Apply
A person creates a private account. The public receives a pseudonymous reviewer identity, not the reviewer’s email.
HUMAN ATTESTATION FOR AI-ASSISTED WORK
AI produces.
Humans attest.
Attested by Humans records who examined an output, under what expertise, against what criteria, using what evidence, and what they concluded.
01 / WHAT IT MEANS
Not a claim that a human made it.
A record that a human checked it.
The platform does not try to prove that something is “AI-free.” It makes human scrutiny inspectable: review scope, reviewer expertise, conflict declaration, evidence, uncertainty, verdict and limitations.
02 / HOW IT WORKS
Control before publication.
Establish scope
The reviewer declares domains and specialties. Expertise can remain self-declared, carry submitted evidence, or be independently verified.
Review
The reviewer identifies the subject, defines what was checked, declares conflicts, records evidence, gives a verdict and states confidence.
Moderate
New reviewers do not publish directly. Attestations enter moderation; reviewer accounts and expertise claims can be approved, rejected or suspended separately.
Publish
The public record shows the attestation, pseudonym, relevant expertise status and stated review basis without exposing private identity data.
Build trust
Only sustained, high-quality participation can later justify a trusted-reviewer tier. Trust is earned per history, not granted by signup.
03 / CONTROL MODEL
Approve people. Verify expertise. Moderate claims.
PRIVATE ACCOUNT
→PUBLIC PSEUDONYM
→EXPERTISE CLAIMS
→ATTESTATION
→MODERATION
→PUBLIC RECORD
A reviewer is not certified as universally trustworthy. A specific account is approved to participate; specific expertise claims have their own evidence status; each attestation remains a scoped judgment that can be challenged or withdrawn.
04 / REVIEW DIMENSIONS
Assess only what is actually examined.
Accuracy
Are factual claims correct and appropriately qualified?
Evidence
Are claims supported by credible and traceable sources?
Reasoning
Do assumptions, logic and conclusions hold together?
Safety & compliance
Are material risks, rules or constraints handled appropriately?
Transparency
Are uncertainty, limitations and relevant AI involvement made clear?
Fitness for purpose
Is the output suitable for the decision or use the review actually covers?
05 / REVIEWER GUIDE
A useful attestation is narrow, evidenced and falsifiable.
1
Define the object. Name the report, code, analysis, design, dataset or other output being examined and link to it where possible.
2
State the scope. Say what you checked and what you did not. “I reviewed the cost assumptions and arithmetic” is stronger than “I reviewed this model.”
3
Declare your basis. Add relevant expertise and disclose any relationship, collaboration, competitive interest or other conflict.
4
Show evidence. Reference sources, calculations, standards, observations or tests that materially support the judgment.
5
Separate finding from confidence. A strong verdict with weak evidence should not masquerade as certainty.
6
Make revision possible. State major concerns and what evidence or change would alter your assessment.
06 / PRINCIPLE
Attested ≠ correct forever.
An attestation is a timestamped human judgment within a declared scope. It does not imply authorship, legal certification, infallibility, complete verification, or approval outside that scope.
07 / QUESTIONS
What the platform does — and does not — claim.
Who can become a reviewer?
Anyone may apply. Applications are not automatically activated. Participation is controlled separately from expertise verification.
Are reviewers anonymous?
Publicly, reviewers use pseudonymous identifiers. The service retains private account information for access control. The pilot does not promise cryptographic anonymity from the platform operator.
How is expertise verified?
Claims begin as self-declared. Evidence can be submitted and an administrator can mark a claim verified or established. The exact verification policy will evolve during the pilot and is shown separately from account approval.
Can a reviewer assess something outside verified expertise?
Yes, but the platform should not present that review as in-domain expert verification. The expertise status remains visible so readers can weight the judgment appropriately.
Does Attested by Humans certify that something is true or safe?
No. It records a human assessment and its scope. It is not a regulator, conformity-assessment body, legal opinion, safety certification or guarantee.
Can a published attestation be removed?
It can be withdrawn or rejected if moderation, conflicts, abuse, evidence problems or reviewer status justify doing so. The pilot currently exposes the current publication state rather than a full immutable revision ledger.
Why no single universal score?
A single number creates false precision across unlike subjects. The pilot prioritizes verdict, confidence, expertise, scope, evidence and disagreement. Scores may be added later only where a rubric makes them meaningful.
08 / VERSION
Pilot v0.2.0
Introduces controlled reviewer accounts, pseudonyms, expertise claims, structured attestations, moderation, public attestation browsing and an admin workflow. This is an experimental trust infrastructure, not a certification service.