The query had been running since 4:47 a.m., which Maya knew because she had woken at 4:47, looked at her tablet, started it, and gone back to sleep for ninety minutes. When she came back down to the kitchen at RMBL365 the kettle was warm — someone had been through earlier — and the synthesis interface was waiting for her with the foraging-trip distributions stacked in a way she hadn't asked for but immediately wanted.
It had grouped 2034, 2037, and 2039 together against the 1998–2022 baseline and broken out 2011 as a fourth panel. The annotation said: 2011 included as nearest-analog low-snowpack year in the pre-warming distribution; suggest checking whether grouping is appropriate.
"You're showing off," she said to the tablet, and made coffee.
The kitchen at RMBL365 in mid-September was a quiet place. The summer cohort had cleared out three weeks ago. Renata, the schools coordinator, came through on Tuesdays and Thursdays; the water-districts coordinator was in Montrose this week. Upstairs Maya had the small room with the slanted ceiling that had been hers in August and September for two seasons now. The aspens above town were turning — not the full burn yet, just the high stands going first, the way they did now in early September instead of late. She had stopped registering this as strange.
She brought her coffee to the table and looked at the four panels.
The 2011 grouping was the right call. It bothered her that the tool had made it before she had. She added a note to the methods slide — low-snowpack analog years; 2011 included on pre-warming community composition — and pulled up the visitation records.
Here was the question she had come for. Her home lab on Sapelo had pollinator-network work going back a decade, but it was salt marsh, Spartina and solitary bees and the wrong axis of stress. The mechanistic hypothesis she wanted to test — that Bombus foraging distance distributions would broaden in low-snowpack years not through a shift in the mean but through a fattening of the right tail, individual workers ranging farther because nectar reward density at any single patch had dropped below some threshold — required paired records of forage trips, floral density, and nectar standing crop at the inflorescence level, across more than a decade, in a system where snowpack varied. Three places on Earth had this. RMBL was one. The basin synthesis tools could pull it in an afternoon.
She had spent four years getting to a place where she could ask the question in an afternoon. The afternoon felt deserved.
The right tail had fattened. She zoomed in. The 2039 distribution had a small but real population of trip-distance estimates above 1.4 kilometers — almost twice the long-baseline maximum — and the tool flagged that thirteen of those trips were from workers tagged in a colony above the Copper Lake talus, where Ipomopsis density had crashed in the July dry-down. She had not asked the tool to make that connection. It had made it.
She sat with it.
The slide for her job talk was open in another window. Mechanistic coupling of snowpack, floral reward density, and Bombus foraging behavior in subalpine meadows. Below it, the placeholder figure she had drawn last week, which she was about to replace with the real one. She did not replace it yet.
At 9:40 her phone vibrated. A text from her advisor on Sapelo: Heard back from Davis search committee. They want a second talk, virtual, mid-October. Congrats.
Maya breathed out. She wrote back thanks, will call later, and put the phone down and looked out the window at the aspens, and then she let herself say fuck yes in the empty kitchen, quietly, because the upstairs room above her had a renter she hadn't met yet who had come in late last night.
She had ten minutes before the call.
She opened the methods note. She and Daniel Whitebear at the Ute Mountain Ute natural-resources office had been working on this draft for fourteen months — a paper about how the basin pollinator records could be cross-referenced with traditional knowledge about plant timing, what counted as evidence on each side, how to write a methods section that didn't pretend the two epistemologies collapsed into one. The methods note was a precursor. Three pages. They had gone back and forth on the second paragraph for six weeks.
Renata came in the back door carrying a box. "Maya. Hi. Sorry. I'm dropping these and going. The fifth graders made them."
"Made what?"
"Phenology cards. For the corner display. They've been tracking serviceberry since March."
"How are they?"
"Better than the grad student last year." Renata set the box down and looked at the tablet. "You getting your tail?"
"I got my tail."
"Good. I have to go pick up Joaquin from the bus. He's home with something."
"Tell him I'm sorry."
"He's playing Minecraft. He's fine." Renata paused at the door. "Are you here Friday?"
"Flying Thursday."
"Atlanta?"
"Atlanta."
Renata made a face Maya had seen a lot of basin people make about Atlanta in September, and then she was gone.
At 9:58 Maya pulled up the call. Daniel came on at 10:01 from an office she recognized — the window behind him showed the Sleeping Ute, and a corner of a poster about uranium-mill tailings she had been meaning to ask him about. He was holding a coffee.
"You saw the redline?"
"I saw it. I want to push back on one thing."
"Push."
She talked him through the second paragraph — her concern that the language he had drafted made the basin records sound like they were checking the traditional observations, rather than the other way around, or rather than neither. He listened. He was quiet for a long time and then he said, "Yes. I felt that too. I couldn't get to what was wrong."
"What if we drop validate entirely and say the two sets of observations are placed in conversation?"
"Placed in conversation is bloodless."
"Yeah."
"Read against each other."
"Better."
They worked on it for thirty-five minutes. At one point Daniel's daughter came in and asked for the car keys and Daniel said, in Ute, something that made her laugh, and then in English: "Sorry, Maya. Saturday. Ask me Saturday."
When they hung up Maya had a paragraph she could live with and a feeling she recognized from the good days, which was that the work was a little bit further along than it had been at breakfast.
She went back to the foraging distributions.
She thought about the search committee at Davis, and about the talk she would give, and about the fact that one of the questions she would certainly get was: can't a foundation model do this synthesis now? what's the basin-specific value-add? She had been thinking about this question for months. The startups had launched in the spring; her postdoc cohort had a group chat where they traded screenshots of the more egregious commercial syntheses, the ones that confidently misidentified subalpine species or invented references. But the better commercial tools were getting better. She had used one in July to check her own pipeline against and it had been — fine. Wrong about two things she cared about, right about a third she hadn't noticed.
What she would say at Davis: the value was not in the synthesis. The value was in the paired records that no general model could synthesize because they did not exist anywhere else. The thirteen long-trip workers above Copper Lake had been tagged by a field crew that had been working the same slope for nineteen summers. The methods note she had just edited would not exist without a relationship the basin had built over a decade of standing meetings in a room she could see from where she was sitting. The general tools were good at the readable surface. The basin was where the surface got made.
She thought this might be a slide.
She drafted, on a napkin from the kitchen counter:
What general models can do: read everything ever written.
What general models cannot do: be in the room when the question is asked.
It was not quite right. She would argue with it for a week. She pulled the napkin closer and crossed out be in the room and wrote help decide what counts as the question, and then she crossed that out too, and then she stopped and laughed at herself and got up to make more coffee, because the kettle had gone cold again and Joaquin's fifth-grade phenology cards were still in the box by the door, and she had not looked at them yet, and she wanted to.