New equations land. The numbers don't cohere. And for the first time in this universe nobody is wrong — which turns out to be the harder problem, because there is nothing to fix except a field that was left blank, a screen that filled it in, and a reasonable person on the other end who has already given you a reasonable no.
Sarah Morrison caught him at the coffee machine. New urban peak-flow equations for Tennessee had gone live in StreamStats — a real study, a good one, twenty years overdue. Her question was the practical one every hydraulics section in the state was about to ask.
"When do we use the urban ones and when do we use the regular ones?"
Lester said he'd find out. He went back to his desk, delineated a basin in Memphis, and ran both. Rural came back higher than urban. He sat with that, then delineated one in Nashville and ran both again. Urban came back higher than rural.
He picked up the legal pad.
Memphis. Germantown. Jackson. Then east — Waverly, Dickson, Nashville, Murfreesboro, Cookeville, Knoxville, Johnson City. Delineate, run both, write down which came out on top. He did it a hundred times over two days, and the pattern held so cleanly it stopped being a curiosity and started being a boundary.
West of the Tennessee River, rural ran high. East of it, rural ran low.
The reason sat in the rural report, in the table nobody reads: which variables ended up in which equation.
In west Tennessee — Hydrologic Area 4 — imperviousness is in the rural equation. It came out statistically significant, it earned its place, and it carries a published calibration range of roughly a quarter of a percent to thirteen point seven.
In the rest of the state it isn't in the equation at all. Not restricted. Not capped. Absent.
So the two halves of the state fail in opposite directions, and both failures are invisible. Out west, hand the rural equation a basin at thirty percent impervious and it doesn't know the response curve flattens up there — it only ever saw the steep part. It keeps climbing at the slope it learned. It overshoots because it's trying. Everywhere else, hand it the same basin and it returns a country-creek number, because nothing in the equation can see pavement. It undershoots because it isn't trying.
That would have been the whole finding. Interesting, worth a memo, done. Then he looked at the parameter table — the box listing every variable with its minimum and maximum, the box that decides whether you get the out-of-range warning.
Zero to one hundred is not a range. It's the full span of a percentage. It's what a field looks like when there was nothing to put in it — because in that region the variable isn't a predictor, so the report lists it as N/A, and N/A has to become something on its way into a database.
And here's what that does. The one region with a documented limit throws a warning. The three regions with no limit at all throw nothing, ever, at any value — so a basin at thirty-five percent impervious in the middle of Nashville comes back clean. Full prediction intervals. No flag. A number that looks more trustworthy than the Memphis one, and is less.
A missing limit reads exactly like a wide-open limit.
GFAS is a confident wrong answer.
This is worse — it's a blank field wearing the face of a passing check.
He took it down the hall. Sarah first, then the section, then the man above him. The response was the one he'd learned to expect.
Are you sure you read the report? Those are PhDs. They know what they're doing.
And they did. That was never in dispute. Lester wasn't arguing with the study — the study was clean, the variable selection was clean, the table was honest about its own limits. He was arguing with what came out of the application when somebody loaded that table into a database. Different object entirely. But "those are PhDs" only answers the first one, and once a room decides you're questioning the authors it stops hearing that you aren't.
They didn't want it going out with the agency's name on it.
"Fine," Lester said. "I won't put TDOT on it. I'll handle it."
He went home, opened his own email, and wrote it as a civilian. And he did the thing that made the difference, which was not to announce that anyone was wrong:
Then the two numbers fifty percent apart, and the observation that the criterion is already enforced in the other direction — the urban scenario correctly refuses a basin below the threshold — so the mechanism exists and works. It just runs one way.
Sent from the kitchen table. Forgive any typos.
The reply came back inside a day, and it was not a yes.
Ray Coburn ran the StreamStats web application out of the Water Mission Area's Enterprise Technology Office — the small name in the footer of the app that most users never click. He didn't stall and he didn't get defensive. He explained: the limits were set correctly, straight off the published table. Imperviousness wasn't a predictor in those three regions, so no bounds were entered, because there weren't any to enter.
That is a good answer. It is internally consistent, it is procedurally correct, and it closes the ticket.
Lester read it twice and understood that he had been told no by someone being careful, which is the hardest kind of no to move.
So he stopped arguing about Tennessee and went looking for somebody who had already done it.
He found it one state over. A 2023 rural peak-flow regression in South Carolina, in the same national application, carrying a cap on imperviousness of zero to ten — on a rural equation, where imperviousness is not a predictor either. Then he did the part that mattered: he swept the value from zero to nine point nine and watched the computed discharge. It never moved. Not once, not at any recurrence interval.
Which proved the cap wasn't doing arithmetic. It was doing applicability. Same threshold, neighboring state, established practice, three years old.
He sent that.
Two weeks after the first email, the answer came back changed.
Both authors had been consulted. Regions one, two and three would carry an out-of-range warning above ten percent. It would ship in the migration the following morning. One of the authors had already started drafting guidance text for the state page.
The next day, a second note nobody asked for: the author had come back again on his own and added a constraint to Hydrologic Area 4 as well — the region Lester had explicitly written that he wasn't asking them to change.
And then, because Coburn figured he'd appreciate the weeds, the thing Lester wasn't entitled to and got anyway. During quality control on the urban study they had discovered the impervious layer feeding the urban regressions had gone through a slightly different set of processing steps than the one feeding the rural regressions. Same source data. Two chains. Rather than reconcile them after the fact, they renamed the urban one alternate and kept both.
Which explained the two impervious numbers sitting on every report Lester had run for a week, a couple of points apart, that he'd assumed was rounding and had been quietly bothered by ever since.
Fourteen days. Two rounds of code changes across all four regions of a national federal application, guidance text in the works, and an internal QA/QC explanation volunteered to a civilian.
There was no entity to hate. That was the strangest part. He'd built a whole vocabulary for institutions that let wrong numbers stay wrong, and this one had turned around in two weeks and then come back a second time on its own.
He'd told them he'd verify it once it landed, so at six in the morning he did.
It was live and it fired. All four rural regions carried the new alternate-imperviousness limit; region four showed its calibration range directly above it, the two constraints side by side in one table — a cleaner answer than the one he'd come in asking for.
He walked Nonconnah Creek in Memphis through the imperviousness bands. Three points, four and a half square miles up to fifty-seven, all of them upstream of the gage and none of them the gage itself — because the urban report already names that gage as an exception, and a finding that lands on it answers itself.
At every one of the three, the urban equations came in twenty-eight to thirty-six percent below the rural ones.
Then he ran the other direction for contrast: Richland Creek in Nashville, at a gage with continuous record back to 1964. Compared both equation sets against the published flood-frequency values. Rural was off by an average of twenty percent across six recurrence intervals. Urban was off by five and a half, and closer at every single one.
And the last thing, the one he almost didn't notice. Two of the three Nonconnah basins tripped the new warning — and both had ordinary imperviousness values under ten percent. Nine point four four. Eight point seven six. The alternate layer runs a point or two hotter every time, so the warning reaches basins the report's own "only one gage" sentence would never lead a reader to expect.
Here is what that cost him, and it is the part he wrote down.
For a week his best material had been an elegant thing: one gage in west Tennessee is simultaneously the only basin above the threshold and the single documented case where the urban equations come in lower, and the sentence directing you across that line sits one paragraph below it. Two reports, side by side. Nobody had put them together. He was proud of it.
Three live runs made it irrelevant. When the tool itself returns a lower urban number on three separate basins, it no longer matters which gage a published figure belongs to. The measurement doesn't need the argument.
He cut it.
A thing you noticed is worth less than a thing you measured,
and if you can't tell the difference it's because you're the one who noticed it.
Kill your best paragraph when a worse one does the job without you.
Earl asked how it went. Lester said they fixed it in two weeks, then fixed a second thing he'd specifically told them not to fix.
"So you won."
"Nobody lost. That's the part I can't get anybody at the office to hear. The paper was right. The table was right. The people who entered the limits entered them right. There just wasn't anything to enter, and a blank came out looking like a pass."
"And the first guy told you no."
"He told me no for a good reason. That's different from telling me no. If somebody gives you a careful no, don't push harder on the same door — go find the room where it's already been done."
Earl considered his drink. The fountain was foamy again. It's always foamy again.
"So what do you tell the next guy?"
Lester pulled the pad over, and under the region numbers and the two impervious values a couple of points apart, he wrote it clean:
Read the report, not the application.
A range that runs zero to one hundred isn't a range —
it's the shape of a question nobody was asked.
Bring a precedent, not a complaint.
Measure the thing instead of proving it.
And when the institution moves: say so, out loud, by name.
The ones who fix it in fourteen days are the reason it's worth sending at all.
Every one of these starts the same way: a number that doesn't sit right, and a man who writes it down before he decides what it means. The pad is the throughline — the surge tower, the SUPERSHEET, the FEMA letter sent from a parking lot, and now a blank field that read as a passing check. One page holds all of it.
Where this one goes next