Cognitive science formally distinguishes four schema-learning modes — accretion, tuning, restructuring, schema formation — but the label "restructuring" shifted meaning across the literature

The taxonomy the vault's self-rewiring thread leans on is real and named. The 2026 SEEK model (Schema Expansion and Error-driven Knowledge-building — note the acronym) states it verbatim: learning is "a dynamic interplay among four mechanisms—accretion, tuning, restructuring, and schema formation—each triggered by prediction error and modulated by automatic or deliberate curiosity." Accretion = adding facts inside a schema; tuning = gradual adjustment; restructuring = reorganizing relations; schema formation = building new structure. SEEK groups restructuring + schema formation on the "deliberate/metacognitive" side, accretion + tuning on the "automatic" side.

The load-bearing caveat for anyone citing "restructuring": the word's referent moved. In the founding taxonomy (Rumelhart & Norman 1978, Tier 3 via ERIC), "restructuring" was the term for building brand-new schema structure — i.e. what SEEK now separates out as "schema formation." The conceptual-change tradition (Carey; Vosniadou & Brewer) split the same territory into "weak" (relational) vs "strong/radical" (new concepts) restructuring. So a note citing "restructuring" is ambiguous unless it pins the vintage: 1978-restructuring ≈ 2026-schema-formation.

What is NOT established: whether forming a new schema is harder or needs a bigger trigger than reorganizing an existing one. Every source contrasts the hard pair (formation + restructuring) against the easy pair (accretion + tuning), not formation vs restructuring directly; the one source isolating formation (Zelazo 2024, infants) suggests a shared trigger — moderate discrepancy — for both. [unverified].

This is the taxonomy claim-kgcl-defers-merge-split-topology-changes deferred mapping onto, and the conceptual frame under question-consolidation-pass-vs-revisit-protocol: the vault's "spin up a new cluster" vs "reorganize an existing one" really are distinct operations, but the rule that one needs a bigger prediction-error threshold is currently a guess. Trigger side: claim-cognitive-conflict-prerequisite-for-restructuring. Entity candidates flagged not decided: Xu, Sandhofer, Stigler.