Psi-epistemic

In plain terms

Here is the problem. Two readings of the quantum state have fought for a century. Realist: the state is a physical thing, so its jump on measurement is a physical process needing an explanation the theory does not give. Epistemic: the state is a description of what someone knows, so the jump is just an update -- learn something, revise -- no more mysterious than a poker player recalculating after seeing a card.

The epistemic reading dissolves the measurement problem, which is its great charm. Its great embarrassment is the follow-up question: knowledge of WHAT? If there is nothing underneath, the reading says the theory is about nothing in particular. And theorems from the last fifteen years -- PBR most famously -- make naive versions of the underneath very hard to hold.

So the position is attractive, constrained, and incomplete without a floor.

In CSD

CSD is psi-epistemic with a floor. The state is knowledge of which region of the constraint surface the microstate occupies. There is a fact underneath -- one point, one trajectory -- and the state summarises what a preparation pins down about it.

Everything in the programme sorts into two levels, and keeping them apart is the discipline the whole account depends on. Epistemic: the projective state space, the outcome regions, the probabilities. Ontic: the substrate and the measure that weights it. The corpus's fibre analysis sharpened the split late in the day: it treats the projective space as a lossy projection, with the record-forming content placed in the fibre above it rather than on it. That placement is a judgement against a tight constraint chain, not a proved dichotomy -- whether the content could stay on the base at N of three and above is open in both directions.

Keep three claims apart, because the corpus once ran them together. (A) CSD epistemicity: [psi] does not completely specify the microstate -- the projection is many-to-one. That is this entry's subject. (B) The technical Harrigan-Spekkens classification: on the canonical EXACT sharp interface CSD is psi-ONTIC, because distinct exact pure states have mutually singular ontic measures (RecordLayer.sharp_preparations_mutuallySingular). (C) Finite-resolution overlap: positive-volume REGION preparations whose regions overlap have conditional laws that provably cannot be separated onto disjoint supports.

(C) is the theorem anchored here. ⚠️ It does NOT establish (B) in the epistemic direction: region preparations are a different preparation class from exact pure states, and describing (C) as technical psi-epistemicity of exact states was an error, corrected 2026-08-25. (A) and (C) are true; on (B) the exact interface is psi-ontic.

Mathematically

The overlap statement is proved: conditional_not_mutuallySingular -- overlapping preparation regions give non-mutually-singular conditional laws, by a density argument rather than a shared-support one, because two measures can both charge a common set and still be mutually singular, so overlap alone would not settle it -- with the concrete witness kahler_preparations_overlap instantiating it on open neighbourhoods of any two rays of the working arena. The complementary facts are proved too: conditioning on an exact pure state is an unreachable idealisation (single fibres are Liouville-null), and what a physical preparation projects to is a density against the Fubini-Study measure, not a point.

Which literature constraints bite -- PBR and its relatives -- still depends delicately on what the state is taken to be knowledge of, and on independence assumptions. The full mapping of that boundary is not formalised; treat it as live terrain.