The Multi-Source Prediction Engine
With the SIRE LLM now delivering live, adaptive predictions, our next phase is to extend its capabilities with the Multi-Source Prediction Engine, a planned optimisation layer designed to combine and dynamically weight insights from multiple contributors in real time. Score (Subnet 44) will be one of the main contributors, with our Vision-language-action (VLA) and Player Value Function (PVF) models. These specialised models will provide high-resolution visual insights and player-level evaluations, adding unique dimensions to the prediction process. Our new approach leverages LLM to uncover orthogonal signals, hidden within massive context windows of data. This isn’t just an incremental improvement, it is a fundamentally new way of extracting value, designed for live, adaptive performance. As Score Co-Founder and “MicroPrediction - Building an Open AI Network” author, Peter Cotton puts it: “The LLMs we have now, we didn’t have six months ago. Statistics can be a form of reasoning, but it’s not the only one at our disposal.”
Planned Design
Information Intake: For each prediction target, the engine will open several slots. Each slot will be filled by an information packet from a contributor, containing signals relevant to the future outcome. Contributors can be models (including VLA and PVS), human analysts, or other data-driven systems.
Model Processing: The SIRE LLM will process all packets, interpreting both the individual signals and the relationships between them to produce forecasts.
Multiple Prediction Runs: Instead of a single forecast, the system will generate several independent predictions per target, with each run using a randomly selected sample of slots.
Outcome Tracking: Once the real-world result is known, the system will measure how much each packet contributed to predictive success.
Performance Scoring: Packets will be ranked based on historical profit and loss (PnL) providing a domain-agnostic measure of their value.
Dynamic Weighting: Packets with strong PnL histories will be up-weighted in future predictions. Weaker packets will be down-weighted or replaced over time.
Iterative Optimisation: This process will repeat for every prediction target, continuously refining the mix of contributors and improving overall forecast accuracy.
Why It Matters
Combines diverse intelligence sources in a single, adaptive framework.
Continuously self-improves based on live market outcomes.
Maximises long-term +EV by prioritising the most profitable contributors.
Scalable and future-proof, able to integrate new models, data feeds, or human experts without changing the core system.
The Multi-Source Prediction Engine will be the next leap forward for SIRE, moving from a single adaptive LLM to a networked, self-optimising intelligence system designed to stay ahead of increasingly efficient betting markets.
Last updated