Fair questions, straight answers

A baseball analytics platform: a plain-English stats search over 20+ seasons of MLB data, a plate-appearance-level Monte Carlo game simulator, and model-priced markets shown next to live Kalshi quotes — with every model call graded in public.

No — deliberately. Search is a deterministic grammar over our own corpus: the same question always returns the same, checkable answer, and every answer links the game log it came from. No hallucinated stats.

MLB today, in depth. Other leagues are on the roadmap; we'd rather do one sport honestly than five sports shallowly.

No. Search, game pages, team and player pages, live scores and the sample simulation are open to everyone. An account adds simulations (3/day free) and one unlocked model pick daily; Pro unlocks the full board.

Licensed sports data feeds (games, box scores, plays, pitch-level data from 2002 onward), MLB's public probables and lineups, National Weather Service park forecasts, sportsbook consensus lines, and Kalshi's public market data — all ingested into Mixl's own store.

Live games update every few seconds and are merged into the corpus within seconds of the final out. Lines refresh through the day; models retrain nightly.

Models are only ever evaluated on data from after their training window — the way real prediction works. It's the difference between a backtest you can trust and a highlight reel. A challenger model replaces a champion only by winning out-of-sample.

That disagreement is the product. We show our number, the de-vigged book consensus and the Kalshi price side by side, in percentage points — when we're with the market you'll see that too.

Calls are logged before first pitch and scored against settlement, misses included. Settled calls stay public — the record page is append-only by design.

It plays full games plate appearance by plate appearance: each PA is an eight-way outcome draw from the batter's rates composed against the pitcher's, with starter leashes, bullpen handoffs, baserunning and extra innings. Do that ten thousand times and you get a distribution, not a guess.

It's a real captured run with a fixed random seed, replayed — so every visitor sees the same genuine engine output. Live runs draw fresh games each time (and Pro can pin seeds to reproduce a run exactly).

Estimates converge with the square root of the run count: 1,000 games gets you within a few points, 10,000 within about one. The convergence chart in every run shows the band tightening as games stack.

Any two teams, yes. Editing the nine lineup slots and swapping starters is a Pro feature.

Everything public (search, pages, live scores, the demo), plus with a free account: 3 simulator runs a day at up to 1,000 games each and one fully unlocked model pick daily.

The full model board (our probability, edge and pick on every market and prop), 50,000-game simulations, lineup and starter editing, pinned seeds and saved runs.

Stripe checkout, monthly or yearly. Cancel from the billing portal anytime; access runs to the end of the paid period. Your price is grandfathered — later price changes never touch an active subscription.

No. Mixl doesn't take bets or hold your funds. Market prices shown are from Kalshi, a CFTC-regulated exchange; if you trade there, that relationship is between you and Kalshi. Optional account connection is read-only and your keys never leave your browser.

No. Model output is research and entertainment — probabilities, not promises. Markets are hard, variance is real, and past model performance doesn't guarantee anything. 18+ only; if gambling stops being fun, seek help (1-800-GAMBLER in the US).

The bot side of Mixl trades demo money only while its public track record accrues — that's a hard gate, not a promise. Nothing it does touches user funds.