This text is a starting draft and should be reviewed by legal counsel before launch.
Methodology and compliance
How expert data is made, protected and delivered
This page is written for legal, security and procurement teams. It describes how we source and verify practitioners, where the data comes from, how custody and intellectual property are handled, and what ships with every batch.
Read the full documentation pack1. Sourcing and verification
Practitioners are recruited through professional associations, academic networks and direct referral from existing contributors. Every candidate holds an active registration with the professional board for their field, and we verify each registration directly with the issuing body before any work is assigned.
Verification is not a one time event. Registrations are revalidated periodically for as long as the contributor remains active in the network, and a contributor whose registration lapses is suspended from new assignments until it is resolved.
- OAB for lawyers, CRM for physicians, CRC for accountants, CREA for engineers
- Direct checks against the issuing professional board, never self reported screenshots
- Periodic revalidation for every active contributor
2. Task design and data origin
Every task is built on a synthetic case constructed by our task design team. We maintain an explicit, contractual prohibition on the use of client files, patient records, case files or any third party confidential material at any stage of production. Contributors confirm this prohibition on every assignment.
Synthetic construction also protects against reproducing copyrighted material. Cases are authored from scratch, checked against public sources during review, and rejected if any element reads as derived from an identifiable real matter or published work. Because we author the case, we hold the ground truth about what it contains, which makes correctness decidable and audit possible.
3. Chain of custody
A trajectory moves through three stages: creation, review and delivery. Each stage is logged. At creation we record the contributor, the task version and the instrument session. At review we record the reviewer, the rubric version, the scores and the written annotations. At delivery we record the batch composition and the recipient.
Batches are versioned and immutable once delivered. Every batch carries a checksum, and every item in the batch carries an identifier that traces back through its review record to its creation record. If a customer ever needs to demonstrate how a specific item was produced, the chain is complete.
4. Intellectual property
Every contributor signs a full assignment of intellectual property to Práxis covering the work product they create in the network. Práxis in turn assigns that intellectual property to the customer on delivery, under the commercial agreement.
The contractual chain is deliberately short and auditable: contributor to Práxis, Práxis to customer. There are no intermediate licensors, no retained contributor rights over delivered work, and no ambiguity about who owns the data the customer trains on.
5. Privacy and data protection
We operate under the LGPD in Brazil and maintain a GDPR posture for European customers. Our practice is data minimisation: we collect only the personal data needed to run the contributor relationship and the customer engagement, and production tasks contain no personal data of third parties by construction.
Data residency options are available for customers with jurisdictional requirements. Our current subprocessors are Supabase for managed database infrastructure and Resend for transactional email, each bound by a data processing agreement. The list is reviewed whenever our stack changes and is available on request.
6. Confidentiality
Customer task specifications are confidential. Access is limited to the people assigned to that customer's work, enforced through role based access control, and contributors see only the tasks they are assigned, never the customer's identity or the broader program unless disclosure is agreed in writing.
Customer projects are separated logically and operationally. Reviewers working on one customer's batch do not access another customer's materials, and we do not reuse one customer's task designs or trajectories for another customer.
7. Delivery formats and provenance
Each batch ships as structured data in the customer's agreed format, together with a delivery report and a provenance record per item. The provenance record states the task version, the review trail, the audit status and the batch checksum, so every item can be independently audited after delivery.
If a customer's legal, security or procurement team needs more than this page provides, we maintain a full documentation pack covering policies, subprocessor terms and audit procedures, available under NDA on request.