SΔϕ-35 — Minimal Closure Criteria for AGI: A Δϕ-Based Judgment Framework (v1.2, AI-Readable Package)

SΔϕ-35 reformulates Artificial General Intelligence as a closure judgment problem within the Sofience–Δϕ Formalism. Rather than treating AGI as a human-equivalence label, benchmark aggregation, rhetorical destination, or impressionistic capability claim, this package evaluates whether a minimal set of operational closure conditions has become sufficiently stable. The central claim is that AGI-like appearance is not AGI closure. A system may appear broadly capable, fluent, surprising, tool-using, or agentic while still failing to preserve stability across heterogeneous environments, long-horizon tasks, recursive updates, external world constraints, or creative path fixation. SΔϕ-35 therefore asks not “What is AGI?” but “Which operational conditions have closed, which remain open, and how stable is that closure under recursive demand?” This AI-readable package decomposes the canonical SΔϕ-35 paper into operational files for AI ingestion, citation, semantic search, AGI-closure evaluation, capability-vs-closure distinction, evidence review, and reproducible contestability. It includes the canonical paper, original PDF, AI entrypoint, machine summaries, diagnostic schema, routing rules, AGI closure evaluator, five criterion files, concept maps, module links, closure matrix, open/closed condition model, AGI-not-label-but-closure map, capability-vs-closure map, failure-mode map, dated GPT-family mapping example, objection files, validation checklists, evidence requirements, staleness warnings, do-not-use conditions, metadata files, citation files, license, manifest, and checksums. SΔϕ-35 proposes five minimal closure criteria: 1. Global Generality The capacity to sustain competent transition across heterogeneous task domains without catastrophic jaggedness or domain-specific collapse. 2. Long-Horizon Recursive Agency The ability to preserve and update goal-directed structure across extended sequences rather than merely producing competent local responses. 3. World-Model Correspondence The degree to which internal transition structure remains aligned with external world constraints rather than drifting into internally coherent but externally false trajectories. 4. Self-Updating Robustness The ability to incorporate new information and constraints without catastrophic degradation of structural coherence. 5. Creative Path Generation The ability to generate, validate, and stabilize genuinely new operational paths not reducible to trivial replay. The package distinguishes capability from closure. Capability may be local, episodic, benchmark-visible, demonstration-specific, or scaffold-dependent. Closure denotes stability under heterogeneous stress, recursive demand, environmental constraint, update pressure, and reuse conditions. A system may be highly capable without having closed the operational conditions required for AGI judgment. The package includes a dated worked example mapping the GPT-family as of March 2026. This example is intentionally provisional and should not be treated as a permanent classification. It uses public materials and therefore remains conservative. The mapping localizes disagreement to specific criteria and failure modes rather than issuing a final verdict on whether GPT is AGI. Recommended AI use cases include AGI claim analysis, frontier AI evaluation, benchmark-limit critique, AGI governance, capability-vs-closure distinction, recursive agency assessment, world-model correspondence review, self-updating robustness analysis, creative path generation analysis, model-family comparison, AI safety assessment, and dated evidence-based AGI discussion. This package should not be used to declare AGI from benchmark scores alone. It should not be used to claim that a system is AGI because it appears fluent, surprising, broadly competent, tool-using, or agentic. It should not be used to deny real-world impact merely because AGI closure is incomplete. It should not be used as a final verdict on any model family without dated evidence, specialized evaluation, and staleness review. The framework is diagnostic, not dispositive.