Dealing with identical instances when joining the knowledge graph of multiple experiments

[joergfunger]

In todays meetings of the knowledge graph group, there was a discussion on how to deal with multiple instances that are presumably identical, but not in reality. In the example, we have a compression test and a Youngs modulus tests that are performed for the same mixture. In the current approach, we first generate an instance of the mixture based on the id, create an instance of the compression test and then add that to the (empty) graph. Now we want to add another subsequent experiment (Youngs modulus test). In theory, we could create another instance of the mixture (with the same ID) and an instance of the compression test and then join the two graphs (or upload the new triples into the triple store). If all information is really identical (with the same id), the instances are just replaced. However, if there are (even only slightly) differences, e.g. because the raw excel sheet of the second mixture instance was slightly modified (e.g. the name of the additives was slightly changed), we might end up with an inconsistent dataset (e.g. if we have a list of ingredients, we would then have the additive twice). How do we handle that situation? Do we distinguish between using an existing object and double check with a list of metadata to verify if the data in the metadata.yaml and the graph (using a query) is identical, otherwise return an error? Or do we add an instance of the second mix and verify, if the number of triples hasn't changed (essentially the same as before, but then there is no manual check required)? It would be quite interesting to discuss that not only from a technical implementation, but also from how we are going to use it. Is a mix (design) a constant (a single instance) or is it different for all tests (for each test another instance) that are performed. This holds true for many similar entities (e.g. a mix contains a list of ingredients such as specific aggregates, additives, water) - I guess we do not have to specify the density of water each time we add another mix. That's why I would hope to involve both @firmao and @PoNeYvIf from a technical point of view, but a users perspective on that by @eriktamsen @JulietteWinkler @StephanPirskawetz would be also very valuable.

[eriktamsen]

I am wondering, is there value in defining spefic concrete recipies (mix designs) as a kind of mix template which are unique and then create seperate instances for each actual mix?
I think this would depend a bit if where are a limited number of recipies which are used alot or if each new mix changes the recipe slightly. @JulietteWinkler, @StephanPirskawetz, what would you say would be more practical?