SABERISTIC
conceptNo visitor data expected

Psych Lab

An AI-assisted questionnaire workshop where creators draft, review, publish, distribute, and score self-report instruments through join-code participant flows.

THE QUESTION

Why this exists

Turning a research or self-reflection idea into a coherent questionnaire requires item design, response scales, reverse scoring, validity checks, interpretation text, publishing, distribution, and reproducible scoring.

THE BUILD

What this prototype explores

Psych Lab explores the full path from an author's idea to a structured self-report questionnaire. A creator can ask for a familiar format or describe a new topic, receive an AI-assisted draft, repair it against a strict schema, edit every word, and publish it behind a join code. Participants answer without interacting with the model; reverse scoring, attention checks, sums or means, and prewritten interpretation bands make the result deterministic. The product deliberately draws a boundary between authoring assistance and evaluating a person. It presents itself for research, education, entertainment, and self-reflection—not diagnosis, screening, treatment, or consequential automated decisions. The repository also explores creator plans, participant report purchases, analytics, marketplace mechanics, and administrative oversight. Those wider capabilities make the current safety bar much higher. The reviewed source has six open HIGH-severity CodeQL alerts and includes questionnaire, identity, account, analytics, subscription, and payment paths. The canonical domain is retained as reviewed provenance, but this publication does not expose it as a launch action. Psych Lab remains a concept until the security findings and the complete data lifecycle are resolved and independently verified.

DECISIONS

What shaped the build

  1. Confine AI to authoring

    AI drafts questionnaire wording and structure before participation; respondent answers are scored with fixed arithmetic and are not sent back to the model for profiling or automated decisions.

  2. Use one validated specification

    Established and novel drafting paths both produce a strict JSON contract with response scales, subscales, reverse scoring, attention checks, interpretation bands, and non-clinical disclaimers.

  3. Keep a human publication gate

    Invalid model output is returned with schema errors for repair, while every valid draft still waits for a creator to edit and approve it before publication.

  4. Separate authoring from participation

    Short join codes separate creator authoring from the participant flow, while deterministic sums or means make reported scores reproducible.

KNOWN LIMITS

What is not being claimed

  • Six open HIGH-severity CodeQL alerts were present at the reviewed commit; production use is blocked until they are resolved and independently rechecked.
  • The application handles questionnaire responses, participant identifiers, creator accounts, analytics, subscriptions, and payment events, but those data paths have not passed launch review.
  • Generated questionnaires are drafts, not validated clinical or diagnostic instruments, and must not drive healthcare, hiring, credit, insurance, admissions, or similar decisions.
  • Reproducing established instruments can raise copyright, licensing, fidelity, and commercial-use questions that schema validation cannot resolve.