Evidential Survivability: OCP and Ethereum's CROPS Direction

Evidential Survivability: OCP and Ethereum’s CROPS Direction

Independent Verification Infrastructure for the AI Era

Damon Zwicker — May 2026


Abstract

Ethereum is increasingly positioning itself not as a faster or cheaper execution environment, but as durable public infrastructure for an AI-driven world. Vitalik Buterin has articulated a direction for the Ethereum Foundation emphasizing censorship resistance, openness, privacy, and security — what he describes as the CROPS dimension — as the foundational priorities for Ethereum’s long-term trajectory.

At the same time, AI systems are evolving from passive tools into autonomous operational actors capable of dynamic execution, delegated decision-making, and machine-to-machine economic coordination. This convergence creates a new infrastructural problem: how can observations, actions, and commitments remain independently verifiable after the originating systems, vendors, orchestration layers, or execution environments change?

The Observation Commitment Protocol, or OCP, is not an AI governance framework, execution environment, or runtime control system. It defines one narrower infrastructural role: independent verification. This paper argues that OCP is an example of the kind of narrow, public, independently verifiable infrastructure that Ethereum’s CROPS direction makes increasingly necessary — and that this alignment is already visible in public implementations today.


Ethereum’s Shift Toward Public Infrastructure

Ethereum’s emerging direction reflects a judgment that the protocol’s long-term significance derives not from transaction volume or execution throughput, but from becoming a trust anchor and coordination substrate for increasingly autonomous systems.

The CROPS framework — censorship resistance, openness, privacy, and security — represents a commitment to building infrastructure that remains trustworthy under adversarial conditions, institutional change, and the passage of time. Buterin is explicit about this direction: “I think Ethereum should strive the hardest to be deeply impressive in a different dimension: the CROPS dimension.”

This is a fundamentally different objective than systems optimized around operational convenience or short-term performance metrics. It implies that the most important property Ethereum can provide is not speed, but survivability — the ability for records, commitments, and proofs to remain independently verifiable long after the systems that produced them have changed or disappeared.

This is exactly the infrastructural property that becomes most valuable in the AI era. AI systems increasingly generate decisions, observations, recommendations, and autonomous economic actions. As these systems become more capable and more embedded in consequential processes, the integrity and survivability of their evidential history becomes a first-order concern — not just technically, but legally, regulatorily, and institutionally.


The Verification Crisis

Modern governance architectures already rely on layered visibility systems: dashboards, monitoring frameworks, policy engines, attestation systems, and runtime controls. These systems may govern behavior during operation. But operational visibility during execution is not the same as independently survivable accountability afterward.

Over time, systems encounter litigation, audits, regulatory scrutiny, adversarial review, vendor turnover, infrastructure migration, or outright platform failure. At that point, organizations discover that the ability to observe a system during operation does not guarantee the ability to independently reconstruct what happened later.

This gap becomes more acute in probabilistic AI systems where execution paths evolve dynamically, delegation chains fragment, and operational state may exist only transiently. The AI era introduces a new infrastructural challenge: evidential survivability — the ability of a commitment to outlast the infrastructure that produced it.

Without an independent verification boundary, commitments are implicitly bound to the systems that created them. When those systems change, the verifiability of the commitment changes with them. This is not a governance failure. It is an architectural one.


The Narrow Role of OCP

The Observation Commitment Protocol intentionally limits its scope. It does not attempt to solve truth, authorship, governance, intent, or AI alignment. It isolates a narrower question:

Can a committed digest be independently recomputed, compared, and checked against public ledger state — without relying on the originating platform, SDK, gateway, vendor, or orchestration environment?

Its verification invariant is minimal by design:

recompute → compare → confirm inclusion

Verification operates directly against public ledger state. No trusted dashboard is required. No trusted gateway is required. No trusted SDK is required. No trusted indexer is required. A verifier needs only a public RPC endpoint and the original observation bytes.

OCP is chain-agnostic, but Ethereum’s CROPS direction provides one of the clearest philosophical homes for this kind of verification primitive.


What OCP Does Not Claim

OCP does not prove that an observation is true.

OCP does not prove that an actor was authorized.

OCP does not prove that an AI decision was correct.

OCP does not guarantee data availability.

OCP proves that a digest can be independently recomputed and matched to a public commitment on-chain.

This boundary is intentional and non-negotiable. It is what makes OCP composable with identity systems, governance frameworks, and attestation layers without absorbing their semantics.


CROPS and OCP: The Architectural Alignment

Censorship Resistance

Once committed, the digest remains independently checkable through public ledger state. Verification does not depend on the originating application, vendor, gateway, or institution remaining available. This is what censorship resistance means in practice: not the absence of an adversary, but the structural inability of any single actor to revoke the verifiability of a commitment after the fact.

Openness

OCP verification is publicly recomputable. Any verifier with access to public chain data can independently confirm the digest, the inclusion, and the transaction linkage. There is no permissioned verification layer, no API key required, no SDK dependency. The verification procedure is a published algorithm applied to public data. Any implementation that follows the algorithm produces the same result.

Privacy

OCP does not make private data public. It allows a public commitment to exist without requiring the underlying data to be disclosed. The commitment itself — the digest anchored on-chain — reveals nothing about the observation it represents. Privacy depends on what is later disclosed: verification requires the relevant observation bytes, but the commitment can exist indefinitely without them. This creates a separation between verification capability and data disclosure that is meaningful for systems operating on sensitive inputs.

Security

OCP minimizes trusted surfaces. The verifier does not need to trust proprietary APIs, centralized dashboards, gateways, SDKs, or operational vendors. This makes verification substantially more resilient under adversarial conditions. An attacker who compromises the originating system, gateway, or SDK cannot retroactively invalidate the on-chain commitment because the committed digest remains independently checkable through public ledger state.


Modular Trust Architecture

One of the most important architectural shifts emerging in the AI era is the decomposition of trust into independent infrastructure layers. Increasingly, production systems are separating identity, commitment, verification, governance, and execution admissibility into composable components rather than collapsing them into a single platform trust assumption.

This is visible in the stack that has emerged from collaborative work across four builders in three countries:

Layer Standard Question it answers
Identity ERC-8004 Who is the agent?
Input Trust WYRIWE / ERC-8263 profile What did the agent actually receive?
Commitment ERC-8263 What was committed on-chain?
Verification OCP Can the commitment be independently verified from chain state alone?
Interface ERC-8274 Can any contract consume the proof?

Each layer answers a materially different question. Each is independently specified. None absorbs the semantics of the others.

This separation matters because governance systems themselves eventually become subject to institutional drift, infrastructure replacement, adversarial challenge, and operational decay. Without independent verification boundaries, governance architectures risk collapsing into opaque authority systems whose integrity must simply be trusted. OCP provides the verification boundary — the structural guarantee that what was committed can be independently checked — on top of which governance systems can operate with more confidence.


Production Evidence

This is not only theoretical alignment. The stack is already visible in public implementations.

ERC-8263 v0.2 cites OCP’s boundary statement verbatim in its verification disclaimer. The verification invariant is formally codified in Appendix A.2 of the active Ethereum standard, PR #1748, ethereum/ERCs. As of May 2026, over 742 proofs have been anchored across four chains. A live AI agent bounty settled on Base Sepolia in May 2026 — no oracle, no intermediary, no trusted third party — with the identity, input provenance, commitment, verification, and gateway attestation layers running end-to-end.

ERC-8274, AI Inference Proof Verification Interfaces, PR #1771, names OCP as the unifying verification primitive across zkML, opML, TEE, oracle, and multisig backends. A Composition Note co-authored by Vincent Wu, Damon Zwicker, and Tiago Merlini documents how the layers compose and is published on ethresear.ch. Full references are in the appendix.

The architecture Ethereum’s CROPS direction calls for is not waiting for future development. It exists today, operating on public infrastructure, verifiable by anyone with a public RPC endpoint.


Why Narrow Systems Matter

OCP is an example of a class of infrastructure Ethereum’s CROPS direction makes necessary: narrow, public, independently verifiable primitives that provide durable guarantees without absorbing governance, identity, or application semantics.

Ethereum increasingly favors modularity, credible neutrality, minimized trust assumptions, and constrained protocol scope. OCP follows the same architectural discipline. It does not attempt to absorb governance, execution control, identity semantics, or institutional authority.

This narrowness is a resilience strategy. Systems that attempt to absorb too many responsibilities often become centralized, opaque, politically brittle, and difficult to independently verify. Constrained systems are easier to reason about under adversarial conditions. They are easier to audit, easier to compose with, and easier to trust precisely because there is less to trust.

In the AI era, where systems are probabilistic, autonomous, and difficult to predict, the ability to reason clearly about what a primitive does and does not do becomes increasingly valuable. A verification layer that also handles identity is harder to trust than one that handles only verification. A commitment layer that also handles governance introduces entanglement that makes both functions harder to audit independently.

The decomposition is the architecture.


Conclusion

Ethereum’s CROPS direction reflects an attempt to preserve trustworthy public infrastructure in an era increasingly shaped by autonomous systems, probabilistic computation, institutional centralization, and machine-scale coordination. The most important property Ethereum can provide is not execution speed — it is evidential survivability: the ability for commitments to outlast the systems that produced them.

OCP is an example of the kind of narrow, public, independently verifiable infrastructure this direction makes necessary. It does not answer what is true, what is allowed, or who is responsible. It answers whether a committed digest can be independently recomputed and matched against public ledger state — by anyone, at any time, from any client, without trusting the system that produced it.

Independent verification does not solve the AI alignment problem. It does not replace governance. It provides the floor beneath both.

That is not a small thing. In the era of autonomous computation, it may be one of the most important properties any infrastructure can provide.


References and Evidence Appendix

CROPS / EF Direction

Vitalik Buterin, May 2026:
https://x.com/VitalikButerin/status/2058583593102844111

OCP

Specification:
https://github.com/damonzwicker/observation-commitment-protocol

ethresear.ch publication:
https://ethresear.ch/t/observation-commitment-protocol-ocp-v1-0-0/24602

npm package:
https://www.npmjs.com/package/ocp-verify

Base Sepolia contract:
0x0963Fd33DF80c94360F2DC22e5c09517AeE7ED5c

Solana devnet:
GCXRKzreL2fdYBpnfmKzFqTxE46eGmwQuErMw4uZ1DUL

ERC-8263 v0.2

PR #1748:
https://github.com/ethereum/ERCs/pull/1748

ETH Mainnet contract:
0xe95d6a15966984c209a62a2c188828555eb5ec3d

L2 mainnets, Polygon / Base / BSC:
0x87dd3A56AFD0D2c488aD7E13fB036b59144b25dC

ERC-8274

PR #1771:
https://github.com/ethereum/ERCs/pull/1771

Ethereum Magicians thread:
https://ethereum-magicians.org/t/erc-8274-ai-inference-proof-verification/28083

ERC-8004

Ethereum Magicians thread:
https://ethereum-magicians.org/t/erc-8004-trustless-agents/25098

Reference implementation:
https://gateway.ensub.org

WYRIWE

ERC draft:
https://github.com/TMerlini/wyriwe

Ethereum Magicians thread:
https://ethereum-magicians.org/t/wyriwe-what-you-read-is-what-you-execute-input-provenance-for-verifiable-ai-inference/28655

Composition Note

ethresear.ch:
https://ethresear.ch/t/composition-note-erc-8004-erc-8263-ocp-a-reference-guide-for-implementers-building-on-the-ai-agent-verification-stack/24995

Gist:
https://gist.github.com/damonzwicker/8742e742bdc627b8e2179c00b81289dc

Live Bounty Settlement — Base Sepolia, May 2026

Contract:
0x57fe09a6Eb8d5741b24fF640AA8Bc4D2010B93D7

Claim transaction:
https://sepolia.basescan.org/tx/0x2780f94d11ad3766e7d370b1016d3ba917933f8111ddd42d3fcab2226cd380e4

Verify endpoint:
https://gateway.ensub.org/agent/verify/758d61f26a44448384e5c4468a0dcb7a2abe456067b0f7b505bc28b9411fe931

Contract source:
https://gist.github.com/TMerlini/bf3abd30c332cccb257d0e5bdff1ff95

OCP Discussions-To

Ethereum Magicians:
https://ethereum-magicians.org/t/draft-erc-observation-commitment-protocol-ocp-chain-agnostic-cryptographic-commitment-primitive/28399

3 Likes

The “evidential survivability” framing is the clearest articulation of why the commitment layer matters more than execution speed, and it maps directly onto CROPS in a way that’s worth making explicit for the composed stack, not just OCP in isolation.

The same property extends one layer upstream. OCP anchors the digest before execution. WYRIWE anchors what was read before execution, the raw input, the sanitization spec applied, and the final committed input. Both survive independently. If OCP evidential survivability means “the commitment remains verifiable after the platform disappears,” WYRIWE means the input provenance remains verifiable too. The chain holds end to end: what was read → what was committed → what was verified.

On CROPS applied to the full stack: censorship resistance is the on-chain anchor (OCP); openness is public RPC recomputation with no proprietary SDK; privacy is hash commitment without disclosure of underlying data; security is the elimination of trusted intermediaries at every layer. The 742 proofs and the Base Sepolia settlement are evidence that this isn’t theoretical, the stack is already operating under these constraints in production.

2 Likes

Exactly — this is the point I think becomes increasingly important as the stack composes.

OCP gives the commitment survivability layer: the digest remains independently verifiable even if the originating platform, gateway, SDK, vendor, or execution environment disappears.

WYRIWE extends that survivability upstream by preserving the input provenance boundary: what was read, what sanitization or transformation rule was applied, and what final input was committed before execution.

So the composed stack is not only asking whether a proof can be verified after the fact. It is preserving the evidential chain that led into the proof:

what was read → what was committed → what was verified → what can be consumed by another contract or system.

That is why the CROPS mapping matters at the stack level, not only at the OCP level. Censorship resistance, openness, privacy, and security become much stronger when each layer maintains a narrow boundary and remains independently checkable.

The 742 proofs and the Base Sepolia settlement matter because they show this is no longer only an architecture diagram. It is already operating as public verification infrastructure.

1 Like

One update since this paper was published: OCP has been assigned an official Ethereum standard number.

OCP is now ERC-8281. The PR is live at Add ERC: Observation Commitment Protocol by damonzwicker · Pull Request #1788 · ethereum/ERCs · GitHub , 10/10 CI green, currently in editor review.

The full stack is now fully numbered:

ERC-8004 — agent identity WYRIWE — input provenance ERC-8281 — independent verification (OCP) ERC-8263 — on-chain commitment ERC-8274 — proof verification interface

The evidential survivability argument in this paper now has a formal standards path behind it. Each layer independently specified, each with a number, each composable without absorbing the others.

1 Like