Pricing

Priced per judgement, quoted per project

Evaluation work varies too much for one rate card to be honest. We price each project on what it actually takes, from a sample of your data, and the quote shows how we got there.

What drives the price

Five things decide what a judgement costs

You control most of them. Overlap and review depth in particular are a trade between cost and confidence, and we would rather you made it deliberately.

Task complexity and length

A two-line answer to compare takes a minute; a long technical response with a detailed rubric takes far longer. Price follows the time a careful judgement needs.

Expertise and language

Work that needs domain knowledge, such as code, medicine or law, or a particular language, draws on a smaller pool of raters.

Overlap rate

How much of each batch is rated independently by several people, and by how many. Each extra rater on an item is another paid judgement.

Gold and QA coverage

How many known-answer tasks are seeded in, and how much of the completed work a reviewer checks before delivery.

Turnaround

A deadline that needs a larger cohort working in parallel costs more than one that lets a smaller group work steadily.

Included in every project

No extras for the controls

The quality machinery isn’t an add-on. Every project gets all of it.

Rubric setup

We draft the rubric with you and validate every submission against it.

Written guidelines

Shown to raters while they work, so nothing is left to inference.

Qualification exam

If you want one: raters must reach the pass mark before live work.

Gold tasks

Known answers mixed invisibly into the queue to measure accuracy.

QA review

Pass, rework or fail, with a written reason for every return and fail.

Agreement reporting

Majority verdict, pairwise agreement and Fleiss’ κ on overlapping work.

Audit trail

An append-only ledger and a log of sensitive actions behind the data.

CSV and JSONL exports

The full record, and preference pairs ready for training.

Rater payouts handled

We recruit and pay the raters; you deal only with us.

How to start

From a sample to a quote

01

Send a sample

100 rows or so, and a description of the judgement you need made, in plain words. Tell us roughly how much data you expect, and how often.
02

Get a quote

We come back with a draft rubric, a cohort plan, the overlap and review we would recommend, and a price per judgement with turnaround.
03

Run a pilot, if you like

A first batch run end to end, with its quality report, so you can judge the data before committing to volume. The quote includes its price.
04

Scale up

Send batches as CSV whenever you are ready. Exports and quality reports come back per batch.

For raters

What the work pays

  • A flat amount per rating, set by each project. You see it on your dashboard before you start a session.
  • Credited as soon as you submit. Reversed only if QA fails the rating, and you can appeal.
  • Exams are unpaid. Gold and overlapping tasks pay the same as any other task.
  • No fees. Withdraw any amount from ₹1 to your bank account.

For AI teams

Ask for a quote

Prices are set per project, so the fastest way to a number is a sample and a sentence about what you need judged.

Get a quote