Conformance checking?

[TathagataChakraborti]

The P4PP setup reminded me of the conformance checking problem in BPM.

As an analogy, playing the student's plan forward in the reference model is a type of "token replay", while the trace alignment view has this type of model move versus log move (>>'s) versus synchronous move which is what the feedback of P4PP reminded me of during the demo. Not sure what my point is here, just an observation. 🤷

Of course, conformance checking also has planning compilations like this, maybe its possible to write a "Automated Grading as Conformance Checking" paper. 🤔

[haz]

Doesn't this align (aha!) closely with the plan rec as planning work? E.g., this. The main difference I see is that it's not a single trace, but rather every trace captured by a domain. It simultaneously looks across them all.

[TathagataChakraborti]

Yes indeed! Not that but this, coz we need to drop observations to align.

[haz]

I go back to the challenge of this all being flipped. If you want to explore the space of modifications to make two theories align, then you are trying to navigate a solvable space until it is unsolvable. If it was reverse (e.g., what must be change so these theories misalign), then we could package it all up as a big planning problem. Not sure how to promote unplannability as planning :P. Well, there is a way (modeling the successor function in the PDDL itself), but it'd never actually solve.