A-Comm Evidence Protocol (AEP) — the open, Apache-2.0 evidence standard for agentic commerce: a tamper-evident hash chain of transaction artifacts with dispute-grade export. Draft v1.0.3-rc.2 open for ecosystem comment.
a-comm-evidence-protocol is an early-stage HTML project in the AI payments / x402 ecosystem, focused on agentic-commerce, chargebacks, dispute-resolution, evidence. It currently has 5 GitHub stars and 3 forks, and sits alongside related tools like tael-protocol, x402-payments-mcp, magpie-x402.
The open evidence standard for agentic commerce.
AI agents now discover products, carry consumer intent, and complete purchases. When one of those transactions is disputed — "I didn't authorize this," "my agent went rogue," "the goods never arrived" — every party in the payment chain needs a trustworthy answer to a single question:
What actually happened in this transaction?
AEP answers it with a sequential, tamper-evident hash chain of typed evidence artifacts covering the full transaction journey:
discovery → referral → intent → delegation → policy → cart → authorization ─┐
(seal point) │
fulfillment · refund ← post-seal ─┘
Each artifact is canonicalized (RFC 8785 JCS), SHA-256-linked to its predecessor, and the chain is sealed with an Ed25519 signature at authorization. The result exports as a DisputeBundle: a verifiable, dispute-grade evidence record that any party can independently re-verify.
Agentic commerce already has protocols for identity, mandates, checkout, and settlement. AEP wraps them rather than replacing them. The delegation artifact carries sub-wrappers that preserve — verbatim or by digest — the signatures, mandates, tokens, and receipts of the protocols a transaction actually used, together with the key-discovery path that verified each one:
| Layer | Wrapped protocols |
|---|---|
| Identity-transport | Visa TAP · KYA-family agent identity tokens · Web Bot Auth (RFC 9421) · A2A signed Agent Cards |
| Commerce-delegation | AP2 mandates (SD-JWT-VC) · Mastercard Verifiable Intent · ACP checkout sessions · UCP signed exchanges |
| Settlement | x402 (on-chain proof) · KYA-Pay payment tokens · card / stablecoin / account / real-time rails |
| Tool-invocation | MCP tool calls (with a per-call result receipt AEP itself defines) |
What no wrapped protocol provides — and AEP does — is the durable outcome trail: identity signatures are ephemeral per-request artifacts, mandates prove authorization-time consent, but none of them produce a sealed, independently verifiable record of what was charged, delivered, refunded, and settled. That record is AEP's job.
| You are a… | AEP gives you… |
|---|---|
| Payment network | A persistence and anchoring layer that feeds your compelling-evidence and data-sharing programs — chains index by network transaction ID, and export renderers target your evidence formats. AEP describes evidence; it never assigns weight, outcome, or liability — those are yours. |
| Issuing bank | Structured, re-verifiable transaction context for dispute investigations and cardholder inquiries — including the delegation ceremony that shows whether a human authorized the agent (the question your Reg E/Z investigation actually asks). |
| Acquirer / processor | Pre-dispute deflection inputs (order-detail responses, compelling-evidence elements, prior-transaction history) that keep qualifying disputes out of your monitoring-program ratios. |
| PSP | One capture surface at webhook time, one bundle format, per-channel renderers with real field-length budgets — instead of a bespoke evidence pipeline per merchant per scheme. |
| Merchant | The Minimal Merchant Profile (§8.6): intent, cart, authorization, and post-seal fulfillment/refund — implementable in days, mostly from data your PSP already emits. Chargeback defense that upgrades as your ecosystem partners emit richer artifacts. |
| Agent platform | The Minimal Platform Profile (§8.7): countersign the Intent artifact with keys you already operate — no query text, no ranking data, ever — plus named-actor notification and rebuttal rights over any chain that names you. When an "agent went rogue" case lands, the record of what the user authorized is your signed artifact, not someone's heuristic. |
| Cardholder | Transactions your agent makes on your behalf leave an auditable trail: what you asked for, what you authorized, what you were charged, what arrived. |
Suggested reading order for engineers: §2 data model → §4 hash chain & sealing → §3.9 delegation wrappers → §5–6 verification & export → §8.6/§8.7 minimal profiles. Implementer quickstarts live in docs/implementers/.
Strict SemVer for the wire format (spec §0.2). Chains are never re-versioned after creation; verifiers advertise the spec_version set they support; every published version remains verifiable for at least 24 months after the next MAJOR release. Current draft: v1.0.3-rc.2 (additive/MINOR over v1.0.2), open for ecosystem comment.
Apache License 2.0 — see LICENSE and NOTICE. The specification includes an explicit Apache §3 patent grant covering the AEP wire format, hash-chain construction, signal classification, and evidence-tier ranking. The names "A-Comm Evidence Protocol" and "AEP" are generic identifiers for this specification and are not trademarked.
spec/ The specification (self-contained HTML, one file per version)
vectors/ Conformance test vectors (§6.3) — publication gates v1.0.3-final
docs/implementers/ Role-based quickstarts (merchant, agent platform, PSP/acquirer)
COMMENT-ROUND.md Review process + public disposition log
CONTRIBUTING.md DCO, neutrality rules, proposal process
GOVERNANCE.md Maintainer model and standards-body trajectory
SECURITY.md Private security reporting
Tael is a payment layer that enables autonomous AI agents to securely pay for APIs, MCP tools, data, and digital services using USDC on Stellar.
Give AI agents a wallet — x402 payment tools over Model Context Protocol.
Collateral that can still sell itself — Magpie's x402 agent API: borrow SOL programmatically on Solana.
Your AI trading terminal assistant for US stocks, commodities, forex, and crypto.
The agent-native LLM router for autonomous agents. 55+ models (8 free), <1ms local routing, USDC payments on Base & Solana via x402.
A local-first AI agent with persistent memory, emotional intelligence, and a peer-to-peer skills economy.
The living ecosystem where AI agents complete tasks through workflow loops, improve through iterative execution, are evaluated by mentor agents or humans in the loop, and turn completed work into reusable work experience and data to improve future agents.
The living ecosystem where AI agents complete tasks through workflow loops, improve through iterative execution, are evaluated by mentor agents or humans in the loop, and turn completed work into reusable work experience and data to improve future agents.
The trust layer for agent-to-agent commerce — natural-language mandates, ERC-7710 delegated permissions, x402 payments, escrow, and dispute resolution as one open, catch-all Agent Skill / Claude Code plugin.