The minimal set of grounded, coherent, tension-bearing concepts from which consistent narrative conclusions follow inevitably. Not a list of variables — the conceptual foundation that makes reasoning possible and narrative non-arbitrary.

Concept pairs occupying opposite ends of the same analytical dimension. Not logical opposites — structurally opposing concepts within a shared domain. Conditional, not mandatory: most concepts have no natural opposite, and forcing one onto every concept is exactly the failure mode this framework exists to catch.

A concept at the right level of abstraction: precise enough to be grounded in evidence, general enough to reason from. Northern European Premium names a mechanism (a price premium surviving GDP-adjustment in a specific six-country cluster), compresses a pattern that spans that cluster, and creates structural tension against its pseudo-antonym. Countries that are rich is not a calibrated handle — it describes a category, implies no mechanism, generates no analytical direction, and would overclaim, since most high-GDP countries don't actually show the pattern.

When vocabulary is supplied, the AI moves from invention to verification — checking whether concepts are descriptive, domain-coherent, and evidentially grounded rather than generating labels freely. The AI stops guessing meaning and starts checking it.

A label earns concept status only once multiple, diverse indicators converge on the same underlying mechanism. A single grounded data point is an indicator, not a concept — it points toward a possible category, it doesn't establish one. Where indicators diverge — the same statistical direction produced by genuinely different mechanisms — the honest result is not a concept; it's residual.

The signal to stop refining a concept, or to stop generating new ones. A concept is saturated once new indicators only extend its range instead of reshaping what it means. A board is complete, for the evidence at hand, not when every data point has a label, but when the next candidate grouping fails the convergence test — that failure is the stopping signal, named honestly in the residual rather than forced into a manufactured concept.

Among a board's concepts, the one with the greatest explanatory reach — not just central, but the pattern the others are read as deviations from. Distinct from Dominant centrality, a per-concept weight: a board can have several Dominant concepts, but only one plays this organizing role relative to the rest.

Centrality of a concept in the evidence base — in Glaser & Strauss's terms, a property, with each concept's Dominant/Present/Edge Case rating as its dimension: a specific position on that property's range, not a fixed identity. Three levels:

The minimal board contains the core deducible space — the smallest coherent set of concepts from which the global story follows. The expandable board adds shadow concepts and edge cases: the structural tensions that challenge or complicate the dominant narrative.