https://github.com/lsst/drp_pipe/blob/main/pipelines/HSC/DRP-RC2.yaml is currently our official HSC pipeline, but its step2 subset cheats in that it is a bag of tasks that are run globally (without dataquery constraints) on our 3-tract RC2 validation runs. It contains tasks that need to be global (FGCM), tract-level (jointcal/isolatedStarAssociation) and visit-level (skyCorr, consolidateVisitSummary, finalizeCharacterization). I say that it cheats because this is only doable because it's 3-tracts.

DP0.2 processing showed us that at 100's of sq deg, we can't submit 150-tracts all at once. The execution team's solution was to submit groups of visits and groups of tracts. As a courtesy. I propose splitting step2 for large production runs into it's visit-level and tract-level components:

step2a: # visit-Level

      - consolidatePreSourceTable

      - consolidateVisitSummary

      - skyCorr

step2b: # tract-Level

      - isolatedStarAssociation

      - jointcal

step2c: # global (can be run any time after 2a)

      - fgcmBuildStarsTable

      - fgcmFitCycle

      - fgcmOutputProducts

step2d:  # visit-Level

      - finalizeCharacterization

      - writeRecalibratedSourceTable

      - transformSourceTable

      - consolidateSourceTable

this has the added benefit of communicating with NAOJ the order of imageCharacterization tasks that need to be run before coaddition.