I suppose this executable and the other one to run on grid are meant to setup tritonserver on the grid/interactively respectibely, but I wanted to understand better what's the scope of both scripts and perhaps I'd recommend adding a short description to both in case people want to use them.

this was actually introduced in the previous PR. These scripts are still under development so it's a bit premature to document them, but if needed we can add a brief description of the intended usage (but maybe not necessarily for this PR as it does not make any change to the scripts?).

Perhaps I'll clarify this on my own after checking the rest of the code, but I wanted to make sure what some of the labels refer to:
nuslhitsCryoE/W are the hits before nu-graph filtering
NuGraphCryoE/W I guess is a "global object/container" for all nu-graph related reconstruction
NCCSlicesCryoE/W is the "not clear cosmic" single slice considered as the neutrino candidate slice based on the trigger/optical flash
NGMultiSliceCryoE/W corresponds to nu-graph being applied to all non clear cosmic slices in a certain event
ngfilteredhitsCryoE/W corresponds to post-nu-graph-filtering hits
is this interpretation correct?
Perhaps, especially as far as NCCSlicesCryoE/W and NGMultiSliceCryoE/W are concerned at least, I'd add a comment to clarify/specify some details in the fhicl.

The proposed chain uses:

• NCCSlicesCryoW/E: a collection of slices that are not tagged as clear cosmic by Pandora;
• NGMultiSliceCryoW/E: this runs NuGraph2 on multiple slices (here, the ones from NCCSlicesCryoW/E) and provides objects and associations with NuGraph2 predictions;
• ngfilteredhitsCryoW/E: this filters the hits in each slice using the NuGraph2 filter score, and creates a new hit collection we then run Pandora over (the second-pass Pandora).

Thanks, we will add some comments and perhaps get rid of labels that we do not use anymore (nuslhitsCryoE/W and NuGraphCryoE/W were used to select the "triggering" slice and run NuGraph2 on a single slice).

Can you clarify how these two fhicl parameters set to false change the settings in pandora?

The bottom line is that, once PandoraPFA/larpandora PR #35 and PandoraPFA/LArContent PR #256 are merged in Pandora/LArSoft, there will be two fcl parameters (CollectHitPredictions and UseHitPredictions) to collect the information from NuGraph2 and inject it into Pandora.

More specifically, we added a new interface in Pandora (larpandora) to collect the hit-by-hit predictions from NuGraph2 at the second Pandora stage, after NuGraph2 was ran, and inject them into the hit object within Pandora (specifically, a LArCaloHit).
CollectHitPredictions steers whether we expect NuGraph2 information from the input we give to Pandora, and UseHitPredictions adds the information to the LArCaloHit.
This operation just adds more information that can be used in Pandora, but does not change the reconstruction (for that, we would need to actually use the information in any algorithm, and change the relevant XML).

I kept the parameters around just to remind us that there is this option, but they are commented out right now since the Pandora PRs are not yet merged. Specifically, false means that we do not inject the NuGraph2 information into Pandora (even if available): this can and will be changed once everything is ready.