A fault that left no line, and a chart that could not mark it
Fig. 1 is drawn from the eight files the satellite sent down. One of the four faults is missing from it, because an outvoted bit flip never reaches the log.
Eight files came down: two event logs, two summaries, two stdouts, the run info, and four lines of systemd journal. The figure on the research entry is drawn from them by a script, not from the text of the entry. It reads the two CSVs and the two summaries, and throws if the two runs do not share their fault ticks or if the stop tick is not one past the last useful tick. Two lanes, a 30 px bar each, and an axis break at tick 300, because every fault lands before tick 250 and the shrinking quorum then works alone to 5,250. The TMR lane is solid to 239 and hatched from 240 to 4,500: one good, one stuck, 4,261 disagreements detected, no results.
Three faults have a dashed line: the self-test at 20, the stuck replica at
120, the kill at 240. The bit flip in replica 2 at tick 60 has none. It was
outvoted the moment it happened and the voter wrote nothing, because nothing
changed. It is in the summary, faults=2 caught=2, and nowhere in the event
log. The caption says so, since a figure with three marks for four faults
reads as a mistake.
Version 0.4.0 writes outvoted,2 when a live replica dissents, and the run 2
schedule moves the flip to tick 30. The second figure will have its fourth
mark.