Source the specialist
Engineers, clinicians, lawyers, scientists — sourced against the work your queue is producing, not a generic talent pool. The bench grows where the field is short.
→The data only people can give — from specialists who passed your rubric, signed for their work, and never share an identity or a dataset with anyone else's.
What credentialed specialists now earn to author tasks, rubrics, and reasoning traces.
Credential-checked and on one contract. Cleared on the rubric your team already runs.
When the EU AI Act asks where your training data came from. Every datapoint already carries the answer.
Source the specialist in your field. Verify and vet them on your rubric. Put them on your queue — a calibrated bench is on the work in weeks, not a hiring cycle.
Engineers, clinicians, lawyers, scientists — sourced against the work your queue is producing, not a generic talent pool. The bench grows where the field is short.
→Identity-verified, credential-checked, and cleared on the rubric your team already runs. Trial cases, reviewer notes, and calibration records sit on file.
→One contract. One payout record. Assigned to annotation, review, and approvals — under the same rubric your model is graded on, with their work signed.
Five stages, the same on every program — annotation, review, approvals, and expert tasks. The work comes in. The proof comes out.
Annotation, review, approvals, and specialist tasks enter one place with the rubric, the rights, and the owner attached. Nothing runs in a side channel nobody can audit later.
Field, calibration, quality history, and current load decide who takes the work next. The reason a name surfaced stays on the task, so the match holds up under review.
Week and month calendars handle shift assignment, scoped budget bands, and conflict detection. The commercial terms stay tied to the shift record.
When a queue opens a gap, the bench recruits against the actual work — vetted specialists in the field that's short, with the reason on the record. Not a generic pool, and not a name from a directory.
Completed tasks update coverage, quality records, and the terms on file. The bench keeps a record of how it gets better — and which failures it should never repeat.
Filter by calibration, trust tier, availability window, or field. The roster surfaces specialists you can put on your queue — not a directory of every name on file.
Names anonymized. Example bench view from the platform workforce directory.
Coverage is read against the actual work — which field is short, on which queue, and by how much. When a gap crosses the line, the bench recruits vetted specialists in that field, with the reason on the record.
No generic talent pool, no name pulled from a directory. The bench grows where your work needs it, and every new specialist arrives identity-verified and signed.
A specialist's tier is not a label. It is the result of calibration, review, incidents, and field clearances — every change reviewable, every change reversible.
First calibration on the rubric the team runs. Trial set scored, reviewer notes filed, tier opened.
Active queue load with reviewer agreement above the floor. The bench treats this as a trusted contributor.
Reopen on a single span. Coached, not penalised. The reason and the resolution stay on the specialist's record.
New calibration activated. Specialist now eligible for clinical NLP queues. Review record updated.
The Roster shows who's on the bench. Detail tells you how they got there. The Scheduler is where the shift gets booked. Coverage is where the next specialist comes from.
The full roster, filterable by field, certification, availability, and trust tier — with coverage and a one-click way to put specialists on a queue.
Identity and credential checks, certification history, earnings, quality trend, active assignments, and the calibration record from your rubric.
Week and month calendars for shift assignment, budget bands, and conflict detection — the commercial terms tied to the shift record.
When a queue opens a gap, the bench recruits against the actual work — vetted specialists in the field that's short, every step on the record before it runs.
The bench is the artifact. Who's on it, what they cleared, what they reviewed, and what they were paid — all on the record. Specialists recreate realistic tasks from clean-room scenarios, so no employer data is ever uploaded.
Every specialist on the bench, identity-verified, the rubrics they're cleared on, and the queues they're on.
Every cleared case, reviewer note, and overturned call — attached to the specialist who made it, with the rights it was done under.
Who passed which rubric, on what date, against which trial set. The record a reviewer — or an auditor — can inspect.
One contract per specialist. One ledger per queue. The money trail matches the work trail.
“We needed experts in our field, verified, whose work would survive an audit. We got a bench on one contract, never pooled, every datapoint carrying who made it and the rights it was done under.”
Six questions, six answers. The same six come up across annotation, review, approvals, and specialist execution.
Every specialist is identity-verified and credential-checked, on one contract. Your data is isolated to your program — never pooled with anyone else's. The thing a competitor skipped the year it lost four terabytes, including who its workers were.
One contract per specialist, one payout record per queue. The money trail matches the work trail. Every cleared task is auditable against the terms it cleared under.
Yours. Specialists are vetted on the rubric your team already runs. Trial cases, reviewer notes, and calibration records sit on file. Recalibration follows the same rubric version.
One contract per specialist, with the terms tied to the shift record. Expert pay is quoted by field and seniority; you see the scoped band before you book.
Yes. Every datapoint carries a chain of consent and provenance — who created it, who reviewed it, the rights it was done under. The answer when the EU AI Act asks where your training data came from in August 2026.
When a queue opens a gap, the bench recruits against the actual work — vetted specialists in the field that's short, with the reason on the record before anyone is booked.
A competitor onboarded its workers like consumers and pooled their data in one place — then lost four terabytes, including who its workers were and the protocols they labeled under. Here, every specialist is on one contract, identity-verified, with your data isolated to your program. Never pooled. The architecture is the answer to the breach.
Most buyers can't trace where their training data came from, and can't validate it. When the EU AI Act asks in August 2026, the chain of consent and provenance is already attached to every datapoint — who created it, who reviewed it, and the rights it was done under. And the bench is independent: a neutral second source, not a vendor absorbed by one of the labs it served.
Test the run. Review the hard cases. Recruit the right specialist. Remember the misses. Approve what's right.
Every override, override note, and signed call — attributed and on file.
See the page →Same rubric. Same reviewers. Same standard across every batch.
See the page →Codify the rubric, score every release against it, and keep review evidence attached.
See the page →Bring the work. Keep the proof. Verified specialists, signed for their work, never pooled — and a chain of consent on every datapoint.