The same 2,745 results, 520 kilometers up

The registration went in on 7 October. ESA confirmed it on the 14th as experiment 4051 and sent a root filesystem, not a container: Alpine 3.21 on 32-bit ARM, entered with chroot under QEMU. The rules were short. One tarball, every file owned by a user called exp, an executable called run in the foreground, output in a folder the runner collects, and five minutes or less. The package went on the 15th. ESA had run it in its sandbox within the hour.

The sandbox is a virtual machine with two Cortex-A53 cores. The flight computer is an Altera Cyclone V with two Cortex-A9 cores at 800 MHz, on a Linux kernel built in 2021. TU Graz ran the package on the FlatSat on the 16th at 05:20 UTC: 45 ticks a second, six times slower than the sandbox and three times slower than the laptop. The shrinking quorum delivered 2,745 correct results over 5,250 ticks and none wrong. Fixed TMR delivered 239, then a detected disagreement on every tick.

The prediction went to ESA that afternoon, in writing: the same events on the same ticks, about 2,700 correct results for the shrinking quorum, 239 for fixed TMR, zero wrong, about 225 seconds, and any real bit flip in the 64 MB sensor block as one extra line in the log.

On 18 October at 09:45:37 UTC, systemd on a host called sepp1 started Experiment 4051. It ended 227 seconds later. 2,745 correct over 5,250 ticks, zero wrong. 239, then disagreement to tick 4,500. The event log is the FlatSat’s, tick for tick. The extra line never came. No real upset in 227 seconds, so the run calibrates nothing about radiation, and the dusk page now says so under its byline: it checks the program, not the model, and 1.34x stays a simulation number.

ESA reported that the run had been completed successfully. Nothing else about the platform was reported. Both runs went three seconds past their 110 second cap, as they had on the FlatSat two days earlier. That was the one defect, and it was already on the ground.