About BBMI · independent sports modeling

Built by an actuary.
Judged by the record.

BBMI builds independent predictive models for MLB, NCAA baseball, football, and basketball — plus WIAA high school basketball. Nightly simulations produce rankings, playoff odds, and game forecasts; picks are one output among many. Everything publishes before the games and is graded after.

10,000 seasons nightly

Simulation engine

Every remaining game, simulated one plate appearance at a time — powering the rankings, playoff odds, and what-if scenarios.

Frozen before first pitch

Public forecasts

Game lines, totals, and win probabilities posted before play, never edited, graded against final scores.

The interesting questions

Research desk

Injury what-ifs, park effects, teams outrunning their talent — the analysis the box scores can’t give you.

How this started

It started with my kids — relentless askers of questions no box score can answer.

“How many championships would the Bulls have won if Jordan never retired?”
“What would the Yankees’ record be if they played their home games at Fenway?”
“How good would the Angels have been if Mike Trout stayed healthy?”

There’s only one honest way to answer questions like those: rebuild the season inside a simulation, change one thing, and play it out ten thousand times. The dinner-table speculation turned into a family NCAA bracket challenge, the bracket model worked better than expected, and the experiment got serious — it now spans 2,005+ documented NCAA basketball games, a full WIAA high school season, NCAA football and baseball models, and a per-plate-appearance MLB simulation.

I’ve spent decades as an actuary building predictive models for healthcare costs and revenue forecasting. The core disciplines — data quality, variable selection, calibration, out-of-sample validation — translate surprisingly well to sports. When the model’s forecasts kept landing closer to final scores than the market’s, the next step was obvious: publish them, track them, and find out if it was real.

The same engine now answers both kinds of questions — the what-ifs on the research desk, and the nightly forecasts for every game on the schedule.

Publish the forecast before the game. Record every result publicly. Let the cumulative record speak for itself.

How the models work

Every model forecasts games from the ground up — runs, margins, win probabilities — without ever seeing a betting line. When a forecast is compared against the market’s number, the gap between them is the edge: a measure of how strongly the model disagrees with the consensus.

edge = | BBMI line − Vegas line |

Lines routinely move 1–2 points between open and tip-off and vary across books, so a gap under 2 points is market noise, not signal. Only picks with edge ≥ 2 are counted in the performance record — everything below that threshold is excluded as a coin flip.

Basketball

Offensive and defensive efficiency, tempo, strength of schedule, home court.

Football

Efficiency ratings, opponent-adjusted box scores, quality wins, home field, altitude, weather-adjusted totals. Validated across two independent seasons with zero overfitting gap.

Baseball

MLB: per-plate-appearance Monte Carlo simulation with pitcher/batter rate models and park factors. NCAA: run scoring, ERA, pitcher adjustments, dynamic park factors.

WIAA

The same basketball framework, with wider error bars — high school stats are self-reported and noisier.

The validation record

Full logs →

Forecast accuracy is tested against the sharpest benchmark available — the betting market. These are out-of-sample results, logged publicly.

NCAA Basketball · Spread65.1%Premium (edge ≥ 6)Free (edge 2–6) 55.0% · 57.0% overall ATS · 2,005 gamesView the log →
NCAA Basketball · O/U62.2%Premium (edge ≥ 4)Free (edge 2–4) 55.0% · edge ≥ 4 pts · 1,279 picks · Live · 2025–26 seasonView O/U →
NCAA Football · Spread64.3%Premium (edge ≥ 6)Walk-forward (≥5 pts, 2 seasons) 59.7% · 544 gamesView the log →
NCAA Baseball · Spread57.3%Premium ATS · edge ∈ [2, 6] runs1,085 premium picks · Live · 2026 seasonView the log →
NCAA Baseball · O/U62.2%Premium O/U · edge ≥ 4 runsUnder 60.7% · Over 51.9% · 172 picks · Live · 2026 season · All picks (edge ≥ 1.5 runs): 54.7% on n=1,052 (575–477)View O/U →
NCAA Baseball · Moneyline+9.4%ROI (flat 1u)51.0% win rate · 835 picks · Live tracking from 2026-04-15 · edge ≥ 1ppView ML →
WIAA Basketball81.0%Outright winner accuracy5,996 games trackedView the log →
MLB · Away +1.5 Run Line (sim)In calibrationF9 RL picks (away +1.5 when home is the Vegas favorite) now generated by the per-PA sim (deployed 2026-05-23). Live track record building this season; prior MVM (FIP/wOBA) model results are paper-tracked off-site, not displayed here.View the log →
MLB · Over/Under (sim)In calibrationF9 O/U totals now generated by the per-PA sim (deployed 2026-05-23). Live track record building this season; prior MVM (FIP/wOBA) results are paper-tracked off-site, not displayed here. Over picks suppress before MLB Week 13 and on posted totals ≥ 9.0; UNDER fires on any UNDER-direction edge (single-tier free product, post-2026-05-23 corrected staircase).View the log →

Spread records count picks with edge ≥ 2 points (1.5 runs in baseball). Every number comes from the public log.

The transparency rules

See the full pick log →
01

Full pick log

Every game picked, every result recorded. No gaps, no selective omissions.

02

No retroactive edits

Picks publish before games start. The record cannot be adjusted afterward.

03

Edge breakdown

Performance is reported by edge tier, so you can see exactly where the model earns its keep.

04

Weekly summaries

Results by week, so you can verify it isn’t one lucky streak doing the work.

Where the betting market fits in

The model never sees the Vegas line when it forecasts a game — the market is a benchmark to validate against, not an input. But it’s worth understanding why an independent model can disagree with that benchmark and win. Vegas has to set a precise number. You only have to decide which side of it to be on — a fundamentally easier problem. And sportsbooks aren’t purely predicting outcomes: they manage liability, shading lines toward public money even when their own numbers say otherwise. A model focused purely on accuracy can exploit those gaps — especially in college sports, where markets are thin and hundreds of teams get little analytical attention.

Your advantages

  • Pick a side — never set the number
  • No liability to manage; pure accuracy focus
  • College markets are inefficient and data-rich
  • Public-money distortions create exploitable gaps

The house edge

  • The juice (−110) means ~52.4% just to break even
  • Vegas gets injury and lineup news faster
  • Sharp money pushes closing lines toward true
  • Variance is real — good models have losing weeks

The honest bottom line

No model wins every bet. The goal is to clear the 52.4% breakeven threshold consistently over hundreds of games, and to bet more when the edge is largest. The documented record on high-edge picks shows that’s achievable — with discipline, patience, and realistic expectations. If someone promises 70%+ on every pick, they’re selling something other than math.

Not a tout service

AspectBBMITypical tout
Track recordPublic, unedited, full historyCherry-picked wins, losses hidden
MethodologyDocumented analytical approachVague claims, no explanation
ConfidenceEdge scores show convictionEverything is a “lock”
Performance filterExcludes line-movement noiseCounts everything, even coin flips
Bad weeksLogged and visibleQuietly buried
Pricing$10 trial / $35 monthly$99–$299+ per month

Evaluate the record, not the marketing.

Model changelog

Picks freeze before games start, so a model update only ever affects future picks.

Major model updates are logged here as they happen. Because picks are frozen before games tip off, any methodology change only affects future picks — never historical results.

Future updates will be logged here as they are deployed.

See the record for yourself.

Every forecast, every sport, logged publicly. Judge it yourself.