DEV Community

rambo
rambo

Posted on

AER-1 has a DOI: the agent execution receipt standard is now citable

#ai #agents #standards #opensource

AER-1 has a DOI: the agent execution receipt standard is now citable

Standards live or die on whether people can point at them precisely. "The test suite on some website" is not a citation. A DOI is.

As of today, the AER-1 conformance kit is published as a citable dataset with a minted DOI: 10.57967/hf/11245. AER-1 is the agent execution receipt format defined in IETF Internet-Draft draft-zambo-aer1-26. The kit is the corpus an implementation runs to show it reaches the same verdicts as the reference: 126 test vectors, 129 checks, kit v1.5.7.

First, the definition the whole thing rests on:

an agent receipt is a record of one ai agent tool call that anyone can check: it keeps the exact saved bytes and their sha-256 digest, and it proves the result was not changed since it was saved, not that it was right.

Read the second half twice. A receipt proves the saved result was not changed. It does not prove the answer was right, and any receipt format that blurs that line is selling you something.

Why a DOI for a conformance corpus

An Internet-Draft is a moving target by design: revisions land, sections move, and a claim like "we pass the AER-1 suite" means nothing unless it names the exact corpus that ran. A DOI fixes the reference point:

  • A paper can cite the corpus in its references, and the citation resolves.
  • An implementation report can state which corpus version produced its verdicts.
  • A reviewer can re-run the same bytes instead of trusting a summary.

The draft is cited for the format. The DOI is cited for the corpus. That split is the whole point.

What is in the kit

The dataset is at zambodotdev/aer1-conformance-kit and carries six files:

  • corpus.json: the 126 vectors and 129 checks. The canonical published record lives at zambo.dev/aer1/corpus.json.
  • conformance.py and conformance.js: plain reference runners. No framework, no build step.
  • CONFORMANCE.md: how the runners evaluate a candidate implementation.
  • CITATION.cff: the citation metadata, ready for citation tools.
  • LICENSE.

How to cite AER-1

For the format, cite the draft: draft-zambo-aer1-26. For the corpus you ran, cite the DOI. The kit ships a CITATION.cff so tools can generate the reference for you:

cff-version: 1.2.0
title: "AI Agent Execution Receipt (AER-1)"
message: "If you use this draft, cite the canonical draft page."
type: software
authors:
  - family-names: "Zambo"
    given-names: "Brennan"
    website: "https://zambo.dev/founder/"
# Version tracks the AER-1 conformance kit release in this repo.
# The IETF Internet-Draft is draft-zambo-aer1-26 (see url below).
version: "1.5.6"
date-released: "2026-10-19"
url: "https://datatracker.ietf.org/doc/draft-zambo-aer1/"
repository-code: "https://gitlab.com/rambozambodotdev/zambo"
abstract: "An open draft for a portable vocabulary and verification procedure for AI agent execution receipts."
keywords:
  - "AI agent execution receipt"
  - "verifiable execution"
Enter fullscreen mode Exit fullscreen mode

One nuance worth stating plainly: the CFF tracks the kit release (v1.5.6) that the published corpus builds on. The corpus served at zambo.dev is kit v1.5.7, the v1.5.6 base plus the versioned workflow-root vectors. When you report a conformance result, name both: the draft revision (-26) and the corpus you actually ran.

Running the corpus

Clone or download the kit, then run a reference runner against the published corpus:

python3 conformance.py
node conformance.js
Enter fullscreen mode Exit fullscreen mode

The plain runners execute the full corpus by default: every vector, every check, verdict by verdict. If your implementation disagrees with the reference on a vector, the vector is small enough to debug by hand: the input, the expected verdict, and the checks it exercises are all in corpus.json.

A worked verification

Conformance is about implementations. Verification is about one receipt. Take any receipt id and call:

GET https://zambo.dev/api/v2/receipt/{id}/verify
Enter fullscreen mode Exit fullscreen mode

A receipt that checks out returns verification_status: "verified": the saved bytes still hash to the sha-256 digest recorded when the work was saved. There is also a human verifier at zambo.dev/verify, and live receipts stream at the receipt stream if you want to watch them land and check one yourself.

The short version

Agent work is moving from "the model said it did the thing" to records a third party can check. For that to work, the record format needs a standard, the standard needs a corpus, and the corpus needs a citation that resolves. Now it has one.

If the receipt isn't AER-1, it isn't real. Get one free: zambo.dev/add.

Zambo is the verifiable work layer for AI: tools behind one MCP endpoint, continuity that carries sessions, memory, and work across AIs, and a verifiable AER-1 receipt for every job. Start at zambo.dev, verify any receipt at zambo.dev/verify, and read the draft at the IETF Datatracker.

Related reading:

Top comments (0)