The 00Z sounding from North Platte wasn't there. I checked the feed twice. It hadn't been there reliably in over a year.

I pulled up the 12Z from Topeka. Six hours old and two hundred miles east of where I needed it. The profile showed a capped atmosphere with decent moisture below 700 millibars, but the cap was the question, and the cap was always local. You can't borrow someone else's sounding and know what's happening above your county.

The office was just me and Ruiz. He was working aviation. Three of the six AWIPS stations were dark. Not broken. Nobody sitting at them. We'd been running two-person shifts since spring, and the fluorescents buzzed the way they always buzzed, and the coffee in my mug had gone cold an hour ago.

I had a line of cells building southwest of the CWA. Maybe sixty miles out, moving northeast at thirty knots. Radar showed decent reflectivity, nothing remarkable. But surface dewpoints were in the low seventies, winds backed at the surface, and on visible satellite I could see towers punching up and shearing northeast along the line. The environment was telling me something the numbers couldn't confirm. The 12Z HRRR had been hinting at supercell potential by late afternoon. The 18Z would tell me more, except the 18Z was initializing with less upstream data than it used to, and I'd learned to hold that loosely.

I checked the mesonet. Four of eleven stations in the western half of my area were reporting. The others showed the same flat line they'd shown since their maintenance contracts lapsed. I used to file tickets on those. Now I worked around them.

The HRRR came in at 1847 UTC. I loaded it into D2D and overlaid it with the RAP and the NAM. The HRRR wanted to blow up the line into discrete supercells by 22Z, tracking them through the population corridor along the interstate. The NAM was slower, weaker, kept things linear. The RAP split the difference.

Normally I'd check the HRRR against the latest sounding to see if the model was handling the cap right. But I didn't have a sounding. I had a six-hour-old profile from Topeka and a surface ob from a station forty miles south. The model was telling me something I couldn't verify about the thing that mattered most.

I pulled up the ensemble playout. Twenty members, and the spaghetti was wide. Couldn't agree on timing by more than three hours or placement by more than a county. That spread used to mean the atmosphere was genuinely uncertain. Now it could mean that, or it could mean the initialization was thin because the observations that should have anchored it weren't there. I couldn't tell the difference.

That was the whole problem.

Ruiz looked over. "What are you seeing?"

"Maybe something. Maybe not."

He nodded and went back to his TAFs. We'd both gotten used to saying it.

I watched the radar for twenty minutes. The line was organizing. One cell on the southern end was starting to show rotation, just a suggestion of it, a slight hook in the reflectivity that could be real or could be an artifact of the beam angle at that range. I needed velocity data to confirm, and the velocity looked suggestive but not definitive. At closer range I'd know. But at closer range it would be over someone's house.

I knew the criteria. Radar-indicated tornado or a spotter report. All I had was a model that wanted to produce supercells in an environment it couldn't fully see, a radar signature that was suggestive at a range where suggestion was all I'd get for another fifteen minutes, and no upstream sounding to tell me whether the cap would break.

I could wait. Fifteen minutes and the cell would be close enough for the radar to resolve what it was doing. Or I could warn now, based on what I thought was happening, and be wrong. Another false alarm in a county that had gotten three this year already. People stop listening after enough of those. I'd watched that happen too.

But fifteen minutes is a lot of time. It's the difference between a family in a basement and a family in a kitchen.

I sat with it. The fluorescents buzzed. The screens showed what they showed, which was less than they used to show, which was all they were going to show tonight. Ruiz was typing something. Outside I could hear the wind picking up against the building, though that didn't mean anything by itself.

I put my cursor on the warning tool and started drawing the polygon.