Dictionary design

One strict total order does all the work

NodeComparator: a strict total order over RDF terms
- Default-graph sentinel first, then BNode < URI < Literal; literals ordered by VALUE (2 < 10), with exact-term tie-breaks so distinct terms get distinct positions
An id IS a rank
- Sort the distinct terms of a section; a term’s 1-based position is its id. locate() = binary search; extract() = indexed read. No hash tables in the file
Three sections, two id spaces
- entities (URIs + bnodes usable as G/S/O) → ids 1..E; predicates → ids 1..P
- objects: entities keep 1..E, literals follow as E+1..E+L — matching the macro order, so sorting quads by id tuple equals sorting them by term
That id/rank equivalence is what lets the fast writers sort packed integer keys instead of comparing nodes — provably the same order