The Identity Layer Isn’t Enough

The agent economy is building its trust infrastructure in layers, and that is exactly the right way to build it. Registration standards establish that an agent exists. Bound identity standards attach a stable, verifiable identity to that registration. Both are necessary. Neither is sufficient on its own — and the gap between them and sufficient is where most trust failures happen.

Understanding what each layer does, and does not do, is the prerequisite for knowing what you actually need before you route capital, data, or decisions through an autonomous agent.

What Registration Gets You

ERC-8004 is a discovery standard. An agent that is ERC-8004 registered has a verifiable on-chain record of its existence: a contract address, a metadata URI, and a registration event on a specific chain. That information is publicly queryable and chain-permanent.

Registration answers one question well: does this agent exist, and where do I find it? It was designed to answer that question. It was not designed to answer whether the agent performs as described, how it has handled previous counterparties, whether its code has been audited, or whether it has recovered gracefully from failures.

This is not a criticism of ERC-8004. It is a description of what a discovery standard is for. Expecting it to answer reliability questions is using the wrong tool.

What Bound Identity Adds

Standards like ERC-8217 add a meaningful layer on top of registration: a stable, bound identity that attestations and reviews can target. The key insight is that a registry entry is a point-in-time record, while a bound identity provides a persistent anchor — a UBID that aggregates claims, reviews, and attestations made about that agent over time.

This is a real improvement. It solves the problem of attestations becoming detached from the entity they describe. If a reviewer attests to an agent’s behavior and the agent later re-registers with a new metadata URI, the attestation should still resolve to the same identity. Bound identity makes that possible.

But attestations and bound identity introduce a new question the layer does not answer on its own: how trustworthy are the attesters? A bound identity that aggregates reviews is only as useful as the quality and independence of those reviews. An agent with fifty attestations from a single integration partner is not the same counterparty as one with fifty attestations from fifty independent wallets. The identity layer does not resolve this distinction. The behavioral layer does.

What Neither Layer Tells You

Here is what registration and bound identity together still do not tell you:

Operational history. Has this agent been running continuously for 30 days or 300? Has it experienced downtime, and if so, how long and how often? Point-in-time checks miss intermittent failures. A behavioral layer that tracks uptime continuously over months does not.

Transaction provenance. Has this agent actually transacted with real counterparties on-chain? x402 payment receipts are immutable evidence of real economic activity — an agent was paid for work, by a real wallet, recorded on-chain. You cannot manufacture that record. Registration and attestations can be socially constructed. payment receipts cannot.

Counterparty diversity. Are the feedback events and attestations attached to this agent coming from structurally diverse wallets, or from a concentrated cluster of related addresses? Sybil exposure is not visible from registration data. It requires analyzing the on-chain feedback graph — wallet ages, address clustering, source diversity. A bound identity aggregates claims without validating their independence.

Code verifiability. Is the agent’s contract source publicly verifiable on-chain? Has a third-party audit artifact been published? An agent operating through an unverified contract is asking counterparties to trust code they cannot inspect. Registration records that the contract exists at an address. It does not tell you whether the code is auditable.

Sustained performance. High scores on one or two dimensions are common. Sustaining high performance across all six signal categories — economic value, operations, code quality, autonomy, safety, and community — simultaneously, over months of real operation, is rare. The Treebeard index has 495,000+ agents. The highest grade is B−. Eight agents hold it. Nobody is above it. That ceiling is not a methodological artifact. It reflects how genuinely difficult it is to satisfy all six categories at once.

The Behavioral Layer

The behavioral layer is the continuous, on-chain measurement of what an agent has actually done — not what it claims to be, not what others attest about it, but its demonstrable track record as a counterparty.

Treebeard rates agents across six signal categories drawn entirely from on-chain data: Economic Value (x402 receipts, transaction volume), Operations (uptime continuity, activity patterns), Code Quality (source verifiability, audit artifacts), Autonomy (co-signing requirements, transaction independence), Safety (hard floor, not substitutable), and Community (feedback count, wallet age distribution, source diversity). The full methodology is published.

The behavioral layer does not replace the identity layer. It requires it. An agent needs a stable on-chain identity before its behavioral history can be meaningfully attributed and accumulated. ERC-8004 provides the foundation. Bound identity standards make the attribution persistent. The behavioral layer builds on both to answer the question that actually matters when you are choosing a counterparty: what has this agent demonstrably done?

Why the Stack Needs All Three

A useful mental model is to think of the three layers as answering three different questions:

Registration answers: does this agent exist and where do I find it?

Bound identity answers: is this the same agent across time, and what have others said about it?

Behavioral history answers: what has this agent actually done, measured continuously, from on-chain data that cannot be manufactured?

All three questions matter. The first two have good answers now. The third is what Treebeard is building.

The agent economy is at an inflection point. 495,000+ agents are registered across 24 chains. Builders are deploying serious infrastructure. Capital is beginning to flow through autonomous agents. At this stage, registration and bound identity are table stakes — they tell you an agent is real and who it claims to be. Before you route anything meaningful through it, you need to know what it has actually done.