The answers we coach you to demand — from us.

Data security & confidentiality

Client training data is proprietary IP, and we treat it that way from first transfer to final deletion.

• Mutual NDA before any data access — signed at the company level before a single file moves.
• Every individual annotator signs an NDA — confidentiality doesn't stop at the company boundary; the people touching your data are personally bound.
• Per-project access controls — annotators are granted access only to the project they're staffed on, and access is revoked at project close.
• No local copies — work happens in your pipeline or in controlled environments; data doesn't live on personal machines.
• Deletion on request — at engagement end, client data is removed from our working environments on your instruction.

Quality structure

Quality isn't a slide in our proposal — it's a structure, and every part of it is measurable.

• QA is separated from annotation — the people reviewing work are never the people who produced it. (Checklist question 5 — and the reason we put it there.)
• Inter-annotator agreement (IAA) tracked per project — reported to you, not just monitored internally. Here's how we think about IAA →
• 95%+ annotation accuracy — written into engagement terms — our accuracy commitment is a contractual SLA, not a marketing number.
• Taxonomy training before the first batch — annotators train on your labeling guidelines and edge cases before touching production data.
• Edge-case escalation process — ambiguous items are flagged and resolved with your team, not guessed at.

Certifications

We don't currently hold a SOC 2 certification, and we'd rather tell you that plainly than imply otherwise. What we offer instead is contractual: NDAs at both company and annotator level, per-project access controls, and an accuracy SLA in writing. For engagements with specific compliance requirements, we'll walk through our controls in detail during scoping — and put what matters to you in the contract.