I spent five years writing rules. The most useful thing I learned is how a rulemaking record gets assembled, because that is where the defects live.
The standard is not the obstacle
Arbitrary-and-capricious review sounds deferential and is, on substance. Courts will not second-guess a technical judgment an agency is equipped to make.
They will, however, insist that the agency did what it said it did. That is a procedural question, and procedural questions have documentary answers.
Where the defects are
Three recur often enough to check first.
The standard that changed. A proposed rule floats one test; the final rule adopts another. If the shift was not fairly noticed, commenters never got to address the rule that actually issued.
The comment that went unanswered. Agencies must respond to significant comments. “Significant” is doing real work in that sentence, and a substantive comment disposed of in a sentence is a live issue.
The estimate that moved. Cost figures that change between proposal and final rule without an explanation in the record are the single most productive thing to look for.
Reading order
Read the docket before the rule. The rule tells you what the agency decided; the docket tells you how, and how is the reviewable part.