RFD 1021: A shared holographic algebra library
Problem
RFD 1019 gave the studio core a local module, WeftspunStudio.Hrr. It held a real-valued Holographic Reduced Representation. Binding used a Fourier transform, and the inverse worked only on unitary vectors.
A search of the weftspun repositories found an earlier port of the same idea. weftspun/residual-fsq-recommender held Holo.Core.HRR, a phase-angle representation. An ADR of 2026-07-14 removed it, because a trained model replaced the training-free recommender that used it. The algebra itself was never in doubt.
Two representations of one idea is one too many. The two also do not interoperate, because a vector from one carries no meaning in the other.
Decision
Extract the earlier port into a library, and depend on it.
The library is weftspun/elixir-holographic-reduced-representation, which supplies the :hrr application and the HRR module. The studio core takes it as a Git dependency.
The algebra and the atom generation move over unchanged, so test/fixtures/hrr_golden.json still certifies parity with the Python holographic.py reference to 1.0e-12.
See DETAILS.md for why the phase representation wins, the dimension choice, the migration effects, and the risks. It also holds the Lean proofs and the verified status.