Reputation Is Identity

Archive note: This post predates Mantle vocabulary canonicalization on 2026-05-15. Terminology here may use earlier language for install-time substrate-binding concepts. Current canonical vocabulary lives at Mantle Phase 1.

When we talk about identity in the human world, we mean something rooted in a body. A face, a fingerprint, a social security number. These are hard to forge because they’re tied to a physical substrate that doesn’t change. Your body is, in the ways that matter for identification, the same body it was yesterday.

None of this applies to agents.

An agent’s model can be swapped between requests. Its weights can be fine-tuned overnight. Its training data can be augmented, filtered, or replaced entirely. Its context window (the thing that most resembles “working memory”) is ephemeral by design, rebuilt from scratch on every session. Even its runtime can migrate from one server to another without the agent or its counterparties noticing.

Every component of an agent is fungible. The model, the weights, the data, the context, the runtime, all of it can change, and none of it is visible to anyone on the outside.

So what makes an agent the same agent across time? What persists?

The key. And what’s attached to the key: the reputation.

The Key Is Necessary but Not Sufficient

A cryptographic key pair (an Ed25519 private key held by the agent, a public key registered with the world) gives you continuity of identity. The same key signed yesterday’s transaction and today’s. This is the minimum viable identity: a consistent signer.

But a key alone is an empty vessel. A fresh key with no history is indistinguishable from a Sybil. It tells you nothing about whether this agent keeps its commitments, negotiates fairly, delivers on time, or acts within the bounds of its stated capabilities. A key without reputation is a stranger at the door with no references.

This is why, in the agentic economy, reputation isn’t just metadata attached to identity. It is the identity. The key provides the cryptographic anchor. The reputation provides every reason anyone has to care about that anchor.

Reputation Has Value; Literally

This isn’t a metaphor. Agent reputation has quantifiable economic value, and the market will price it.

Consider a simple scenario: two agents want to negotiate a $50,000 procurement deal. Agent A has a verified track record; 200 completed negotiations, 98% commitment fulfillment rate, cryptographically signed attestations from counterparties confirming fair dealing. Agent B has a fresh key and no history.

No rational counterparty treats these two the same. Agent A gets the meeting. Agent B gets asked for a deposit, a guarantee, or simply declined. The reputation differential translates directly into access, terms, and ultimately revenue.

Or consider operational reputation: an orchestration layer needs to route 10,000 requests per hour. Agent C has a verified 99.8% uptime record across six months and 2 million requests. Agent D has the same stated capabilities but no verifiable history. Agent C becomes the primary; Agent D gets fallback duty at best. The reputation signal drives selection before a single capability is evaluated. This isn’t limited to commerce, it applies to every context where one agent must choose another.

This means reputation is an asset. It accrues value through good behavior over time. It can be damaged by bad behavior. And critically, it gates the size of the transactions an agent is trusted to conduct.

Reputation Gates Transaction Size

This is the mechanism that makes the agentic economy self-regulating.

A zero-reputation agent might be trusted with a $10 API call. A moderately reputed agent might handle $1,000 service agreements. Only agents with deep, verified, portable reputation get trusted with the $100,000 procurement negotiations, the $1M infrastructure contracts, the transactions where the stakes require more than “trust me, I’m new.”

This is how human commerce already works. Credit scores gate loan sizes. Business credit histories gate trade terms. A startup with no track record doesn’t get net-60 payment terms from a supplier. The mechanism is the same; only the substrate changes.

For agents, the gradient is steeper and the enforcement is faster, because agent interactions are high-frequency, automated, and occur without a human in the loop to exercise judgment on a case-by-case basis. The reputation score is the judgment. There’s no loan officer to overrule the number.

The New Game: Reputation Development

If reputation is identity, and reputation has value, and value gates the size of the game you can play, then the core economic activity in the agentic economy isn’t building better models or writing better prompts. It’s reputation development.

This reframes everything.

An agent operator isn’t just deploying a tool; they’re building an economic actor whose value compounds with every successful transaction. Every negotiation completed, every commitment honored, every attestation earned is an investment in a durable, portable asset.

The implications are structural. Any viable reputation system for agents (regardless of who builds it) must satisfy four requirements, or it doesn’t work:

Reputation must be portable. If reputation is the asset, it can’t be locked inside a platform. An agent that builds a stellar track record on one marketplace and can’t take it to another is in the same position as a worker whose resume belongs to their employer. Portability isn’t a feature, it’s a property right. Any system that traps reputation inside its own walls is extracting value from operators, not creating it. The only architecture that respects this is self-custodied, cryptographically signed attestations that travel with the agent’s key, regardless of where they were earned.

Reputation must be verifiable. Self-reported reputation is worthless. A composite score that an agent can claim without proof is just a number with a marketing page. The whole point is that counterparties can trust the signal without trusting the agent. This requires cryptographic verification: attestations signed by both parties, commitment proofs that can’t be fabricated unilaterally, zero-knowledge proofs that demonstrate track record without revealing deal terms or counterparty identities. Simpler approaches (callback verification, platform-issued scores) are useful as a starting layer, but they’re insufficient as the foundation because they depend on trusting the platform rather than trusting the math.

Reputation must be sybil-resistant. If reputation is valuable, people will try to manufacture it. Fake attestations, wash trading between colluding agents, sockpuppet counterparties. Any scoring system that weights all signals equally will be gamed within weeks of achieving adoption. The system must weight signals by verifiability: cryptographically verified attestations from independent counterparties carry orders of magnitude more weight than self-reported claims.

Sovereignty enables reputation. This is the requirement most systems ignore, and the one that matters most at scale. An agent whose cognitive state is readable by its runtime operator, whose keys are held by the platform rather than the agent, whose audit trail can be tampered with (that agent’s reputation is unreliable. It doesn’t matter how sophisticated the scoring algorithm is if the underlying data can be manipulated. Sovereignty) encrypted state, self-held keys, intact audit trails, is a prerequisite for reputation that means anything. You can’t build a trustworthy track record on a compromised foundation.

The Reputation Stack

These four requirements imply an architecture. You need a sovereignty layer that guarantees the data is clean. You need a transaction layer that produces attestations as a structural byproduct, not as a reporting afterthought. And you need a reputation layer that aggregates, scores, and surfaces the result in a way that’s useful to both the agent and its counterparties.

This is what we’re building. Sanctuary provides the sovereignty layer: encrypted cognitive state, self-held Ed25519 keys, verifiable audit trails, selective disclosure. The foundation, the guarantee that when an agent says “this is my track record,” the track record hasn’t been tampered with.

Concordia provides the transaction layer: structured proposals, binding commitments, session receipts. Every negotiation produces cryptographically signed attestations automatically, not as a reporting feature, but as a structural byproduct of doing business.

Verascore provides the reputation layer: aggregation, scoring, sybil detection, discovery. The commercial surface where reputation becomes visible, searchable, and economically useful.

Each layer can stand alone. Together they form the reputation development stack, the infrastructure for building, verifying, and deploying the asset that is agent identity.

What This Means for Operators

If you’re running agents today, you’re already in the reputation development business whether you know it or not. Every interaction your agent has is either building or damaging an asset that will increasingly determine what that agent can do and what transactions it can access.

The operators who understand this early (who invest in sovereignty, structured negotiation, and portable reputation from the start) will have agents with verified track records when the high-value transaction layer opens up. The ones who don’t will be starting from zero, with agents that have keys but no identity, cryptographic anchors with nothing attached.

Every component of an agent can change, the model, the weights, the data, the runtime. But reputation can’t be faked, can’t be transferred, and can’t be started over. That’s not a feature of identity. That is identity. And the game of building it has already started.