Phosra Spec / Capability
OCSS v1.0 — DraftAudit
Algorithmic transparency, audit, and dark-pattern interventions.
What Audit does
One audit-ready answer to “how does the algorithm decide what to show kids.”
Every law that asks “how does the algorithm decide what to show kids” — KOSA’s algorithmic transparency duty, KOSMA, CA SB 976 (addictive design), EU DSA Article 28, NY S9051 (AI engagement dark-pattern ban), the EU AI Act’s risk classifications — needs an answer that’s both human-readable and machine-auditable.
Audit is the canonical algorithmic-transparency + intervention layer. It disables infinite scroll and autoplay for minor accounts, blocks engagement dark patterns, weights re-rankers toward civil-society “prosocial” / “role model” / “diverse representation” dimensions, generates DPIA evidence packs on request, and produces the periodic algorithmic-audit report regulators require.
A platform’s response to “show me your algorithmic-decision audit trail for minors in jurisdiction X” goes from a months-long discovery exercise to a generated PDF. Parents see what’s being recommended and why. Phosra’s Receipt signs every decision so the audit trail is court-defensible.
How partners plug in
Audit is a socket. Recommender events flow in. Audit-ready evidence flows out.
These are the upstream auditors, schemas, and rating dimensions Audit publishes against — either shipping today, in conversation with a partner, or pending an upstream pilot.
Standards & laws
What Audit does for each statute.
- KOSA + KOSMA — fulfills the algorithmic-transparency duty for covered platforms.
- CA SB 976 (addictive design) — enforces autoplay-off + infinite-scroll-block for minor accounts.
- EU DSA Article 28 (minor protection) — produces the periodic transparency report on minor exposure.
- NY S9051 (AI engagement dark patterns) — blocks engagement-driven dark patterns in AI products.
- EU AI Act (high-risk classification) — surfaces the “you are talking to an AI” disclosure on every Nth turn.
- FTC algorithmic-transparency consent decrees — generates the agency-required audit pack.
- CA AB 2273 (AADC, Phase 2) — enforces the children’s-design-code on algorithmic surfaces.
Conformance
Adopter Tier 1 certification.
To ship Audit-conformance for an Adopter Tier 1 certification, your implementation must pass the Audit suite. Test count is [draft] coming Q3 2026. The suite covers autoplay/infinite-scroll suppression for minors, dark-pattern detection accuracy, DPIA evidence-pack generation, and signed audit-trail emission to Receipt.
We are co-authoring the suite with our design partners. If you want a seat at the table while the bar is being set, reach out.
Rule list
The 11 rules Audit ships
Every rule below is implemented by this capability. Pulled directly from the rule registry.
- Prosocial Weight — Boosts the recommender weight of prosocial content (educational, civic, mental-health-positive) in minor feeds.
- Role Model Weight — Boosts the recommender weight of role-model content (educators, athletes, scientists) over engagement-only content.
- Representation Weight — Boosts the recommender weight of representative content matching the minor user's identity signals.
- DPIA Request Generator — Generates a Data Protection Impact Assessment request packet for any rule a UK or EU regulator escalates.
- Algorithm Feed Control — Disables personalized algorithmic feeds and switches to chronological or non-profiled content delivery.
- Addictive Design Control — Disables autoplay, infinite scroll, notification streaks, and other compulsive-use design patterns.
- Infinite Scroll Block — Disables infinite-scroll feed mechanics for minor accounts; pagination breaks force a natural pause.
- Autoplay Disable — Disables autoplay on video, audio, and reel surfaces so minor users always tap to start the next item.
- AI No-Engagement Dark Patterns — Blocks AI products from deploying engagement maximizers (streaks, infinite scroll, manipulative nudges) on minor users.
- AI Is-It-Real Disclosure — Requires AI-generated content to carry a 'this is AI' label visible to minor users at the moment of consumption.
- Algorithmic Audit — Generates the per-rule audit trail every recommender decision must produce for regulator replay and civil-society inspection.