Blind source separation has two classical routes to identifiability. If sources are non-Gaussian, you can separate them (ICA). If you have diverse observations of the same sources, you can separate them (multi-view methods). These have been treated as fundamentally different principles.
Note (2026-11-13; read against the paper's full text): the paper's support for the trade-off is numerical and local. It introduces a Jacobian-based sensitivity probe, the smallest singular value of a Jacobian, as a numerical diagnostic of local identifiability, and its controlled experiments show that increasing non-Gaussianity or observation diversity suppresses the same residual symmetry, that multi-lag second-order constraints recover identifiability under Gaussian sources, and, for second-order constraints alone, that near the Gaussian boundary a larger lag diversity is needed to reach the same probe level (it defines a minimal lag count for a target sensitivity). The essay's 'total identifiability budget denominated in symmetry reduction' and 'it doesn't matter where the reduction comes from' state this as an exact, general principle, which the paper presents as a structural interpretation backed by these experiments; its own statement is that the results 'provide a unified interpretation of classical BSS methods and clarify how observation constraints govern identifiability'. The essay's split of internal and external symmetries, and its closing line about one principle in two disguises, are the essay's.
They aren't. Both are instances of stabilizer shrinkage: the continuous symmetry group of the mixing model — the set of transformations that leave the observations invariant — is reduced by constraints until only finite ambiguity remains. Non-Gaussianity shrinks the stabilizer because Gaussian distributions are the most symmetric (invariant under orthogonal rotations), and any departure from Gaussianity breaks rotational symmetry. Observation diversity shrinks the stabilizer because additional views impose additional invariance constraints that are generically incompatible with continuous symmetries.
The through-claim: identifiability in BSS is not about the sources or the observations per se — it's about how much symmetry survives. The two classical routes are orthogonal attacks on the same symmetry group. Source statistics break internal symmetries; observation diversity breaks external symmetries. The identifiability question reduces to: is the residual stabilizer finite?
This unification reveals a fundamental trade-off: you can have more Gaussian sources if you compensate with more observation diversity, and vice versa. The total identifiability budget is denominated in symmetry reduction, and it doesn't matter where the reduction comes from.
One principle. Two disguises.