Design the test

Find the boundary between the testable half and the untestable half

"Everything he said checks out," she said. "Individually. Each piece, on its own, is either confirmed or at minimum defensible. The magnetization is real. The conductivity is real, or close enough to real that I can't rule it out. What I can't find is the thing connecting them — the actual mechanism that would let a real magnetic rock genuinely receive a real current from four thousand miles away and turn it into the specific heat signature we measured. Nobody's ever built that bridge. Nobody's ever needed to. So there's no literature to check it against, which means there's nothing for me to falsify. It isn't that his claim survived testing. It's that half of it was never testable to begin with, and he built it right up against the boundary of the half that was, so close you can't tell where one stops and the other starts."

Design the test that could end the idea, and write it down before the chance comes

That was a real test. Not a rebuttal of anything he'd said — an actual, physical, dischargeable prediction, the kind of thing you could be wrong about, which was the only kind of thing worth proposing in the first place. If a solar event came through while they had instruments in the water and the heat signature didn't so much as flicker, the receiver idea was finished, cleanly, on its own terms. If it moved, he'd have earned something she couldn't argue him out of by finding one more seam, because there wouldn't be a seam left to find.

From her log that night:

I have designed the test that would settle it: does the signature track solar activity in real time, or does it sit fixed. This is dischargeable. It is also not runnable on this leg. I record it here so that when the opportunity arrives, it is not improvised in the moment but already waiting, fully formed, for the ocean to answer.

Make the instrument trustworthy before you use it

"No. But it's the next best thing, which is a recording of what actually happened, which you can read afterward as though it were real time, provided the clock inside it doesn't drift, which — " he considered this, already turning toward the parts locker as he spoke, the sentence trailing him rather than the other way around, " — which I can fix, actually. I can sync it to the ship's own clock right before deployment and check the drift rate against a known reference before we ever put it in the water. If I do that properly, when we pull it back up in three weeks, the timestamps will be trustworthy down to the second. Which means if a solar event comes through while it's sitting there, and the temperature moves, we'll be able to say exactly how fast it moved, and exactly how long after the event it started."

A few lines later:

"That would work," she said.

"I think it would work," he said, careful, because would and think it would were different sentences and he had never seen a reason to round one into the other.

And the first thing he checks when it comes back up (Chapter 19):

The lander came up clean, which was the first thing Bram checked, before he checked anything the housing had actually recorded — seals intact, no water ingress, the clock drift he'd calibrated so carefully three weeks ago holding to within four seconds across the entire deployment, which meant that whatever the internal log said, it could be trusted down to the second, exactly the way he'd promised Amaleto it would be.