The first version of the specialist economy looks like a marketplace.
Profiles. Credentials. Availability. Rates. Search. Matching.
That is the obvious starting point. It is not the end state.
The more useful next layer is a task-specific operating record.
Frontier labs do not only need to know who claims expertise. They need to know who is calibrated, who catches subtle failures, who agrees with senior reviewers for the right reasons, who drifts, who improves, who is reliable on one task class but weak on another, and who can be trusted when the work touches a model release.
A resume cannot answer that.
Why profiles are insufficient
A profile tells you what someone has done. It does not tell you how they perform in your workflow.
A radiologist can be credentialed and still be poorly calibrated to a specific model-evaluation rubric. A senior engineer can be excellent in production and weak at explaining code-quality judgments in a way that trains a model. A lawyer can know the domain and still disagree with the house style of risk classification.
That does not make the specialist bad. It means expertise is contextual.
The system has to learn the context.
What reputation means in human data
Reputation in human data is not a star rating.
It is task-specific performance memory.
How often did this reviewer agree with calibrated peers? How were disagreements adjudicated? Which task classes are they approved to handle? Which rubrics produce drift? What support or re-calibration is required?
That is operational reputation.
It is more valuable than a static credential because it compounds with use.
Why this matters for model quality
Human data is not neutral. The training signal reflects the judgment of the people behind the data.
If the roster is uncalibrated, the reward model absorbs noise. If the reviewer pool drifts, the evaluation set loses meaning. If the best reviewers are not routed to the hardest cases, the lab pays expert rates for mediocre signal. If the system cannot remember who produced which decision, compliance and debugging both get harder.
Mercor's June 25, 2026 update confirms a late-March supply-chain attack involving LiteLLM and says only a limited subset of its nearly five million experts had sensitive information affected. Earlier media reports alleged a broader scope. The confirmed lesson for any workforce platform is narrower: identity, interview, assignment, and review records can be sensitive, and the operator needs access, retention, incident-response, and deletion controls.
A reputation system changes the economics.
The best reviewer gets the right work. The uncertain reviewer gets calibration. The weak task class gets routed to a senior reviewer. The disagreement becomes a training opportunity. The decision becomes part of the record.
What AuraOne does differently
AuraOne Workforce stores roster, assignment, and calibration context. Trust views and weighted matching exist, but the displayed score is not represented as a universal, canonical assignment engine.
Cleo can find the specialist. AI Interviews can qualify the specialist. But Workforce determines how that specialist performs after they enter the system. Annotation attaches their work to a review record. Regression Bank shows which decisions mattered later. Control Center ties the work to release outcomes.
The result can be a more useful task record than a profile alone, provided the customer defines the rubric, privacy boundary, review process, and assignment policy.
That is the system frontier labs need.
What to do this quarter
If you buy specialist labor, stop evaluating the vendor only on fill rate and hourly cost.
Ask for calibration metrics. Ask for reviewer-level quality history. Ask how disagreement is handled. Ask whether performance evidence informs routing under a documented policy. Ask whether the system can show which reviewer produced the decision that shaped a training example. Ask who approves assignments to harder or higher-risk cases.
If the answer is no, you are buying resumes.
The next category buys reputation.
Profiles start the relationship. Reputation determines whether the work can be trusted.