One source of truth. The authoritative, always-current security posture — subprocessor
list, DPA, SOC 2 status, incident history — lives at the
Phosra Trust Center. This page mirrors it for developers
and adds the runnable erasure flow. Where a claim is forward-looking rather than
code-verifiable today, it is tagged policy preview.
Data minimization — collect the least that works
Phosra is a control plane for parental-controls policy, not a data lake. It stores the metadata needed to compile and enforce a policy, and nothing it does not need. The core object model is the whole surface — there is no hidden telemetry object.What each object stores
Family — the household root
Family — the household root
No address, no payment data, no location. A family is a container.
FamilyMember — a guardian on the account
FamilyMember — a guardian on the account
The parent’s login identity (email) is held by the auth provider
(WorkOS), stored hashed, and referenced here
only by
user_id. Phosra’s product tables do not store the parent’s raw email.Child — a kid's profile
Child — a kid's profile
This is the most sensitive object. It is deliberately thin.
Never collected on a child: device identifiers tied to advertising, precise geolocation,
biometrics, contacts, browsing history, or message content. Enforcement happens against
categories, not against a log of what the child did.
ChildPolicy + PolicyRule — the ruleset
ChildPolicy + PolicyRule — the ruleset
Policy state is configuration, not behavior. A
PolicyRule records which of the
123 OCSS categories is on and its config — never an event
that a rule fired against a specific child action.Enforcement + audit — the durable record
Enforcement + audit — the durable record
The data-plane
EnforcementReceipt is identity-free by construction — no child ref,
device id, or user id (see Data model → EnforcementReceipt). The
router that moves OCSS signals cannot decrypt the payload it carries; the envelope is
sealed to the recipient’s key. See Content monitoring.What Phosra never does with child data
No sale or brokering
Minor data is never sold, rented, or brokered — ever.
No ad targeting
Child data is never used for advertising or ad targeting, and never shared with ad
networks or data brokers.
No external model training
Child data is never used to train external ML models.
No content logging
Phosra enforces against categories. It does not keep a log of a child’s messages,
browsing, or activity.
Retention
Retention windows are parent-configurable, and defaults follow the most conservative
applicable statute (COPPA / UK AADC). The exact numeric windows and the backup-rotation
period are governed by the DPA rather than the API — request them from
security@phosra.com for a security review. They are tagged
policy preview here because they are not readable from the
sandbox or OpenAPI spec.
Right to erasure — a real, runnable flow
Erasure is not a support ticket. Deleting aFamily cascades to every Child,
ChildPolicy, PolicyRule, and enforcement record beneath it, and deleting a single Child
erases just that child’s subtree. Both are ordinary DELETE calls that return 204 No Content.
Every request and response below was captured live against the public sandbox
https://phosra-api-sandbox-production.up.railway.app on 2026-07-06 — so you can run this
end-to-end right now. In production these routes require a WorkOS session JWT for the signed-in
parent and an owner/parent role on the family — they are consumer routes, so a phosra_
developer key does not authenticate them. See Authentication and
Environments.Erase a single child
DELETE /children/{childID} removes the child profile and its entire policy subtree.
404, not a soft-deleted tombstone:
Erase the whole household (cascade)
DELETE /families/{familyID} erases the family and everything beneath it in one call —
proven live: a child created under the family reads 404 immediately after the family delete.
Removing one guardian, not the household: to detach a single parent/guardian without
erasing anyone’s data, use
DELETE /families/{familyID}/members/{memberID} (also 204).
That removes the person’s access; it does not delete the children.Data portability / export
Machine-readable JSON exports for portability (GDPR Art. 20 / CCPA) are available on request — email security@phosra.com. There is no self-serve export endpoint on the API today (aGET …/export returns 404) — we say so plainly rather than imply one
exists. A self-serve export API is policy preview.
Encryption
In transit — TLS 1.3 + HSTS
HTTPS is enforced end-to-end with TLS 1.3; HSTS at the edge. A conservative
Content-Security-Policy restricts script, frame, form, base, object, and worker sources.
Plain-HTTP requests are redirected, never served.
At rest — AES-256-GCM
Postgres with tenant scoping enforced at the application layer on every query.
AES-256-GCM is applied to sensitive fields (OAuth tokens, provider credentials) on top of
standard at-rest disk encryption.
Sealed data-plane envelope
OCSS signals are sealed end-to-end to the recipient’s key. The routing layer reads
only the headers it needs to move a signal — the router structurally cannot decrypt the
payload. See Content monitoring.
Short-lived auth
Production auth runs on WorkOS with short-lived JWTs and a separate sandbox tenant for
development. Admin actions are gated behind explicit role checks on every request.
Data residency & subprocessors
US-primary data residency across all subprocessors. Phosra updates this list when the data path changes; the Trust Center is the authoritative copy.
Need the signed subprocessor DPA for your vendor review? Request it at
security@phosra.com.
Regulatory alignment
Phosra maps its controls to the statutes that govern children’s data. Conformance is evidence a regulator can weigh — not an approval Phosra issues, and not a safe harbor.
See the compliance hub for the per-statute detail pages.
Report a vulnerability
Phosra welcomes good-faith security research.Email security@phosra.com
Send findings to security@phosra.com. We triage within
one business day and keep you updated through remediation.
Stay in scope
Don’t access data that isn’t your own, don’t test against production parent or child
accounts, and don’t run scanners that degrade service. Use the open
sandbox for testing — it holds only synthetic data.
Coordinated disclosure
Give us a reasonable window to remediate before public disclosure. With your permission,
we credit researchers in our disclosure log.
Where to go next
Trust Center →
The authoritative, verifiable security & governance posture.
Core objects & data model →
Every field of every object this page classifies.
Environments →
Sandbox vs. production, and why the sandbox holds only synthetic data.
Authentication →
Keys, roles, and the owner/parent gate on destructive calls.