South Test Court had no place to hide a good first result.
The alternate branch from the annex was here on a separate proof path because Tarin had paid to make it separate.
Selka ran one finger down the procedure.
"You declared the inputs, work surface, preparation state, branch target, counter condition, and acceptance threshold."
"Yes."
"And this is the complete public procedure?"
Tarin felt the old urge to answer faster than the question required.
"Complete enough to reproduce the result."
Selka looked up.
The words were already bad by the time he heard them.
"That is what we are testing," she said.
The piece under glass had passed Tarin’s own run that morning. He had seated the Anchor, waited for the preparation to settle, laid the counter line, then committed the branch. The court indicator went green under independent observation. The page, he had thought, contained everything that mattered.
Now a court replicator in a gray tab read that page once and began.
Tarin knew the replicator only by the gray court tab on their sleeve. Selka had selected them without asking him.
Their handling was clean: seals opened at the rail, marks matched intake, fixture calibration already signed. No stupid mistake arrived to rescue Tarin.
The operator placed the Anchor. Checked alignment. Set the disclosed counter-preparation line. Then seated the Donor.
Tarin's thumbnail pressed into the side of his index finger.
Something about the order bothered him—not according to the page, according to his hands.
Wrong according to his hands.
He opened his mouth.
Selka said, without turning, "Do not coach the replicate."
"I wasn't—"
She looked at him.
He shut his mouth.
The operator continued.
The branch took.
For three seconds Tarin thought the clean replicate would prove the method belonged to the page more than to him. Then the counter fixture flashed.
The prepared control line sheared where it should have held its intended tolerance. The output did not explode or collapse. It simply moved outside the accepted behavior band, a small technical failure made enormous by the red indicator above the bench.
This story is posted elsewhere by the author. Help them out by reading the authentic version.
The operator stopped immediately.
Selka raised one hand. No one touched the piece.
Tarin stared at the failed output. Almost good was still outside the band.
The operator stepped back from the bench. "Procedure completed as written."
Selka checked the timing strip, then the fixture log.
"Any deviation you observed?"
Tarin heard the trap in the question and answered anyway.
"Not a deviation."
"Then what?"
He looked from the failed piece to the successful one under glass.
The same declared conditions had produced two different results. Tarin found the missing distinction before he found an excuse.
"The setup order."
Selka's face did not change.
Tarin moved one hand toward the clipped procedure, then stopped before crossing the rail.
"I seated the Anchor first. Then I let the preparation settle before I drew the counter line. They set the counter line before the Donor was seated."
The replicator looked at the page.
"The procedure lists all three steps."
"It doesn't fix the interval."
"No," Selka said. "It does not."
He had written the sequence as components, not as a timed state change. His hands had carried a rule the procedure did not.
"We do not know yet that order is causal," he said.
"Correct," Selka replied.
The relief lasted less than a second as she turned the result sheet toward him.
INDEPENDENT REPLICATE: FAIL
DECLARED CONDITIONS: SATISFIED
ACCEPTANCE BAND: NOT MET
Below it, she wrote by hand:
CANDIDATE VARIABLE — SETUP ORDER / INTERVAL NOT SUFFICIENTLY SPECIFIED
He studied the word fail.
"You could mark it inconclusive pending the timing check."
Selka capped her pen with a click. "The cause is inconclusive. The replicate is not."
The sponsor status changed from REPLICATION PENDING to INDEPENDENT REPLICATION FAILED — CORRECTION REQUIRED.
"My run passed," Tarin said.
"Yes."
"Under the declared setup."
"Under one implementation permitted by the declared setup."
Selka pulled the public record form from beneath the test sheet.
"I am retaining both pieces and both observation logs. Your successful run remains a successful run. The independent failure remains a failure."
"I know."
"Do you?"
"I'm not asking you to throw it out."
"You asked me to relabel it thirty seconds ago."
The red indicator above the failed bench held his attention. "I asked before I saw the gap clearly."
Selka let that sit between them.
The replicator had already moved away from the bench. They did not look embarrassed. They had done exactly what the test required and produced information Tarin had failed to put on the page.
That made his own embarrassment harder to place anywhere useful.
A sponsor clerk approached with the status sheet. "Until corrected replication, the alternate branch remains non-deliverable under the amended scope. Any further proof run requires a revised disclosed procedure and fresh observation. Existing commercial acceptance is unchanged."
The client had lost nothing; the alternate had never been authorized for delivery. Tarin had lost the comforting claim that one successful public proof was enough.
"Fee treatment?" Tarin asked.
The clerk checked the sheet. "New proof window is applicant-side. No credit for the failed replicate."
Selka slid the publication acknowledgment toward him. "If you want my name on the retest, this failure goes out with the correction note now."
The acknowledgment was blunt: author-run proof passed; independent replicate failed; cause unproven; setup order and timing were candidate variables; no production claim. Tarin signed it. Selka signed beneath him.
The clerk stamped the failed record and carried copies to the public file and sponsor desk.
"For the next run," he said, "I want the order explicit. Anchor seated first. Counter preparation after the hold settles. Interval bounded, not by feel."
Selka did not write it down yet.
"That is a correction hypothesis."
"Yes."
"Not a result."
He pressed his thumbnail into his finger once, then stopped.
"Yes."
She finally pulled a blank correction sheet from the stack.
Across the top she wrote:
RETEST REQUIRED — ORDER / INTERVAL TO BE SPECIFIED BEFORE EXECUTION
Then she clipped it to the failed replicate record instead of the successful one.
His attention settled on the two pieces side by side: one accepted, one red-tagged. Neither result erased the other.
Selka closed the public file tray.
The red-tagged replicate stayed on the bench beside the blank correction sheet, waiting for a procedure that did not require Tarin's hands to know what the page had forgotten.