Sports interfaces often display a probability without showing which evidence produced it, which assumptions changed it, or how the result should be evaluated afterward. That presentation makes uncertainty easy to mistake for certainty and removes much of the educational value of a model.
Mundial Lab uses the 2026 World Cup as an accessible environment for probabilistic literacy. Participants act as Data Directors: they inspect a match, choose a model, adjust parameters the model can actually use, preserve a probabilistic reading, and compare it with observed results and model-generated baselines.
Mundial Lab is an independent Hadox and Polisplexity educational prototype. It is not an official tournament data service, a betting product, or a promise of predictive certainty.
The source-available repository combines a public experience with a working analytical and operational stack:
The differentiator is not a claim to know a winner. It is the ability to show what a model believed, which inputs supported that belief, what changed, and what can be learned after an outcome becomes known.
The frontend communicates with a FastAPI backend backed by PostgreSQL and Redis. Background workers execute larger simulations and persist artifacts; scheduler and ingestion workflows collect and normalize available public evidence. The repository provides development and production-shaped Compose definitions, but public operation still depends on reviewed operator infrastructure.
Model availability is not proof of external validation. Elo, Poisson, and Ensemble are available in the application, while Dixon-Coles and Bayes Live remain prototypes. Several experiences still depend on seed, derived, or simulated inputs, and identity is currently alias-based rather than a durable authenticated account.
Every forecast must remain tied to a data snapshot, model version, parameters, random seed, confidence level, and known limitation. Public, estimated, manually entered, and simulated evidence must never be presented as equivalent. Prediction surfaces should encourage calibration and post-result examination rather than certainty or wagering behavior.
Production exposure requires centralized identity and authorization, a fail-closed administrative boundary, strict CORS and secret management, abuse and rate controls, retention and deletion rules, operator audit policy, verified primary results, model calibration, and independent review of public claims.
Mundial Lab is an integrated research and showcase prototype, not a calibrated forecasting service or production-grade multi-tenant platform. Priority work includes durable identity, stronger operator authorization, complete primary-source verification, fit and calibration pipelines, real-time evidence adapters, accessibility and localization review, backup restoration tests, and responsible-use evaluation with educators and domain specialists.
The public repository exposes working software and documentation for transparency, portfolio review, and technical evaluation. It is proprietary and source-available, not open source. Public visibility does not grant rights to copy, modify, distribute, host, resell, or create derivatives beyond the repository's license.
Third-party datasets, software, names, tournament references, and marks remain subject to their respective rights and terms. Their presence does not imply official status, sponsorship, endorsement, or affiliation.