17.8.2 Entity registration (CTTE–1)

17.8.2 Entity registration (CTTE–1)

CTTE–0 gave every occurrence of its mediator the same stable name. Scientific literatures rarely offer this convenience. The same mechanism may be named by an assay, gene product, pathway role, historical convention, or local abbreviation; conversely, two reports may use similar functional language for biologically distinct objects. Before a document compiler can add a generator to its sketch, it must therefore decide which documentary occurrences present the same object. Entity registration is part of theory construction, not clerical preprocessing.

CTTE–1 removes the registered mediator name while retaining the supplied claim parser, four contexts, and one-mediator extension grammar. It contains 800 independently seeded ledgers, 200 from each of four families. A closed ledger needs no new object. In a shared-alias ledger, two different names in each context denote one latent mediator. A distinct-entity ledger has the same relational pattern, but the names denote genuinely different mediators. An ambiguous-identity ledger contains two close candidates measured too imprecisely to justify either merging or separating them. Each occurrence carries noisy four-dimensional assay profiles in addition to its signed, source-bearing relations. Target relations in contexts \(C\) and \(D\) remain sealed until the identity decision, sketch action, and predictions are frozen.

The typed resolver first constructs an equivalence proposal over aliases. For each pair it places a confidence region around the distance between assay profiles. An upper bound below the registered identity tolerance supports equivalence; a lower bound above it supports distinction; overlap with the boundary produces abstain-identity. Only a supported equivalence class may become the new sketch object \(Z\). Contextual factorization and admission are then performed as in CTTE–0. Thus

This order matters. Exact-string matching fragments the shared object. Matching only by relational role merges the distinct controls. A nearest profile rule can recover easy identities, but converts a near-boundary measurement into an unjustified categorical decision. CTTE–1 instead exposes identity uncertainty as an epistemic status that blocks downstream accommodation.

Under the preregistered gates, CTTE–1 was admitted. It achieved \(99.875\% \) four-family action accuracy, constructed the shared mediator and recovered its complete alias equivalence class in every positive ledger, and predicted both sealed target relations in every positive ledger. It retained the closed theory in \(99.5\% \) of closed ledgers, made no false merge in any distinct-entity ledger, and returned abstain-identity in every ambiguous ledger. The single action error was a conservative failure to retain one closed theory after noisy relation reports obscured its direct invariant; no unsupported extension was admitted. Provenance was complete, source splits were disjoint, and deterministic replay reproduced identical artifacts.

Experiment: CTTE–1: alias registration under uncertainty. Input: 800 generated claim ledgers across closed, shared-alias, distinct-entity, and ambiguous-identity families.
Withheld structure: the equivalence class presenting the candidate mediator, followed by the contextual factorization and two target relations.
Structural result: \(99.875\% \) action accuracy, \(100\% \) shared-alias recovery, and \(100\% \) positive exact-pair confirmation.
Epistemic controls: zero false merges of distinct entities and \(100\% \) identity abstention at the registered uncertainty boundary.
Verdict: admitted as a controlled test of \(P_{\mathsf D}\to S_{\mathsf D}\) with supplied assay semantics and grammar; not open-domain entity linking, causal identification, or biological truth.