Redescription at the Edge of Completion: At the Limits of Saturation — A Structural Framework for Human–AI Co-Creation

Intellectual leaps are often described as moments of intuition, creativity, or chance and treated as irreproducible. Yet many occur at the edge of near-complete explanation: refinement has made a frame feel coherent, while something still does not fit. This paper proposes a structural framework for that pre-leap region by treating it as a pattern in how explanatory structures behave under an explicitly fixed frame (premises, evaluative axis, and admissible connective relations). When refinement is pushed within the same frame, progress can stall not by insufficiency but by completion: structure-preserving repairs proliferate—exceptions, case splits, auxiliary clauses—while the same within-frame failure condition persists rather than shrinking. We identify what remains as a residual discrepancy and define the boundary diagnosis, non-connection/discontinuity, as the point where eliminating the residual requires either open-ended growth in structure-preserving patching (patch burden) or an explicit reorganization of admissible connections. AI is used only as an external saturation and stress-testing instrument: under a fixed frame and a non-sycophantic diagnostic interface, it can make closure dynamics and patch burden legible as repeatable diagnostics without treating any output as authoritative. At the boundary, the key issue is whether the residual discrepancy can be sharpened into candidate constraints and a subset adopted as binding. The targeted discontinuity is diagnosed as seed emergence: a minimal connecting move that begins to satisfy adopted constraints while reducing the dominant patch-burden regime rather than renaming it. The contribution is an operational route from saturation diagnostics to constraint-governed reconstruction, with explicit failure modes and retrospective recognition via stabilization and independent reconstructibility.