WrongWorlds

WrongWorlds is an educational game concept built as a self-reported prototype for the OpenAI 2026 hackathon. The author describes it as a five-act reasoning game where learners enter worlds governed by their misconceptions, test those beliefs, and rebuild stronger causal models. It uses GPT-5.6 for interpreting learner language and deterministic TypeScript engines to govern simulation outcomes.

The project is presented as an experimental educational tool with no evidence of revenue, customers, or traction beyond its own submission. The system architecture separates generative interpretation from deterministic authority, aiming to preserve reproducibility and trustworthiness in educational evaluation.

The description states that WrongWorlds is a five-act illustrated reasoning game. It begins with a learner's misconception, which becomes a "Belief Core" and activates a portal to a world where that belief is true.

In this world:

• Learners must investigate the consequences of their belief.
• Collect contradictory evidence.
• Fracture false laws.
• Rebuild a stronger explanation.
• Escape by proving reasoning transfers elsewhere.

The system uses GPT-5.6 for interpreting learner language into structured belief models, and deterministic TypeScript engines to control simulations, evidence generation, scoring, and state transitions.

It is described as a complete educational game experience, including:

• Two visually distinct worlds.
• Keyboard navigation.
• Mobile layouts.
• English/Spanish localization.
• Accessibility features (reduced motion, 2D mode).
• Deterministic replay capability.
• Public Evidence Archive.

Not evidenced No mention of actual users, revenue streams, or adoption metrics. The system is presented as a prototype built for a hackathon.

The author positions WrongWorlds as an alternative to traditional correction-based learning, where learners receive the right answer but do not change their mental models.

Instead, WrongWorlds treats misconceptions as testable realities—a world that can be entered, explored, and escaped from once better reasoning is demonstrated.

It claims to:

• Make misconceptions visible.
• Allow testing of belief predictions.
• Enable cinematic reality fracture.
• Support causal model reconstruction.
• Validate transfer through Bloom Station.

The positioning evolves from a conceptual framework (a "museum of possible realities") to a functional prototype, with the author emphasizing that it's not just about correcting errors but about changing how learners reason.

Inference This is a novel pedagogical approach, but no evidence exists that it has been tested or validated beyond its own claims.

The description implies that WrongWorlds targets learners in educational settings, particularly those dealing with causal reasoning—such as students studying marketing, science, economics, history, statistics, systems thinking, or media literacy.

It is designed to help learners:

• Understand why their initial belief was flawed.
• Engage deeply with the logic behind causality.
• Apply learned reasoning patterns in new contexts.

The system supports English and Spanish, suggesting a global audience, and includes accessibility features like keyboard navigation and reduced motion support.

Not evidenced No specific customer segments, age groups, or institutional use cases are identified. The target is implied rather than defined.

There is no evidence of any business model or pricing structure in the description.

The author mentions:

• Teacher-authored WrongWorlds.
• Classroom cohorts.
• Learner progress histories.
• Adaptive worlds.
• Controlled studies.

But these are speculative future features, not current offerings.

Not evidenced No revenue streams, monetization strategies, or pricing tiers are described. The project is presented as a prototype for a hackathon.

WrongWorlds was built using:

• Codex and GPT-5.6 for interpretation.
• TypeScript, React, Next.js, Node.js for implementation.
• Playwright, Vitest for testing.
• Vercel, GitHub for deployment.

Key technical decisions include:

• Separation of generative interpretation from deterministic authority.
• Deterministic replay with zero POST requests.
• Public Evidence Archive and reproducible screenshots.
• Git-backed production release.
• Accessibility parity (2D mode, keyboard support).

The system architecture is described as:

• GPT-5.6 interprets learner language into a belief model.
• Deterministic engines govern simulation outcomes, evidence, scoring, etc.

Inference The technical stack suggests a strong focus on reproducibility and educational integrity, but no indication of scalability or infrastructure beyond the prototype.

The project is described as a complete five-act game experience, with:

• Two worlds.
• Transfer assessment via Bloom Station.
• Public GitHub repository.
• Evidence Archive.
• Localization in English/Spanish.
• Accessibility features.

It passed:

• 143 Vitest tests.
• 26 Playwright tests.
• Production build verification.
• Public QA checks.

However, the author explicitly states that:

• The current Evidence Archive is exploratory.
• No educational efficacy data exists.
• Larger studies are needed to measure learning outcomes.

Not evidenced No user base, usage metrics, or performance tracking. The system is presented as a prototype with no traction indicators.

The description does not mention competitors directly. However, the concept aligns with:

• Educational AI tools focused on reasoning and misconception correction.
• Gamified learning platforms that use simulation or narrative to teach concepts.
• Causal reasoning frameworks used in STEM education.

It appears to be a novel approach within the space of educational games, using AI to interpret learner intent while maintaining deterministic control over outcomes.

Inference The product is positioned as an experimental innovation rather than part of an existing competitive landscape. No known direct competitors are named.

1. Prototype-only status: The system is described only as a hackathon prototype with no evidence of real-world testing or adoption.
2. Unproven educational impact: No data on learning gains, retention, or reasoning improvement.
3. Limited scalability assumptions: The architecture implies strong control over outputs, but there's no indication how this scales beyond one developer.
4. AI dependency without clear governance: While GPT-5.6 is used for interpretation, the deterministic engine controls outcomes—this balance may be fragile if AI quality degrades.
5. No monetization strategy: No business model or pricing structure is evident.

Not evidenced No evidence of market demand, user feedback, or competitive positioning beyond its own claims.

1. What specific educational outcomes have you observed in pilot testing (if any)?
2. How do you plan to scale the system beyond one developer?
3. Are there plans for formal pedagogical validation or controlled studies?
4. What would constitute a minimum viable product (MVP) for this concept?
5. How do you intend to monetize or deploy this in real-world educational environments?
6. Can you demonstrate how the deterministic engine ensures consistent evaluation across different learner inputs?
7. What are the key assumptions about user behavior and engagement that underpin the design?