Coherent bodies of work that are not prerequisites for anything. Do them because they are worth doing, not because a track told you to.
The tracks are the main paths: ordered, cumulative, and designed so each one ends in a capability you can describe.
A side quest is different. It is a coherent body of work that does not sit on any track’s critical path. Finishing one does not put you further along a track, and you do not need to have finished a track to start one. It is worth doing on its own terms, and in at least one case it is worth more to a lab than most of what surrounds it.
They are listed separately rather than folded into a track for a reason. Material buried inside a sequence gets read as “step 7 of 12,” which is exactly the wrong frame for something you might reasonably spend a month on and never touch the rest.
Two. One is built; the other is named because the gap is real, not because it is ready. Naming a category and then padding it would defeat the point of having one.
Correcting automated reconstruction errors, treated properly: as an allocation problem under a fixed budget rather than as cleanup that ends when the data looks right. Around 17,000 words of reference material, a full technical unit, five worked scenarios, and a portfolio artifact a lab can read.
A drill on native-resolution EM crops with labels derived from released segmentations, scoring into a confusion matrix and reporting accuracy within your own high-confidence calls. Units 05–07 describe how to build one and argue for why it matters; the corpus does not exist. Listed here so the gap is visible.