Marcus reviewed the data for the third time. The values were clean. Every sensor in range, every timestamp aligned, every reading consistent with the model they'd published in April.
Yolanda's report sat on the other side of his desk, handwritten, which was itself an act of aggression. She'd been in the field for eleven days. She said the eastern marsh was dying.
The data said the eastern marsh was stable. Not borderline stable — confidently, boringly stable. Dissolved oxygen, turbidity, pH, conductivity, chlorophyll. Five channels, eleven days, all green.
He called her.
"Your sensors are fine," she said, before he could ask. "I calibrated them myself on Tuesday."
"Then what are you reporting?"
"What I saw."
He waited.
"The herons left," she said. "All of them, over two days. The cattails at the northeast edge are the wrong green — too bright, like they're spending something. And there's a smell on the south bank I've never smelled there before. Not sulfur. Not rot. Something before rot."
"Those aren't measurements."
"No."
He looked at the printout. Five channels. All green.
"Yolanda, if I take this to the board—"
"Don't take it to the board. Go to the marsh."
He didn't go. He pulled up the satellite archive instead, ran a comparison between this season and the previous three. Found nothing. He ran a spectral analysis on the chlorophyll band. Found nothing. He cross-referenced the heron migration database for regional patterns. Found a plausible explanation involving construction noise from a highway project six miles north.
He typed up a paragraph about the highway project and attached it to her report as a response.
Three months later, the eastern marsh crashed. Dissolved oxygen dropped to zero in nine days. The kill was total — fish, invertebrates, the submerged vegetation. The cattails were the last to go.
The post-mortem found the cause: a slow anaerobic process in the sediment, invisible to surface sensors, building for months under the mud. By the time it breached the water column, the chemistry moved too fast for intervention.
Marcus read the post-mortem twice. Then he pulled up Yolanda's report again.
Something before rot.
She hadn't been wrong. She hadn't been using a better instrument. She'd been reading something his instruments weren't pointed at — not a different answer to the same question, but a different question entirely, one that didn't have a column in his spreadsheet.
He could add the column now. Heron departure timing. Cattail color shifts. Olfactory anomalies. He could formalize the whole thing, turn her eleven days into a protocol.
He started the document. Got two pages in. Stopped.
The protocol would catch this failure next time. It would not catch the next one. Because the next one would be whatever Yolanda noticed that she couldn't name yet — the thing she wouldn't know to put in a report until she'd stood in the mud long enough for it to bother her.
The instrument wasn't her senses. The instrument was her willingness to say something she couldn't defend.
He closed the document. Picked up the phone.
"Can you come in Tuesday? I want to talk about the southern watershed."
"What does the data say?"
"The data says it's fine."
There was a pause.
"I'll be there Monday," she said.