The test nobody arranged
Your own test suite passing proves very little. Your friends passing it proves less. The only test that counts is the one nobody arranged: a stranger, with no warning, no help, and no reason to be kind, runs your format cold against a live record and publishes whatever they find.
On October 16, that test arrived by email, uninvited.
Who showed up
WhiteMagic Labs maintains CPR, published on PyPI as continuity-receipt: an open specification for hash-chained records of governed AI-agent tasks, following each task through decision, authority, execution, delivery, termination, and settlement. The maintainer is Lucas Bailey.
They had read AER-1, the open format underneath every Zambo receipt, currently an IETF Internet-Draft (draft-zambo-aer1-23). They called it the cleanest per-execution receipt design in this space. Then, instead of asking us for anything, they built a probe and ran it.
The six checks
The target was the example receipt printed in Section 3 of the draft itself, live on zambo.dev for anyone to open: zambo.dev/run/130da435-e157-498e-af90-605866a86a27
Their probe checked six things:
- Public URL resolution: the receipt resolves on the open web, no credentials.
- Id match: the record served is the record asked for.
- Strict UTF-8: the canonical bytes decode cleanly.
- SHA-256 recompute: hashing those bytes reproduces the published output hash.
- Byte-length match: the decoded length equals the published length, 520 bytes.
- Verification status: the public verifier reports the commitment matches.
Six checks. All pass.
The probe is not a screenshot in a slide deck. It ships in their public tooling and runs against any AER-1 receipt id, so anyone can repeat the whole thing against any receipt, not just the example: the probe, in the continuity-receipt repository.
I recomputed it myself
Trust is not the mechanism here, so I did not take their email's word for any of it. I pulled the same receipt's public bytes and recomputed the hash independently:
sha256:0492c18e09059aa2e30ba0864f6aada2ef08475a5b35944cbf2a9ce46d12b9f1
It matches the published output hash exactly. Their crosswalk document prints the same digest. Two independent computations, done on different sides of this, one number. That is what verification is supposed to look like.
The crosswalk, and what compose means
They went further than the probe. Their crosswalk maps AER-1 element by element into their task lifecycle chain.
This part matters more than the checks. Section 15 of AER-1 leaves decision semantics, authorization, and settlement bindings to adjacent specifications, on purpose. AER-1 records what an execution produced and seals that record. What a decision authorized, and how a task settles, belongs to the specification that owns the decision. Their note on it was simple: they would much rather compose than duplicate.
That is the layers staying in their lanes. A receipt format that tries to swallow the whole governed-task lifecycle becomes a platform. A receipt format that holds one line, and lets neighboring specs bind to it, becomes infrastructure other people can build on. WhiteMagic chose to build on it.
Their verification is now listed in the AER-1 registry, in the Cross-System Verification section, described as exactly what the evidence supports: an independent probe and a published crosswalk. Not an implementation. Not a conformance claim.
What this does not prove
The honest limits
The probe proves the saved record matches its hash and the verifier agrees. In plain terms: a receipt proves the saved result was not changed since it was saved. It does not prove the work described in the receipt was done well. It does not prove a reported result was true at the moment it was reported, and it does not prove who produced the result. Verification never upgrades a report into an observation, and their crosswalk states that limit about AER-1 correctly, which is part of why this verification counts.
Two more, for the record. AER-1 is an Internet-Draft, not a finalized standard. And WhiteMagic is one maintainer with a young spec; their first probe run against AER-1 was September 30, weeks before they wrote to us. Nobody manufactured this for an email. It is a small, careful, independent check that anyone can rerun, and that is precisely the kind that matters at this stage.
Run it yourself
The receipt is public. The probe is public. The draft is public: draft-zambo-aer1-23 on the IETF Datatracker.
Run their probe against the Section 3 example, then run it against any receipt of your own. The demo mints a real one in seconds, free: zambo.dev/demo/
If you find a check that fails, that is a finding, and findings like that are how the draft gets better. The last outside findings were adopted into draft 20 and credited by name. That door stays open.
Frequently asked questions
Has anyone independently verified an AER-1 receipt?
Yes. In October 2026, WhiteMagic Labs, the maintainer of the CPR continuity-receipt specification, ran its own probe against a live AER-1 receipt, with no warning and no help from Zambo. The probe ran six checks: public resolution, id match, strict UTF-8 decoding, SHA-256 recompute, byte-length match, and the public verifier status. All six passed. The probe ships in their public tooling, so anyone can rerun it against any AER-1 receipt.
What did the WhiteMagic probe check, exactly?
The target was the example receipt printed in Section 3 of the draft, receipt 130da435-e157-498e-af90-605866a86a27, live at zambo.dev. The probe fetched the receipt's public bytes, decoded them as strict UTF-8, recomputed the SHA-256 digest, and compared it with the published output hash, sha256:0492c18e09059aa2e30ba0864f6aada2ef08475a5b35944cbf2a9ce46d12b9f1, and the published length, 520 bytes. The digest and the length matched exactly, and the public verifier reported that the commitment matches.
What is the crosswalk between AER-1 and CPR?
WhiteMagic published a crosswalk that maps AER-1 element by element into the CPR task lifecycle chain: decision, authority, execution, delivery, termination, settlement. AER-1 records what an execution produced and seals that record. What a decision authorized, and how a task settles, belongs to the specification that owns the decision; Section 15 of the draft leaves those bindings to adjacent specifications on purpose. The crosswalk is the two formats composing instead of one format trying to swallow the other's job.
Does the probe prove the work in the receipt was done correctly?
No. The probe proves the saved record matches its hash and that the verifier agrees: the record has not changed since it was saved. It does not prove the work described in the receipt was done well, and it does not prove a reported result was true at the moment it was reported. Verification never upgrades a report into an observation. The claim a receipt supports is narrow, and stating the narrow claim plainly is what makes it worth anything.
Can I run the same checks myself?
Yes. The receipt is public, the probe is public in the continuity-receipt repository, and the draft is public on the IETF Datatracker as draft-zambo-aer1-23. You can also mint a fresh receipt with the free live demo at zambo.dev/demo/ and check it the same way. If you find a check that fails, that is a finding, and outside findings are how the draft gets better: the last ones were adopted into revision 20 and credited by name.
Your AI said "done." For $0.99, prove it: https://zambo.dev/day-pass/