‐abstract decomposition axes
 ‐type
  ‐determined by the set of properties
   ‐the properties must have a fixed interpretation
   ‐a set of typed columns is not enough, because the atomic types are
    ambiguous representations of underlying property values
  ‐underlying structure is a tree
   ‐hierarchy via
    ‐join/set union
    ‐meet, but is always empty
 ‐logic
  ‐disjunctive vs. conjunctive
  ‐relational is in conjunctive extensive form
   ‐essentially a model satisfying the theory
   ‐is there such a thing as disjunctive extensive form?
  ‐negation
   ‐relational uses the closed world assumption
  ‐underlying structure is a lattice
   ‐hierarchy via
    ‐join
    ‐meet, but might be empty
 ‐normalization
  ‐underlying structure is a cartesian/set product
   ‐hierarchy via
    ‐set union
    ‐set intersection
 ‐partitioning/tiling/chunking
  ‐underlying structure is a (euclidean) space/manifold
   ‐hierarchy via
    ‐set union
    ‐set intersection
 ‐references/interdependencies/links
  ‐underlying structure is a graph
   ‐hierarchy via
    ‐transitive closure
 ‐coarse graining/cubing/summaries/derived data
  ‐underlying structure is a lattice
   ‐hierarchy via
    ‐join
    ‐meet
   ‐dwarves are essentially functions from lattice relations represented as
    minimum covering graphs to subtotals
   ‐is this an instance of a part‐whole/aggregation relationship?
   ‐how does this fit in with mipmap/ripmap/multiresolution/wavelet
    decomposition?
 ‐versions
  ‐underlying structure is a tree
   ‐hierarchy via
    ‐join
    ‐meet
   ‐how does this fit in with time?
 ‐time/history
  ‐underlying structure a linear order
   ‐hierarchy via
    ‐join
  ‐how does this fit in with versioning? particularly linear chains within a
   version tree?