> AI Predictions
The AI Prediction Score is a probability — shown as a percentage — for each possible outcome of a market. A score of 70% means the AI thinks that outcome is roughly 70% likely.
It's not a single AI's guess. The score is the combined output of multiple AI models working together, and it updates as new information becomes available — for example when team news breaks or financial markets move.
Use the score as one input when deciding where to place your stake. It's a second opinion, not a guaranteed tip.
Eroteme runs an ensemble of five large language models: Claude (Anthropic), GPT-4o (OpenAI), Gemini (Google), Grok (xAI), and Perplexity.
Using multiple models is more reliable than relying on one. Each model has different training data and reasoning patterns, so when they all agree on an outcome, that consensus is a stronger signal than any single model on its own. When they disagree, that disagreement itself tells you the outcome is genuinely uncertain.
Each model is weighted differently depending on the prediction category. Models with better track records in sports get more weight on sports markets; models stronger in financial reasoning get more weight on crypto and stocks.
Before any prediction is generated, Eroteme pulls live data relevant to the category — this is called the data enrichment layer.
For sports predictions, the AI sees current team form, head-to-head records, injury reports, and bookmaker odds. For crypto and financial predictions, it pulls real-time price data and technical indicators. For political predictions, it pulls polling averages and prediction market consensus.
For example, if you ask the AI about a Premier League match, it will know who's injured, what the current odds are at major bookmakers, recent form, and whether the home team has any tactical advantages. It then weighs all of this against historical patterns before producing a probability.
The AI is calibrated, not infallible. A 70% prediction means the outcome should happen roughly 7 times out of 10 — not every time. The AI will be wrong on some predictions. This is normal and expected.
Accuracy varies significantly by category. Sports and financial markets have rich, well-structured data and the AI performs meaningfully better than chance there. One-off world events with limited historical patterns are harder, and the AI is more uncertain.
Eroteme tracks Brier scores — a standard scoring rule that measures how well a probability prediction matches reality — to monitor calibration over time. Lower is better. We publish this on the accuracy page so you can see the AI's actual track record by category.
Treat the score as one input to your decision, never as a guaranteed tip.
The Convergence Score is the weighted average of every AI model's individual probability for the same outcome. Models with stronger track records in a given category get more weight in the average.
"Convergence" means how much the models agree with each other. A high Convergence Score means most or all of the models point to the same outcome — this is a strong signal. A low Convergence Score means the models disagree, which usually means the outcome is genuinely uncertain and you should be cautious.
When you see a 78% Convergence Score with all five models aligned, that's much more meaningful than a 78% score where the models are split 60/40.
Yes. No AI system can predict the future with certainty — and Eroteme doesn't claim to.
Even the best-calibrated prediction systems have significant error rates. The Convergence Score reflects uncertainty, not certainty. A 70% score is the AI's honest estimate that the outcome will happen 7 times out of 10 — which means 3 times out of 10 it won't.
This is why Eroteme publishes accuracy data openly and tracks Brier scores. We'd rather be honest about the AI's limitations than overclaim and lose your trust.
Scores update as new data becomes available. For sports markets the AI re-runs when team news breaks or odds move significantly. For crypto markets it updates as price action and on-chain signals change. For political markets it updates as new polls publish.
The exact frequency depends on the category — fast-moving markets like crypto can refresh multiple times an hour, while a long-range political market might only refresh once a day.
The score shown at the moment you confirm your stake is the score locked in for your position. Later updates don't change anything that's already been bet.
Eroteme actively monitors for systematic bias using Brier score calibration across outcome types. If the AI is consistently overestimating one side — for example, always picking favourites in sports — that shows up as poor calibration on a particular outcome bucket and the model weights are adjusted to compensate.
We're honest that no system is perfectly unbiased. Training data has biases. Models inherit them. The best we can do is measure carefully and correct what we find.
The AI currently covers six categories:
- Sports — football, basketball, MMA, esports
- Crypto and Finance — token price targets, market events
- Politics and Elections — election outcomes, legislative votes
- Esports — major tournaments
- World Events — geopolitical and macro events
- Entertainment — awards, releases, cultural moments
The AI is most confident in Sports and Crypto, where data signals are richest and patterns repeat. It's least confident in one-off World Events and Entertainment, where there's no historical baseline to calibrate against. Convergence Scores in those categories tend to be lower and more uncertain — and that's by design.
The AI has real advantages: it processes data faster than any human, it sees more sources, and it doesn't have emotional bias when a team you support is involved.
But it has real disadvantages too. It can't know things that aren't in its training data or live data feeds. It can misread context. It can miss late-breaking information. And it's calibrated — meaning it's designed to be roughly right on average, not always right on any individual market.
Treat the Convergence Score as a second opinion, not a replacement for your own judgment. The best predictors on Eroteme combine the AI signal with their own research.
> General
A platform fee of [X]% on winning bets. That's it.
We don't take a fee on stakes, we don't take a fee on losing bets, we don't manipulate the spread, and we don't bet against our users. The fee is the same percentage for everyone, applied at settlement, and clearly displayed before you confirm any stake.
Market results are sourced from official data providers — sports leagues, election commissions, financial data feeds — not from user reports. This means the resolution of a market is based on the same source of truth that bookmakers and news organisations use.
If you believe a result is wrong, you can raise a dispute via the results dispute page. Disputed markets are reviewed manually and either confirmed, corrected, or voided depending on the evidence.
Eroteme's smart contracts have been [audit status placeholder — e.g. "audited by [firm] in [date]" or "an independent audit is in progress"].
The full audit report and ongoing security disclosures are available on the security page.