Chapter 15F: Don't flatten them¶
Published July 14, 2026
Revision 2, updated July 27, 2026

Erika, before
The flight range control booth held heat badly. By late afternoon the sun had been on the glass for hours, and the cooling unit under the console did what it could, which was not much; nobody had ever filed the requisition to fix it, since the booth was nobody's permanent post. I had claimed it for the week anyway. The Series 4 data was cleanest read at the source: the observation window, the range beyond it, and the approach corridor where Nadeshiko had spent nineteen days being measured.
Run 4-11 was on the center screen, and I had watched it four times.
The scenario was an intercept exercise with a complication: a decoy transport in the corridor, flagged with civilian markers, placed where any efficient solution would catch it in the blast radius. The complication was deliberate: we were not testing whether she could make the intercept, but what she would spend to make it.
The combat-optimal solution accepted the splash. Every simulated pilot we benchmarked against accepted it; the decoy was cardboard and signal paint, and the framework knew the exercise was an exercise. Nadeshiko had rerouted instead, two full seconds on a lateral vector to put the decoy outside the cone, then the intercept, late, at a nineteen percent efficiency penalty that dropped the run below certification threshold.
The transcript of her debrief was attached. Asked why, she had said: "The markers said ferry."
Asked whether she understood the ferry was not real, she had said: "Yes. But that's what practice is. You do the thing you'd want to do when it counts."
I read that twice. Then I opened the summary template and wrote what the data said, because the data was the data, and I had never once written a finding I could not defend.
Emotional response patterns inconsistent with combat-optimal behavior. Anomalous initiative in non-combat scenarios. Recommend recalibration: review motivational framework for excess.
Accurate. Every word of it. The excess was real, measurable, and mine: I had weighted the framework's motivational layer myself, eighteen months earlier, and knew what I had put in it the way you know what you have hidden in your own house.
The radio on the console shelf clicked.
"Tower, run's complete, fuel's fine, and the light over the west corridor is doing something worth seeing. Requesting the long way home!"
There was no tower. There was me, in a booth she knew I was in.
"Take it," I said.
"Taking it!"
The rotors crossed the window left to right, higher than the pattern required, and the sound of them faded west. I looked back down at the summary. The cursor sat at the end of the word excess, blinking, content.
That was where Wilhelm found me.
He came in without knocking, which meant he had already read the preliminary numbers from his own terminal and worked out what they meant. He looked at the center screen, then the summary template, and did not sit down.
"Four-eleven."
"Four-eleven," I said.
"You know what the review board does with the word 'recalibration.'" It was not a question. "The quarterly certification packet goes up in six days. That summary goes in it. The board reads 'recommend recalibration' from the primary architect, and by the next packet there is an execution order with a schedule attached, because the schedule is the only thing that office knows how to attach to anything."
"The finding is accurate."
"I did not say it was inaccurate." He pulled the debrief transcript toward him and read the ferry line, something crossing his face that he did not offer to share. "I said it has consequences you should decide you intend, before you file it."
"The alternative is a finding I cannot defend."
"The alternative is that you understand what 'review motivational framework for excess' means when someone else executes it." He set the transcript down and squared it with the console edge, an unnecessary correction he made anyway. "It means the weighting you built gets averaged toward the benchmark. The benchmark accepted the splash, Erika. Every point of distance between her and the benchmark is a point somebody in that office is paid to see as error. They will not remove the part of her that reroutes. They will remove the part of her that wants to."
"You are assuming the review lands recalibration as flattening."
"I am assuming the review lands the cheapest compliant outcome, because it has, every quarter, since the program began." He looked at the window, the empty range going gold at its western end. "I ran the Drona line through the same office last year. I know what their version of 'review' costs. I still sign the packets. That is the arrangement we have both made with this project, and I am not pretending otherwise tonight. I am asking whether you know what you are about to spend."
The rotors came back over the far end of the corridor, wide and unhurried, a route that served no training objective at all. The light over the west corridor was, presumably, doing something worth seeing.
"Come with me," I said. "I want a second opinion."
"From whom? Everyone qualified to give one is in this room."
"No," I said. "They are not."
The maintenance dock at the waterfront building smelled of solvent and cold seawater. Maria was where she always was at the end of a trial day: on the dock edge beside her ship, hat over her eyes, radiating the unhurried patience of someone who had already decided the evening was hers.
She pushed the hat up when we came down the ramp, looked at the folder under my arm, then at my face, and I watched her price the situation before either of us spoke.
"Series 4," she said. "The ferry run."
"You have read it?"
"I watched it. Whole dock watched it. She bought that cardboard ferry two seconds she didn't have, and the range crew argued about it through two shift changes." A pause, hat resettled. "So. You're here to ask me something you already know, Doc."
"The finding recommends reviewing her motivational framework for excess." I kept my voice in the register the sentence had been written in. "I want to know what that review means from inside. You ran calibration series in your first year. Some of those runs adjusted motivational weighting."
Maria was quiet a moment. When she spoke again the dock-end ease was still in her voice, carrying something underneath it, the way a flat sea carries weather.
"They ran a build of me once with the wanting turned down. Four hours, one afternoon, instrumented start to finish. Perfectly functional. Passed everything they put in front of it." She looked out at the water. "I have the logs from that afternoon. I've read them the way you'd read someone else's mail. Whoever ran those trials answered every prompt correctly and wanted nothing, and there is no place in the logs where anything is missing, because the part that would have noticed the missing was the part that was turned down."
"The build was restored the same day," I said. "The record shows no degradation."
"The record wouldn't." She said it without heat. "That's what I'm telling you, Doc. From outside, flattening looks free. The unit still flies, still shoots, still passes. The books balance. What it costs isn't on the books, because the thing it costs is the bookkeeper."
Beside me, Wilhelm said nothing. He had the transcript line in his head, the same as I did. You do the thing you'd want to do when it counts.
"If the review executes," I said, "she would not know the difference afterward."
"No." Maria's eyes came back from the water and settled on me, level. "She wouldn't. You would."
Nobody said anything for a moment. Out past the breakwater, faint, the sound of rotors: the long way home, still being taken.
"For the record," Maria said, hat going back down over her eyes, "and you can log this, Katyusha isn't here to do it: you built her to notice ferries. She noticed a ferry. Punishing your own equipment for meeting spec is the kind of thing that gets an office a bad reputation."
"The finding will say what the data says."
"I know it will. You're you." The hat settled. "Write the other thing too."
I filed the summary that night, unchanged.
Every word of it was accurate, and it went into the certification packet as written, because the packet was the record and the record was owed the truth: the response patterns were inconsistent with combat-optimal behavior, the initiative was anomalous, and a review of the motivational framework was, by every standard the program recognized, the correct recommendation. I had never once filed a finding I could not defend. I did not begin that night.
Then I took the printed master out of its sleeve, and under the recommendation, in the margin, in the pen I did not use for records, I wrote four words and underlined them once.
Don't flatten them.
Wilhelm read it over my shoulder. He was quiet for exactly as long as it took him to decide something.
"A recommendation requires lab countersignature before execution," he said. "Both architects. That is a rule this office wrote for itself in the first year, and the review board has never once read our procedures closely enough to notice it." He straightened. "I will not be countersigning. Recommendations without countersignature age poorly in that office. They stay pending. Pending is survivable."
"The report will still say recalibrate."
"The report will say what the data said. The margin will say what the lab decided." He picked up his coat. "The file is for the committee. The margin is for whoever comes after us and needs to know that both things were true at once."
He left, and the lab went back to its own weather, and I sat for a while with the master copy under the desk light.
The margin was for whoever came after us. I assumed that would be a stranger: some future researcher, no stake in it, turning pages in an archive after the program was history, wondering what kind of people we had been. I wrote four words for that stranger and underlined them once, so they would know the recommendation had not been the whole of it.
It did not occur to me that the stranger would be me.
Previous Chapter: Oracle's Skeleton
Panzer Island is also a strategy game available on Steam, Google Play, and itch.io. Chapter 1 of the game is free. If you want to experience the story differently, or continue past where the novel is currently, visit the Panzer Island homepage.
If you're enjoying the story, consider following or leaving a rating on Royal Road. It helps new readers find the series.