RFD 1065: Taskweft domain schema in essential tuple normal form

Problem

RFD 1064 asks Claude to inspect 15,000 dataset rows and write one problem.ex per row. Many rows share the same hair color, eye color, and pose, worded in different ways. A fixed trait list misses new values, and free text per row repeats the same fact many times.

Decision

Design the domain.ex/problem.ex schema per essential tuple normal form (ETNF), from Darwen, Date, and Fagin (ICDT 2012). See DETAILS.md for the definition and its worked example. Three rules follow.

1. @variables holds a trait map, keyed by trait name, one value per character, matching the have/handle/loaded pattern in RFD 1044’s domain.ex. The key is a superkey, so the map sits in BCNF.
2. The trait taxonomy comes from the training data, not from preconceived categories. capabilities starts empty, and a domain action grows it as it resolves each trait value to a capability id. problem.ex stores a :ref to that id, never text.
3. The resolve step runs inside taskweft, through the HRR/ HRR.Cleanup library RFD 1021 already supplies, not through WeftspunStudio.FactVector outside it. A near-duplicate caption binds to the existing id, instead of creating a new one.

Rule 2 and rule 3 remove the redundancy together. A functional dependency anchored on the capability id, a superkey, blocks the redundant tuple that ETNF targets, with no fixed enum to maintain.

Related

RFD 1064 sets the domain/problem split. RFD 1037 gives the :ref/capabilities rules. RFD 1021 gives the HRR library. RFD 1044 gives the worked domain.ex this schema follows.

Details

The measurements and the retractions