You've seen how to design a chain: focal steps, declared dependencies, unique output keys. Now Orbit shows you how they break in production when any of those principles slack.
On the right, five designs that on paper look fine. Each one fails for a different reason, and all the reasons appear in real systems. Your job: before revealing the diagnosis, look at the pattern and try to recognize which of the track's principles is being violated.
When you reveal all five, you'll have the mental catalog to audit your own chain before shipping it.
This catalog isn't exhaustive. But it covers 80% of typical failures in chains and agents. The other 20% (poorly calibrated eval problems, prompt drift over time) come in the track's final unit.