Claim only what was observed.
A WorkProof is evidence from a defined simulation. It is not identity verification, a degree, background check or guarantee of future performance.
MANDORKA WORKPROOF STANDARD / PUBLIC DRAFT
Proof must survive scrutiny.
Mandorka is building a public, inspectable specification for AI-assisted work evidence. A score without versioning, integrity controls and canonical verification is not enough.
Protocol v2Immutable assessment versionsServer-side scoringMachine-readable verificationNo fake percentile
CORE RULES
Seven rules every Mandorka proof must obey.
Never rewrite history.
Every result keeps the assessment version that produced it. New scoring rules create a new version instead of changing old records.
Answer keys stay server-side.
Public challenge payloads contain randomized identifiers. Correct-answer mappings are never sent to the browser.
One session. One completion.
Assessments use short-lived, single-use sessions with replay rejection and human verification at protected write surfaces.
Explain the dimensions.
The 100-point score is decomposed into Accuracy 30, Judgment 25, Instruction following 20, Verification 15 and Communication 10.
Canonical record beats screenshots.
Every current WorkProof exposes a live URL, verification JSON and SHA-256 fingerprint derived from the stored result.
No scientific claim before evidence.
Mandorka will not claim predictive validity, accreditation or population percentile until study quality and sample size justify it.
EMPLOYER USE
Use the signal. Do not worship the number.
A WorkProof should reduce uncertainty, not replace lawful human judgment. Employers should use role-relevant tracks, inspect dimensions, compare versions and combine the result with other job-relevant evidence.
| Control | Mandorka requirement | Employer interpretation |
|---|---|---|
| Version | Exact assessment version is retained. | Do not compare materially different versions as if identical. |
| Integrity | Current proofs record session integrity metadata. | Legacy proof stays visibly legacy; newer controls must not be invented retroactively. |
| Dimensions | Five dimension scores remain visible. | Use dimension gaps to drive targeted follow-up. |
| Thresholds | Mandorka does not publish universal employment pass/fail thresholds. | Any threshold must be job-relevant, validated for the use case and applied consistently. |
STATUS
A standard earns authority through evidence.
Current status: public Mandorka specification, not external accreditation. The validation program exists to test whether the measured signal relates to real work quality before stronger claims are made.