WHY WE BUILT
A SPINE.

I · WHAT'S BROKEN

Three categories. Each missing one thing.

Touts sell picks with screenshots and selective memory. Cherry-picked weeks, hidden losses, fabricated ROI, winning streaks resold as proof of skill. The audience pays for nothing. Honest people stay away.

Verified-record platforms — Pikkit, Betstamp, Juice Reel, WinnerOdds — publish honest tracking. Good for sorting the touts from the real ones. They don't build a model. They report whose record holds up.

Model-sellers — OddsJam, Unabated, BetQL, Props.Cash — sell genuinely sharp math. Ten-thousand-sim engines, fair-line tools, projection optimizers. What they don't do is bet their own model. The math sits on a shelf next to the cash register.

The thing nobody combines: a deep simulation engine and a per-engine model-performance log, from the same entity. Model + log, under one roof — settlements the model is measured by. We started building it because nobody else was.

II · WHAT THE SPINE IS

The spine is not a person.

There's no capper, no analyst, no “trust me on this one” Twitter account. BetFly is a sabermetric simulation engine — a spine — that runs every MLB game five thousand times a night. It models confirmed lineups, pitcher quality, bullpen depth by team, defensive runs saved, catcher framing, umpire K-zones, and per-batter rates regressed toward true talent.

One spine, many engines. The spine simulates the game; eleven engines ride on top of it — each one surfaces edges in a specific market (home runs, strikeouts, hits, total bases, winners, totals, run lines, and more). Same model, different lenses. We track each engine independently — if one stops working, it comes off the slate without touching the others.

When the spine finds a gap between its number and the market, we publish it. Every edge gets graded by Kalshi's own settlement after the game. You see the same signal, the same window — and the same receipt the next morning.

III · WHAT WE WON'T DO

The lines we won't cross.

• We won't pad the slate. If the engine sees no edges, we say so. Your inbox doesn't get manufactured action because we need something to sell.
• We won't hide losses. Cold streaks ship in the same digest as the wins. The model log is everything we've called — every read, every settlement, graded by Kalshi.
• We won't claim guarantees. There are no locks, no hammers, no easy money. Every edge has variance, and we say so out loud.
• We won't sell tools we haven't earned. Every engine on this site has cleared a paper window with the gates that decide when it's ready to publish — or when it's ready to come off.

IV · WHAT WE WILL DO

Model + log. From one entity.

The two things the picks market never combines from the same entity. We do both — under one roof — and we let the settlements speak.

• Model. A 4-layer Monte Carlo MLB engine — 5,000 sims per game, 17 adjustment layers (form, Statcast, Stuff+, park, weather, opener detection, handedness splits) — that prices out every market we publish.
• Log. Every published read gets graded by Kalshi's own settlement market. Per-engine confirmed rate, paper-portfolio ROI, per-market calibration. What stops working comes off the slate — visible to every subscriber, every day.
• Calibration. Per-market shrinkage and XGBoost calibrators wired into the model's output, retrained against the settled record. The same evidence that grades the reads also tunes the engine that produces them.
• Discipline. New engines run in paper for two weeks with a [PAPER] tag before any subscriber sees them live. The NBA engine got disabled when its calibration turned. We retire faster than we ship.