STEP 01
WHO WE RECRUIT
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.
→HUMAN DATA · WORKFORCE · EXPERTS IN YOUR FIELD
Experts in your field. Ready for AI work.
Find vetted specialists for the training work your model needs: labeling, ranking, writing, evaluation, voice, robotics, and domain review.
BENCH COVERAGE
FIELD · QUEUE
ONE PROGRAM · ONE BENCH · DATA ISOLATED TO YOUR TEAM
● READY
EXPERT PAY
Quote-scoped expert rates
What credentialed specialists now earn to author tasks, rubrics, and reasoning traces.
IDENTITY-VERIFIED
Every specialist
Credential-checked and on one contract. Cleared on the rubric your team already runs.
DATA SOURCE
Aug 2026
Know which specialist created the work, what task they completed, and which rights came with it.
HOW IT WORKS
Vetted experts. On the record.
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.
STEP 02
WHAT WE REVIEW
Verify and vet
Identity-verified, credential-checked, and tested on the task your team already runs. Trial cases and calibration results sit on file.
→STEP 03
WHERE THEY LAND
Put them on your queue
One contract. One payout record. Assigned to labeling, ranking, evaluation, voice, robotics, or domain tasks under the same standards your model is graded on.
METHODOLOGY · FIVE STAGES
Queue, match, schedule, close the gap, learn.
Five stages, the same on every program — annotation, review, approvals, and expert tasks. The work comes in. Your team gets usable training data out.
01 · QUEUE
One queue, one record of the work
Annotation, review, approvals, and specialist tasks enter one place with the rubric, the usage terms, and the owner attached. Nothing runs in a side channel nobody can find later.
Task batch · rubric · usage terms → Queued work with source details
02 · MATCH
Assign the right specialist, with the reason
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.
Field · calibration · capacity → Named specialist with a reason
03 · SCHEDULE
Book the work with terms on the record
Week and month calendars handle shift assignment, scoped budget bands, and conflict detection. The commercial terms stay tied to the shift record.
Shift · budget band · conflict rules → Booked shift with terms on file
04 · CLOSE THE GAP
Recruit when the field runs short
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.
Coverage gap · field · priority → New specialists sourced against the work
05 · LEARN
Every outcome feeds the next run
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.
Decision history · review feedback · misses kept → Better coverage, failures remembered
ROSTER · BENCH VIEW · ANONYMIZED
The bench, by field.
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.
CODING & AGENTS
CLINICAL & MEDICAL
LEGAL
FINANCE
ROBOTICS
SAFETY & RED-TEAM
FIELD: CLINICAL
TIER: CALIBRATED
AVAILABILITY: 48H
IDENTITY: VERIFIED
Names anonymized. Example bench view from the platform workforce directory.
COVERAGE · WHERE THE FIELD IS SHORT
When a queue opens a gap, the bench recruits.
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 with clear project terms.
GAP TO QUEUE · ONE FIELD
GAP
Robotics short
SOURCED
In the field
VERIFIED
Identity + rubric
ON THE QUEUE
Ready
COVERAGE · FIELD × QUEUE
FIELD × QUEUE
EVAL
ANNOTATE
APPROVE
CLINICAL
8
22
48
VISION
0
16
34
ROBOTICS
26
42
12
SAFETY
6
18
30
● LOW
● MEDIUM
● HIGH
TRUST TIMELINE · ONE SPECIALIST
Trust is earned on the record.
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.
W-8 · CALIBRATION
Gold-set pass · 0.91
First calibration on the rubric the team runs. Trial set scored, reviewer notes filed, tier opened.
↳ TIER ACTIVE
W-4 · REVIEW
184 tasks · 96 QA
Active queue load with reviewer agreement above the floor. The bench treats this as a trusted contributor.
↳ TRUSTED
W-2 · INCIDENT
1 span reopen
Reopen on a single span. Coached, not penalised. The reason and the resolution stay on the specialist's record.
↳ COACHED
NOW · TIER CHANGE
Clinical cert active
New calibration activated. Specialist now eligible for clinical NLP queues. Review record updated.
↳ CALIBRATED
SURFACES · FOUR PILLARS
Four surfaces. One bench.
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.
ROSTER
Who's on the bench
The full roster, filterable by field, certification, availability, and trust tier — with coverage and a one-click way to put specialists on a queue.
DETAIL
How they got there
Identity and credential checks, certification history, earnings, quality trend, active assignments, and the calibration record from your rubric.
SCHEDULER
Where the work is booked
Week and month calendars for shift assignment, budget bands, and conflict detection — the commercial terms tied to the shift record.
COVERAGE
Where the next specialist comes from
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.
WHAT COMES OUT
What your team leaves with.
The bench is the artifact. Who's on it, what they can do, what work they completed, and what they were paid. Specialists recreate realistic tasks from clean-room scenarios, so no employer data is ever uploaded.
01
Specialist roster
Every specialist on the bench, identity-verified, the rubrics they're cleared on, and the queues they're on.
↳ ARTIFACT
02
Work records
Every completed task, quality note, and correction attached to the specialist who made it, with the rights it was done under.
↳ ARTIFACT
03
Certification reports
Who passed which task test, on what date, against which trial set. The record a program lead can inspect.
↳ ARTIFACT
04
Payout records
One contract per specialist. One ledger per queue. The money trail matches the work trail.
↳ ARTIFACT
PROGRAM NOTE · ON THE RECORD
“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.”
QUESTIONS · BEFORE YOU STAFF
What teams ask before they put the bench on a queue.
Six questions, six answers. The same six come up across annotation, review, approvals, and specialist execution.
01 · ON IDENTITY
How do you keep our data — and our workers' — from leaking?
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.
02 · ON CONTRACTS
How are specialists contracted?
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.
03 · ON CERTIFICATION
Whose rubric clears them?
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.
04 · ON PRICING
How does the bench price?
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.
05 · ON SOURCE DETAILS
Can we show where the data came from?
Yes. Every datapoint carries the useful source details: who created it, what task they completed, and the rights it was done under. That is the answer when the EU AI Act asks where your training data came from in August 2026.
06 · ON COVERAGE
What happens when a field runs short?
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.
WHY TEAMS USE IT
Bring the work. Get the expert data.
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. We keep the simple source details with the dataset: who created the work, what task they completed, and which rights came with it. And the bench is independent: a neutral second source, not a vendor absorbed by one of the labs it served.
WHERE IT FITS
In the loop, this is where you recruit.
Test the run. Review the hard cases. Recruit the right specialist. Remember the misses. Approve what's right.
01
Test
02
Review
03
Recruit
● YOU ARE HERE
04
Remember
05
Approve
RELATED MODULES
Next to this in Human Data.
CLEO
The chair of record for every review.
Every override, override note, and reviewed call — attributed and on file.
See the page →ANNOTATION
Annotation that doesn't drift.
Same rubric. Same reviewers. Same standard across every batch.
See the page →EVALUATION STUDIO
Review it before release.
Codify the rubric, score every release against it, and keep review evidence attached.
See the page →WORKFORCE
Experts in your field. Ready for AI work.
Bring the task. We find the specialists. You get the training data, the roster, and the usage terms in one place.