ifc-0249

17.9 A corpus-to-theory evidence hierarchy

CTTE–0 isolates the value of contextual composition, but it supplies most of the language machinery. The evidence hierarchy hides these supports one at a time:

Stage	Withheld structure	Construction	Primary test
CTTE–0	one named mediator	source-bearing contextual factorization	locked relation prediction
CTTE–1	entity registration and aliases	typed object and equivalence class	entity precision, coverage, and abstention
CTTE–2	context boundaries	local-model cover and restriction maps	contradiction preservation and transfer
CTTE–3	decisive source or query	active acquisition policy	power and time to decision at matched cost
CTTE–4	domain-level constructor choice within a supplied grammar	finite sketch extension and experimental question	irreducibility and sealed benchmark-probe admission
CTTE–5	domain presentation	transport to a generated target presentation	provenance-safe reuse without semantic leakage

Within their registered generated worlds, CTTE–1 establishes controlled entity registration; CTTE–2 nearly recovers a minimal context cover but exposes an epistemic-status ordering defect; CTTE–3 shows that a support-first obstruction can select the missing overlap efficiently within a supplied query language; CTTE–4 constructs a finite sketch extension and an external question rather than choosing among completed theories; and CTTE–5 transports that construction and its experimental operations through a changed typed interface. The sequence therefore reaches domain transport, but only under supplied calibration semantics. Real literatures pose the harder problem of discovering which populations, doses, time scales, assays, and tissues define the relevant cover, which measurements are feasible, and which parts of the interface can be aligned at all. Those unresolved tasks define the transition to the GLP-1 challenge.