GDPR, including the clause most tools skip.

What the platform evidences

Art. 5(1)(b) — Purpose limitation

A declared attribute is readable on the fairness path and on no other. The limitation is a mechanism rather than a policy statement, which is the difference between enforced and asserted.

Art. 22 — Automated decision-making

Which decisions ran without a person, which reached one, and the safeguard that produced human intervention when it was required.

Art. 25 — Data protection by design

Authority bounded in advance, information-flow gating, and residency enforced by the deployment rather than by configuration.

Art. 30 — Records of processing

The agent and model inventories, with provenance. The automated portion of the Art. 30 record; the wider register stays yours.

Art. 32(1)(d) — Testing effectiveness

The adversarial coverage grid against our own identifier detector, published including the cells it fails. This is the clause that asks for a result rather than a process, and it is the page we are proudest of.

Art. 17 — Right to erasure

Payloads erased on request while the structural record survives and every proof already issued keeps verifying — so a deletion obligation does not cost you the audit trail.

Retention is resolved, not configured once

The retention window comes from the resolved market profile, and a record past its window appears on a queue with the regime that set the clock named beside it. A queue built from scans knows nothing about what was never scanned, so the unscanned count sits next to the answer rather than under it.

What is not detected. Names, addresses and dates of birth in free text carry no checksum and are not looked for. A clean scan means no verified identifiers were found — never that a payload contains no personal data.