How noisy is normal, and what a threshold cannot tell you.

Here are twelve consecutive readings of the tracked satellite count, taken from a stretch of the flight where nothing notable is happening. Work out the mean, then the spread. The arithmetic is deliberately small enough to do without a calculator, and doing it once by hand is the difference between sigma being a quantity and sigma being a word.

Those twelve values were not special. They were the first twelve of a great many, and the calculation you just did by hand is the calculation that runs on all of them. Watch what happens to your answer as the sample count grows. Nothing about the method changes.

Same calculation. More data.

The estimate from twelve samples is not wrong. It is uncertain, and by an amount you can also compute: the standard error of a sigma estimate falls roughly as one over the square root of the sample count, which is why twelve samples move it around by tenths and eight hundred barely move it at all. Every threshold you set from now on inherits the uncertainty of the sigma you set it from.

A time series and a histogram of the same channel are not two pieces of information. They are one set of numbers, sorted two different ways. Drag across either one and watch the other.

Three sigma either side of the mean contains most of what this instrument does when it is behaving. That is an observation about this record, not a law: the interval only covers the fraction people quote if the quantity is distributed the way people assume, and a satellite count is an integer that cannot go below zero. Check the histogram before you trust the rule.

Now set a line. Anything past it gets flagged for a human to look at. Drag the handle in the plot itself, because a threshold that is typed into a box somewhere else is a threshold nobody can see against the data it applies to.

A threshold detects a deviation. It does not explain one. Move the line down and you catch the event sooner and cry wolf in quiet air. Move it up and the quiet air is clean and you find out later. Neither position tells you what happened, and there is no setting that does.

Switch the channel and watch how differently the same rule behaves. Bus voltage is continuous, so the flagged count moves smoothly as you drag. The satellite count is an integer with four values in it, so the count does not move at all and then jumps by hundreds: there is no threshold between those two states, because there are no measurements between them. Choosing which channel to watch is a bigger decision than choosing where to put the line.

Here is the position channel on its own, which is how most people meet a flight log for the first time. Look at it, and say what you think happened.

What you computed, what you set, what you first thought and what you thought afterwards. The last two are recorded separately and both are kept, because a position that changed when evidence arrived is the thing this mission is actually assessing.