Over the past decade, research into distributed systems, cryptographic proofs, and decentralized computation has produced a structural insight that is rarely stated but frequently overlooked:

verification is tractable only after the evidence required for verification is itself made available in a form that prevents silent omission.

This insight is central to recent developments on the Ethereum protocol — specifically the deployment of Peer Data Availability Sampling (PeerDAS) on mainnet and the emergence of production-quality Zero-Knowledge Ethereum Virtual Machines (ZK-EVMs) at the level of performance readiness. The significance of these developments lies not merely in performance or capacity, but in the fact that they embody an architectural progression from making governing information recoverable to making correctness proofs meaningful.

This essay examines that progression and explores its relevance for governance in agentic artificial intelligence systems.

The Data Availability Problem and Its Solution

In 2015, Ethereum researchers articulated a fundamental challenge confronting light clients and sharded protocols: how can a participant become confident that block data exists and is retrievable without downloading it in full? Traditional approaches relying on fraud proofs or replication run into a core difficulty: withholding data is not inherently distinguishable from benign absence, and a protocol cannot safely advance without a mechanism that makes withholding detectable.

The research response was the application of erasure coding and data availability sampling. In this model:

• Original data is expanded into a set of redundant chunks using polynomial interpolation or multidimensional coding.

• Participants randomly sample subsets of chunks and verify their availability via Merkle proofs.

• If sufficient sampled chunks are retrievable, probabilistic confidence in the availability of the full data can be established without full download.

• Dual Merkle tree constructions and optimized coding reduce proof size and computational costs.

Crucially, these techniques do not prove that the data is correct. They prove that the data is sufficiently available that a correctness proof could, in principle, be generated and verified. This decoupling of availability from correctness enabled scaling without imposing onerous storage or bandwidth requirements on every participant.

PeerDAS and ZK-EVM: Evolution Not Revolution

The recent convergence of PeerDAS and ZK-EVMs represents a maturation of these insights in deployed systems. PeerDAS operationalizes decentralized, probabilistic availability sampling on mainnet. ZK-EVM implementations aim to provide succinct proofs of execution validity that can be checked efficiently.

Together, these components support a network in which:

• Decentralization is maintained,

• Consensus is preserved,

• Bandwidth and throughput are increased.

This is often described as resolving the “blockchain trilemma” in practice. But it is not the result of a single breakthrough. It is the consequence of correct sequencing: first, make the information that would be needed for verification observable, then build mechanisms that compress and prove correctness.

Without data availability guarantees, validity proofs are economically and operationally fragile. Proof systems need retrievable inputs. If the inputs cannot be shown to exist and be accessible, proof generation and verification are moot. PeerDAS makes absence detectable; ZK-EVMs make correctness verifiable. The two work together precisely because the former makes the substrate for the latter meaningful.

Community Responses and the Real Structural Shift

Community reactions to Vitalik Buterin’s recent announcements illustrate different emphases:

• Optimistic responses highlight throughput and decentralized scaling.

• Others emphasize the need for safety and risk considerations.

Both points are valid, but they understate the deeper architectural shift:
the protocol now treats data availability as a first-class property rather than an implicit assumption.

This is consequential for systems that require trust and verification across distributed actors. Scale without observability is brittle; correctness without access to evidence is hollow.

An Analogue in Agentic Systems

Contemporary research on verifiable AI often focuses on inference proofs: demonstrating that a model executed faithfully, or that a neural network’s computation can be verified via cryptographic means such as zero-knowledge machine learning proofs. These approaches aim to ensure correctness of reasoning.

However, in practice, deployed agentic systems fail not because of subtle inference errors but because the information governing decisions is not itself verifiable. An agent may depend on inputs that were not recorded, apply policy rules that are implicit rather than formal, or involve human approvals outside of any auditable log. These failure modes are not about computation errors; they are about withholding relevant governance information.

Here we can observe an important structural parallel with the data availability problem in blockchains. In both cases, the absence of evidence makes verification unattainable. In distributed ledger systems, the absence of block data prevents correctness proofs from being checked. In agentic systems, the absence of governing evidence prevents understanding, reconstruction, or audit of decisions.

Thus, a direct translation of the data availability insight to agentic AI is the requirement that governing information must itself be made available in a structured and verifiable form before correctness proofs concerning reasoning or inference can be meaningful.

Governance Primitives for Agentic Verification

At Obsqra Labs and in the development of Obsqra.fi, this insight has led to the identification of a set of primitives that collectively satisfy this requirement of governance availability:

1. Deterministic Constraints
   These are explicit, enforceable boundaries on agent actions. They are designed to fail closed, making violations detectable rather than silently permissible.

2. Declared Intent and Declared Retrieval
   Agents are required to explicitly commit to the inputs and context they use for reasoning. This makes the epistemic foundation of decisions observable and subject to audit.

3. Identity-Bound Policy Evaluation
   Policies must be evaluated prior to execution and tied at the cryptographic level to identifiable actors. This ensures that authority and scope are verifiable.

4. Receipts and Trace Artifacts
   The system emits hash-linked records capturing intent, policy decisions, retrieval contexts, and execution outcomes. These artifacts enable replay and selective audit without requiring full replication of all agent activity.

Taken together, these primitives create a substrate in which the information governing behavior is non-withholdable. This is analogous to the way data availability sampling ensures that block data cannot be silently concealed. Once the governing information is thus rendered observable, it becomes feasible to attach correctness proofs concerning inference or reasoning in a meaningful way.

Sequencing Guarantees: A Structural Principle

The Ethereum research community’s work over the past decade exemplifies a disciplined sequencing of guarantees: first make the evidence necessary for verification available and observable; only then build mechanisms to prove correctness of computation. This sequencing reduces trust assumptions and enables scaling without replication.

Agentic systems require a similar staging. Attempts to prove inference correctness in the absence of verifiable governance information are premature. Correctness proofs that lack an accessible substrate of governing evidence are limited in practical utility.

Just as data availability sampling provided probabilistic confidence without full replication, governance availability primitives provide auditability of agent decisions without requiring replication of all internal reasoning processes.

In both systems, verification is not a monolithic property. It is a sequence of layered guarantees, each grounded in the preceding one. Availability must come first. Correctness follows.

Conclusion

The recent developments in Ethereum’s data availability and verification infrastructure illustrate a structural pattern of staged verification guarantees. When the protocol makes the evidence needed for verification itself available, it becomes possible to build meaningful correctness proofs and to scale trust without replication.

Agentic AI systems face an analogous requirement. Without governance availability — that is, without explicit, retrievable, and auditable records of constraints, policy decisions, identities, and input commitments — attempts at inference verification are epistemically hollow.

The lesson from blockchain research is clear: make what you need to prove available before you attempt to prove it. Only then can systems with real-world consequences be made trustworthy at scale.