want to understand how you weighted the variables.”
He rolls his chair over, careful to leave a professional distance between us. But ‘professional distance’ turns out to be close enough that I can smell him—something clean and woodsy, his soap or shampoo—and close enough that I’m suddenly, acutely aware of the heat radiating off his body.
My breath catches. My skin prickles. My brain sends up a red alert that I aggressively ignore.
Focus. Work. Professional friendship.
“Here,” he says, pointing at a data cluster. “I used a modified Bayesian framework for the initial weighting, but I adjusted the priors based on the empirical data from the most recent trials. The traditional model assumes uniform distribution, but our neural signal data shows clear clustering around specific frequency ranges, so?—”
“So you weighted those frequencies higher in the priority algorithm,” I finish. “That’s why your latency numbers are better than what I was projecting.”
He looks at me. That look—the one that used to make my chest tight, as if he was seeing something no one else could see. “You were projecting latency numbers?”
“I had some ideas.” I shrug, trying to play it off. “Jet lag is messing with my sleep. So...”
“Can I see them?”
I hesitate. My projections are rough, more napkin math than rigorous modeling. But his eyes are bright with genuine curiosity, and I remember this feeling—the thrill of two minds working on the same problem, building on each other’s ideas.
“They’re not polished,” I warn.
“I don’t need polished. I need the thinking.”
I pull up my notes. They’re messy, scattered across three different files with contradictory annotations, but Logan doesn’t seem to care. He leans in closer, scanning the data, and I watch his expression shift as he processes.
“This is good,” he says softly. “This is really good. You were approaching the frequency distribution from a completely different angle than I was, but if we combine your framework with my Bayesian model?—”
“We might be able to optimize for both latency and stability.”
“Exactly.” He’s already reaching for my keyboard. “May I?”
I nod. His fingers fly across the keys, pulling up a split screen, dragging my files next to his. He’s muttering under his breath—numbers, variables, half-formed hypotheses—and I lean in too, caught up in the momentum.
“If we adjust the weighting here,” I say, pointing, “and use your prior distribution as the baseline?—”
