Built on math, not hype.
Kirkova exists because most sports "prediction" content is either a gut-feeling take dressed up as analysis, or a black box asking you to trust it blindly. We wanted neither — so we built a system that shows its work.
What we do
We train gradient-boosted ensemble models — XGBoost, LightGBM, and CatBoost — on historical NBA and MLB game and player data to produce win probabilities and player prop projections. No paid data feeds, no large language models generating plausible-sounding guesses. Just statistical models trained on real outcomes, retrained daily, and held to a public, auditable track record.
Why we show the losses too
Every pick on this site is logged and graded — wins and losses both — and the full record is public on the Performance page. A model that only shows you its wins isn't a model you can trust. Ours isn't perfect (no model is), but it's honest about exactly how good it is, tier by tier, in real time.
The confidence tiers
We don't treat every prediction as equally actionable. Picks are bucketed into Sharp, Lean, and Passtiers based on how confident the model actually is — so you always know whether you're looking at our strongest call or a coin-flip game we're showing purely for transparency. See the FAQ for the full breakdown.
Where we're headed
NBA and MLB are live today. NFL and NHL are next, alongside deeper player prop coverage and more transparency tooling around how each model version performs over time. This is an actively developed product, not a static report.