How AIR-SPJ Works

The platform serves four distinct user types, each with progressively more access. Public visitors can explore without logging in. Contributors are approved researchers who submit evidence. Authors are the three-member core team who vote on submissions. Admins manage the platform.

Contributors submit evidence through a guided three-step wizard. The AI assists throughout — extracting metadata from PDFs, suggesting element connections, and classifying source type and tier. Only the title is required; everything else is AI-suggested and can be corrected.

Becoming a contributor starts with creating a free account, then applying at /contribute. Any author or admin can approve an application — approval immediately grants the requested role on your existing account, and you'll get an email the moment it happens. No separate invite link required; just sign back in and start submitting.

Every submission is reviewed by up to three platform authors. As soon as two of the three agree (approve or reject), the item resolves automatically and — if approved — is published to the public Evidence Library and Evidence Map immediately, no further step required. The site admin can also override at any point to finalize a decision immediately, for example when consensus can't be reached in time. Approving or rejecting an evidence item automatically cascades that status to all of its element connections.

The AIR-SPJ framework is organized into 12 clinical domains with 87 total elements. Each element is a discrete behavioral risk indicator assessed in cases where AI interaction is a contributing factor. Evidence connections tie peer-reviewed and practitioner literature directly to individual elements, passage by passage.

The platform recognizes 14 distinct source types and a 6-level evidence tier hierarchy. AI classifies both automatically during submission; authors can correct them during review. Tier 1–2 represents the strongest evidence (systematic reviews, RCTs). Tier 5–6 captures expert opinion and practitioner field observations.

The admin console contains four review queues. The Evidence Queue is the primary workflow — every vote and metadata correction happens there. The Connection Queue is read-only status tracking (connections are decided automatically when parent evidence is decided). The Observation Queue handles practitioner field submissions. The Application Queue manages contributor onboarding.

Practitioner observations are a distinct submission type from evidence. They don't require a published source document — they capture anonymized field experiences from threat assessment professionals who have encountered AI interaction as a case factor. They go through the same 3-author review process but use a lighter form tied directly to a single framework element.

The AIR-SPJ is a structured professional judgment tool — not a checklist or scoring instrument. Practitioners use it to systematically consider AI-mediated risk factors within a broader clinical or threat assessment formulation. The four-step model below describes the intended workflow for professionals using the framework in active cases.

The Copilot at /copilot is a RAG-powered research assistant that answers questions about the AIR-SPJ framework, its evidence base, and AI-mediated risk topics. It draws exclusively from the platform's approved content — it does not search the open web — and cites specific evidence items and element codes in every response. No login required.

The platform is built on a principle of radical transparency for the evidence base and strict privacy for individuals. All approved research and connections are public by design — the goal is an open, auditable record of how the framework is built. Contributor identities, review deliberations, and rejected submissions are never exposed.

Most of the platform is publicly accessible. Anyone can browse the framework, read evidence, explore the network map, and use the AI Copilot. Login is only required to submit evidence or access admin functions.