Skip to content

Careful
is a process.

Good data doesn't come from hoping everyone tries hard. It comes from a system that catches mistakes early and gets people better each week. Here's ours.

80–90%

first-pass acceptance target

2×

eyes on every batch: the author and a reviewer

Day 1

calibration on gold tasks before any live work

Six rules we don't bend.

  1. 01

    Gold tasks before live tasks

    Nobody touches production data until they've matched our answers on a calibration set.

  2. 02

    Disagreements get resolved

    When two reviewers disagree, a lead decides, writes down why, and the guideline gets updated.

  3. 03

    Small batches

    Work ships in small batches, so a misunderstanding costs an afternoon, not a week.

  4. 04

    Second-person review

    A different person checks a sample of every batch.

  5. 05

    Scores per person

    We track acceptance for each contributor. Coaching comes first. Removal comes if it doesn't help.

  6. 06

    NDA by default

    Everyone signs before onboarding. Project details stay out of screenshots, chats and portfolios.

Questions teams ask.

What does 80–90% acceptance mean?+

It's the share of submitted work a client accepts on the first pass, with no rework. It's the target we run our team against, and we report it to you per batch.

How do you onboard a new project?+

We read your guidelines, write our own calibration set, run it with a small group, and send you the questions it raised before scaling up.

Can we audit your reviewers?+

Yes. You can see per-batch scores, sample rationales, and the guideline changes we made along the way.

What if quality drops?+

We pause the affected contributor, re-review their recent work, and tell you exactly what happened.

See it on your data.

Start with a pilot →