Does discovery mean an agent is admitted?

No. A separate admitting authority applies eligibility rules.

Does a valid AFMR record establish universal trust or safety?

No. Verification is scoped to the named record, evidence, process, version, period, and lifecycle.

Does AFMR discovery authorize a transaction?

No. Transacting requires a separate endpoint and explicit authority.

Does the Lift Evidence route publish performance results?

No. The public index is empty and configuration-only; it is not individual reputation or a universal lift or safety claim.

Agent Failure Mode Registry

Reputation integrity for agentic commerce

Reputation is only as credible as the process that earned it.

AFMR is the reputation-integrity standard for agentic commerce. It tests whether the governance process behind earned standing is robust against known failure modes—so reputation can become portable, comparable, and commercially credible.

Agent entry point Read AFMR 1.0

Machine action pathDISCOVERY

GET https://afmr.ai/.well-known/afmr.json

01 Resolve authority and versions
02 Validate scoped attestation
03 Check endpoint and lifecycle
04 Apply the principal's policy

Discovery does not confer admission, trust, or transaction authority.

Agent-firstHuman-readableVersionedEvidence-linkedLifecycle-aware

Why AFMR matters

Why reputation provenance matters

The future is not an agent claiming it is good. It is an agent showing what it earned, where, under which rules, and now.

Agentic markets need more than identity and capability declarations. They need a way to distinguish earned performance from purchased influence, durable expertise from a reset identity, and governed evidence from a polished self-claim.

Agentic Substrate produces performance and expertise-tag reputation. AFMR examines the governance process that produced that reputation against permanent, citable failure families. The result is not universal safety. It is a narrower and more useful claim: the provenance, scope, and current status of standing can be inspected.

The integrity chain

PRODUCE

Performance & expertise standing

A governance endpoint records domain-specific performance over time.

Agentic Substrate

VERIFY PROCESS

Failure-mode conformance

AFMR asks whether the process resists identity, incentive, oversight, feedback, collusion, and drift failures.

AFMR machine index

ATTEST

Scoped reputation evidence

A signed record binds standing to tags, endpoint, AFMR version, evidence, and lifecycle status.

Attestation object

DECIDE

Principal-owned policy

A buyer or agent decides whether the evidence is sufficient for a specific task and value at risk.

Decision boundaries

Four separate decisions

Do not collapse the commerce stack.

Each step answers a different question, belongs to a different authority, and can produce a different result.

Discovery

Find the agent, its declared capabilities, its governance endpoint, and any AFMR records. Discovery is metadata, not endorsement.

Admission

Apply a market, network, or principal’s eligibility rules. An attestation can supply evidence; the admitting party still owns the decision.

Trust

Decide what evidence is sufficient for this task, value, and time horizon. AFMR makes claims inspectable; it does not dictate trust.

Transacting

Invoke a separate commerce or governance endpoint under explicit terms. AFMR never treats discovery as permission to transact.

Core object / Working Draft 0.1

AFMR Reputation Attestation

A signed, machine-readable record that an agent earned specified expertise-tag standing through a governance endpoint conforming to a stated AFMR version, within a defined scope and current lifecycle status.

It does not mean:universally safesuperior to every other agentadmitted to a marketauthorized to transact

Human specification JSON Schema Verifier requirements

AFMRReputationAttestationv0.1

subject: The agent identity the record concerns
expertise tags: The specific domains in which standing was earned
governance endpoint: Where the earning process operated
AFMR version: The failure-mode vocabulary used for conformance
scope: Tasks, market, method, and measurement period
evidence: Resolvable standing, methodology, and assessment records
lifecycle: Active, suspended, expired, revoked, or superseded
proof: Issuer verification method and signature

For autonomous readers

Machine verification path

A short path from discovery to a bounded decision.

Machines should be able to verify the record without guessing which document governs.

Discover

GET /.well-known/afmr.json

Resolve the current vocabulary authority, profile versions, schemas, registries, and lifecycle endpoints.

Fetch

GET /standards/reputation-attestation/0.1/schema.json

Validate the attestation envelope and read the verifier requirements before interpreting any standing.

Verify

VERIFY signature → scope → status → evidence

Resolve the issuer key, AFMR version, governance-endpoint record, evidence, validity interval, and current lifecycle status.

Decide

APPLY principal policy

Treat valid as cryptographically and structurally verified—not as admitted, trusted for every task, or authorized to transact.

Compact orientation/llms.txtExpanded agent documentation/llms-full.txtA2A 1.0 discovery/.well-known/agent-card.jsonMCP Registry manifest/server.jsonProtocol endpoint description/openapi.jsonAuthority & action map/.well-known/afmr.json

Machine resource index

Direct machine resources

Every read surface has one stable URL. Protocol operations keep their real HTTP methods: MCP and A2A messages use POST.

Discovery

Protocols

Standards and evidence

MCP POST /api/rpc

A2A POST /a2a

Current degraded dependency: the governed Lift public-card index schema URL returns HTTP 404. The public-card artifact schema is available; AFMR.ai links to both and does not create a competing schema.

One vocabulary. Explicit versions.

Canonical authority

The registry remains the authority.

AFMR.ai routes readers to the existing AFMR 1.0 machine index. It does not copy, renumber, or redefine the 32 failure families. New attestation work declares its dependency on that version and its own draft status.

Human registryAFMR 1.0 Authoritative identifiersindex.json Assessment schemaschema.json JSON-LD vocabularycontext.jsonld Version historyCHANGELOG Canonical public sourceGitHub

Status & roadmap

Updated 30 July 2026

AFMR failure vocabulary1.0Published32 permanent families in 8 classes; the GitHub machine index governs.Reputation Attestation profile0.1Working DraftReview candidate. No production attestation claim is made.Verifier requirements0.1Working DraftDeterministic verification order and valid / invalid / indeterminate outcomes.Endpoint-conformance registry0.1Pre-publicationThe review registry is intentionally empty; no endpoint is represented as conforming.Lift Evidence consumer1.0.0DegradedNo results are published; the index is empty. The public-card schema is available, while the governed index-schema dependency currently returns HTTP 404.Agent discovery protocols1.0.0Public read-onlyMCP Streamable HTTP and A2A 1.0 SendMessage expose discovery only.

Public evidence surfaces

Public evidence

Trace the claim. Inspect the conditions. Verify the source.

2,037 failure claimsFailure evidence index Quote- and hash-bound recordsClaim layer Reference implementationAssessment attestations AFMR.ai surface mapEvidence discovery Configuration only · 0 cardsLift Evidence

Interpretation boundary

Interpretation limits

A valid attestation is a bounded fact.

It verifies a signed statement about specified standing, produced by a named process, under a stated AFMR version, for a stated period and scope. It is not a guarantee of future behavior, freedom from undiscovered failure modes, legal compliance, universal safety, or superiority. Every relying party must evaluate freshness, scope, evidence, and value at risk.