Record layer

In plain terms

Here is the problem. Suppose outcome probabilities really are volumes of regions. You still owe an account of the outcome itself: after the experiment, something in the world -- a pointer, a grain of silver, a memory -- holds which result occurred. A theory that prices outcomes but cannot say where the outcome is written has stopped one step short.

The record layer is that last step. It asks for two things. The outcome regions must be fixed by the apparatus alone, the way a referee's rulebook does not depend on which team is playing. And the record must be a physical configuration of the arena -- an ontic fact, selected by the one real trajectory.

Nothing mystical is on offer: a record is structure the arena already has, singled out by where the trajectory went.

In CSD

This is the programme's near frontier, and it is partly built. The context-fixed partition exists: basins defined from the apparatus's rate field, with no preparation anywhere in the definition, whose weights come out as Born weights. At the qubit the whole story is proved end to end; at higher dimension the analysis indicates the contextual structure does not sit on the base alone, and the corpus places it in the fibre. That placement narrows where the mechanism can be, though it rests on refuting one natural class of base candidate rather than on a no-go.

What remains open, and is stated as open: deriving the fibre mechanism from de-isolation dynamics rather than exhibiting it. The record layer is where the reconstruction is completed or not.

Mathematically

The proved tier: a measurable simplex-valued rate field on the base defines pairwise-disjoint basins whose epistemic probability at preparation psi is the Born weight -- a partition that never mentions psi. Sequential and mixed-state versions run through the Luders tier. The open tier is dynamical, not kinematic: no interaction Hamiltonian generating the basins is constructed, and the corpus says so in the module that would need it.