REF

What the company appears to be:

The description states that REF is an AI-powered referee for fitness competitions that aims to level the playing field between athletes of different strength levels. It uses a deterministic system to compute handicaps and fairness scores, with the AI's reasoning process visible in real time through a trace console.

What changed:

This project was submitted as part of the OpenAI 2026 hackathon. The authors describe building a full planner chain using AI tools like Codex for orchestration, while keeping core math outside the model to ensure transparency and auditability.

The single most important open question — the commercial due-diligence read:

Is there evidence that REF has traction or adoption beyond the hackathon context? There is no indication of revenue, customers, or usage metrics beyond the self-reported project description.

The description states that REF is an AI referee for fitness competitions. It analyzes performance, applies fair handicaps, validates every rep, and explains its calls — so any two athletes can compete fairly.

It uses a five-tool planner chain:

1. get_lift_history
2. compute_handicap
3. generate_challenge
4. validate_attempt
5. adjudicate_result

These tools run in a capped regeneration loop (max 2 attempts) and stream decisions via Server-Sent Events to a trace console that users can watch live.

The AI's role is to reason through the plan and narrate it in plain English; the numbers underneath are deterministic and reproducible every time.

The description states that REF positions itself as an alternative to existing fitness apps that ignore fairness or hide their logic behind opaque recommendations.

It claims to make the fairness calculation the core product, not just a feature. The AI’s reasoning process is presented as the main event — a trace console styled like a referee call rather than a debugging tool.

The authors also state they wanted an AI feature where the reasoning itself was the demo, not a hidden step before a result screen.

This evolution from cosmetic competition features to transparent, auditable fairness mechanisms suggests a shift toward building trust through explainability and determinism.

The description states that REF targets athletes competing in fitness environments — specifically those who want fair competitions between individuals of different strength levels.

It implies a user base interested in competitive fitness apps where fairness matters. However, there is no evidence of specific customer segments, personas, or market size claimed by the authors.

Not evidenced.

The description does not mention any pricing model, monetization strategy, or business model. No information about how REF would generate revenue or what customers might pay for it is provided.

The description states that REF was built with:

• Frontend: Neo-brutalist design system (thick borders, hard-offset drop shadows, high-contrast palette)
• Backend: Lightweight relational schema (athletes, lift_history, matches)
• Streaming: Server-Sent Events (SSE) for one-way event push from planner to trace console
• Contract: A single locked JSON event schema shared across all five tools

It also mentions:

• Use of OpenAI Codex for scaffolding the planner chain and SSE pipeline
• Deterministic functions for handicap and fairness math, not LLM output
• Capped regeneration loop (max 2 attempts) with explicit fallback handling
• Single-attempt fallback ready per risk plan

The authors emphasize that the AI's job is to plan, call tools, and narrate; arithmetic underneath never touches a prompt.

Not evidenced.

There is no mention of users, customers, revenue, or adoption beyond the hackathon submission. No data on usage, retention, or product-market fit is included in the description.

The description states that competitive fitness apps assume a level playing field that doesn't exist — if one athlete lifts 40% more than another, a straight head-to-head isn't a competition but a foregone conclusion.

It contrasts REF with apps that either ignore the gap entirely (static leaderboards, raw number comparisons) or hide their fairness logic behind an opaque recommendation.

However, there is no evidence of competitors named or described in the text. No market analysis or competitive positioning beyond this contrast is provided.

Inferences based on self-reported information:

• Lack of traction or commercial viability: The project was submitted to a hackathon and lacks any indication of real-world usage, revenue, or customer base.

• Limited scalability assumptions: The system uses deterministic math and capped loops. If the app were to scale beyond its current prototype form, it may face challenges in handling more complex scenarios without additional complexity.

• Dependency on AI tooling: Heavy reliance on OpenAI Codex for scaffolding raises questions about long-term maintainability and whether the solution can be replicated or adapted without continued access to those tools.

• Design focus over functionality: The emphasis on visible reasoning as a UX feature may not translate into broader utility unless there is clear demand for such transparency in competitive fitness settings.

1. What is the intended path from this hackathon prototype to a commercial product?
2. Are there any existing users or pilot programs beyond the hackathon?
3. How would you monetize this product? Is there a pricing model or revenue plan?
4. What are the technical limitations of scaling this system for larger numbers of athletes or more complex competitions?
5. Has the team considered how to integrate with existing fitness tracking platforms or devices?
6. What is the long-term vision for the fairness algorithm — will it evolve beyond simple strength-based handicaps?