Skip to topic | Skip to bottom

Provenance Challenge

Challenge
Challenge.ModelDiscussionRelationships

Start of topic | Skip to actions

Model Discussion: Relationships

Does a causal relationship between E and C in the model mean that everything or something of E was caused by everything or something of C? Ideally, I would expect "everything" in both cases, otherwise the relationship is ambiguous, but I don't know if this requirement can be fulfilled in practice.

In Section 3, you state that the "used edges out of process cons mean that both 3 and 7 were required for the process cons to take place", which seems fine. However, you then seem to assume that this AND relationship applies to all edges from process to artifact, and I do not see why the generalisation works. Could you not model that "cons would not have occurred as it did if it wasn't for the 3"? Considering construction of a model, it may be that one contributor to the model knows about the 3, another about the 7, and so the AND relationship is not in the model. This does not invalidate the value of the separate assertions about the 3 and 7, nor is it a substitute for the assertion that cons required both 3 AND 7. As mentioned by Jim in ModelDiscussionTime (with regard to streams and arrays), it may be useful to model artifacts that are compositions of other artifacts. Is "3 AND 7" not just a composition of 3 and 7?

I'm also not yet convinced about the parallel generalisation, given in Constraint 1. As a (perhaps not great) example, consider the sound produced when two pool balls collide. The processes of both balls moving towards each other are independent and can have independent triggers, but both (AND) are required to generate the artefact (the sound). This AND relationship may be good for documenting the provenance of an exception raised by one process trying to access a resource another had a lock on (though I agree it can be modelled other ways).

If I am allowed approximated relationships for practical purposes (which seems reasonable though the requirement behind it is not yet explicit) then the approximating relationships allowed in Section 2, triggeredBy and derivedFrom, may not be enough. I may want to relate an artifact to a causing process that did not generate that artifact. For instance, I am sat at a desk in King's College London because I went through the PhD? process at Warwick. However, while it was a necessary pre-requisite, the process did not put me in the state of sitting here: much else happened inbetween.

The fact that derivedFrom relationship implies an intervening process (Section 5), assumes that everything is constant in the absence of a process to change it. This seems a reasonable assumption, but may be good to make explicit?

Does one process triggering another always require an intervening artefact (Section 5)? This does not seem intuitive to me, and more explanation would help.

-- SimonMiles - 17 Aug 2007

I think this discussion overlaps with the agents one - going backwards a bit, I argued that there might be economic agents, intentional agents, and executable agents and thus the Warwick agent catalyzing your hiring process rather than your PhD? being an input to the process would be the way to talk about it (yes there were lots of states and process executions involved and there was probably some physical artifact that documents your PhD?, but using the agent route says that we don't want to talk about that detail).

I think some of the other points you raise are covered in the model - different accounts can report a 3 input or a 7 input to a process which we don't make any statement about (do they disagree or did they each see a part of the process?). By putting a role on the inputs we're trying a bit to leak info about what 3 and 7 inputs mean without trying to classify the range of 'degrees of causality' that occurs. We had an example of the max function - it's value depends on both values showing up but in some way only one value has a direct connect to the output.

-- JimMyers - 18 Aug 2007

> I think this discussion overlaps with the agents one - going backwards a bit, I argued that there might be economic
> agents, intentional agents, and executable agents and thus the Warwick agent catalyzing your hiring process rather than
> your PhD? being an input to the process would be the way to talk about it (yes there were lots of states and process
> executions involved and there was probably some physical artifact that documents your PhD?, but using the agent route
> says that we don't want to talk about that detail).

I don't really understand your suggestion. If I wanted to model the two facts I know, fully admitting it to be an approximation of the full details: "me sitting here was caused (but not generated) by my PhD? process", "my PhD? process was caused (triggered) by my undergraduate degree process". I'm not sure if the conclusion is that I am not allowed to include the former approximation in the model, or if I use agents somehow to bridge the gap between process and artifact. In the current model, agents cannot directly cause artifacts, so I'm not sure what would be the replacement cause of "me sitting here".

-- SimonMiles - 20 Aug 2007

> I think some of the other points you raise are covered in the model - different accounts can report a 3 input or a 7 input to a process which we don't make any statement about (do they disagree or did they each see a part of the process?). By putting a role on the inputs we're trying a bit to leak info about what 3 and 7 inputs mean without trying to classify the range of 'degrees of causality' that occurs. We had an example of the max function - it's value depends on both values showing up but in some way only one value has a direct connect to the output.

OK, I understand. Perhaps I misunderstood the explanation in Section 3: I took it to be making a statement about a difference between used and generatedBy relationships, when it is really just saying that relationships in separate accounts should be treated as independent ("OR") while those in the same account should be treated as mutually dependent ("AND"). This makes more sense, though I still don't understand why Constraint 1 holds.

-- SimonMiles - 22 Aug 2007
to top


You are here: Challenge > OPM > ModelWorkshop > ModelDiscussionRelationships

to top

Copyright © 1999-2012 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback