If the customer just wants arc flash labels for electrical panels and switchboards only. What data do you need to get on the panels and switchboards? Do you still trace it all the way to the equipment? I’ve seen some reports that just end at the panels. What do you guys normally do?

I rarely did more work than the customer was willing to pay for.
But I always tried to get all of the source data required. For example, I would get motor nameplate data even though the customer did not plan to put a label on the local disconnect.

NFPA 70E & NFPA 70 tell you what has to be labeled, and OSHA refers to these standards. When we do a study, we try and track down all 3-pole circuits and include what they feed in the study.

NFPA 70E 130.5(H) - Electrical equipment such as switchboards, panelboards, industrial control panels, meter socket enclosures, and MCC's that are in other than dwelling units and that are likely to require examination, adjustment, servicing (troubleshooting) or maintenance while energized shall be marked with a label etc.

NFPA 70 (NEC) - 110.16 - Arc-Flash Hazard Warning.
(A) General. Electrical equipment, such as switchboards, switchgear, panelboards, industrial control panels, meter socket enclosures, and motor control centers, that is in other than dwelling units, and is likely to require examination, adjustment, servicing, or maintenance while energized, shall be field or factory marked to warn qualified persons of potential electric arc flash hazards. The marking shall meet the requirements in 110.21(B) and shall be located so as to be clearly visible to qualified persons before examination, adjustment, servicing, or maintenance of the equipment.

Unless the customer has requested it, I don't include all 3 phase loads in the model. I exclude them if there is nothing about that feeder that needs to be labeled or doesn't have significant fault contribution. Examples could include commercial kitchen equipment, or small 3 phase exhaust fans or an in-duct electric heating coil or electric water heater etc. when the customer says they don't label disconnects and their electrical safety program says live work on anything not labeled is strictly prohibited.

Yes, if a circuit feeds duct heaters, receptacles etc. we simply put a text block on the drawing stating that and no label is printed, but we do the due diligence during data collection to confirm that. Now granted most field disconnects will come up less than 1.2 cal however if someone will need to verify absence of voltage at a 480-volt RTU for example, they still need to understand the hazard to do that whether doing live work or not. I have had instances where a long cable run will drop the arcing fault current and it can be greater than 1.2 cal so for that reason we try and document as much field equipment as possible.

I don't want to get too far off topic but RE: packaged RTUs, I don't think you can accurately calculate the incident energy on many of those. They often have a molded case switch built into the unit in the panel next to the door or access panel or have a door panel type disconnect in the access door. When you open the access door the control section is pretty much an industrial control panel with relays, contactors, terminal blocks, fuse blocks, control transformer etc. with multiple conductor sizes and bus gaps and voltages. All you can do is calc to the line side of the MCS or disconnect and put a disclaimer in the report. That line side value might be way too low. See my YouTube video on industrial control panels. I think the actual potential incident energy in that case is incalculable. Too many variables.

Most of these prepackaged RTUs have all of the compressors and motors listed on the outside. It's a PITA, but I've typically added these loads up and used the sum for the calculations, since the motor loads tend to be the lion's share of the contribution. Would you say that's an inaccurate way of trying to capture the potential there?

Thanks for taking the time to reply. I understand that, my point was about calculating incident energy, not motor fault contribution. I do the same as you, I model them as one motor. There are so many parts in there (contactors, starters, relays, terminal blocks, different size wires, varying bus gaps etc) that, in my opinion, you cannot accurately calculate incident energy for something like that if you needed or wanted to label the unit. I don't see them as different than an industrial control panel.