Evidence-Carrying Cognitive Mesh on DePIN

This preprint proposes an evidence-carrying cognitive mesh for DePIN-style decentralized compute networks that must continue operating when frontier-scale LLMs and centralized inference services are unavailable (e.g., monopoly, regulation, export controls, supply shocks, or service withdrawal). The core idea is to treat “intelligence” as policy-bounded, locally verifiable capability built from small local models (and other compact AI modules) plus a shared substrate of content-addressed, provenance-rich evidence objects—not as reliance on any privileged evaluator or oracle. The system is designed for an observable-only / no-meta setting: nodes act only on locally verifiable artifacts (messages, signatures, receipts, deterministic computations, bounded statistics), and there is no claim of absolute truth. Instead, the network maintains relative, evidence-conditioned knowledge and a queryable semantic claim graph constructed from verifiable web retrieval, deterministic preprocessing (fetch → render → extract → cite), probes, sensing challenges, and optional human attestations. “High-performance” is defined operationally as sustaining (i) integrity, (ii) capability, and (iii) public accessibility above pinned thresholds, while preserving a measurable margin above an individual-saturation baseline, using only local evidence. The paper provides a self-contained specification (schemas, task families, proof sketches, and control rules) and hardens key attack surfaces relevant to web- and embodiment-coupled intelligence: authenticated lies, context-window DoS, partitioned “mirror worlds,” grinding/withholding randomness, poisoning and backdoors, hub capture, and audit/verification DoS. Optional extensions include verifiable freshness via TLS-notarization / zkTLS-style proofs, challenge–response sensing (proof-of-physicality), and actuation insurance to bound irreversibility risk beyond escrow/slashing. An evaluation plan is included (Ising-style mixing testbed and an adversarial DePIN simulator), with empirical validation left for future work. (Phase-theoretic intrinsic evaluation framework builds on: Takahashi, Collective Phase Transitions beyond Individual Saturation (2025), DOI: 10.5281/zenodo.17853555.)