Updated Sep 9, 2026 · Week 1 board.
Studio's first slot offers eight answers. The game total is one of them, under the name Total over/under, and it predicts whether a game total lands over or under the posted number. Choosing it changes what the other seven slots should hold.
A totals model is a pregame model
Total over/under belongs to the same family as the spread pick. Both are pregame targets, so both live under the same honesty law.
That law rules the closing line out. The closing price arrives after the moment a build has to call the game, so feeding it back would be reading tomorrow's paper. Settlement data is barred under a second law.
Two sports, not four
The board carries four sports across the top: NFL, MLB, NBA and NHL. Total over/under runs for two, NFL and MLB.
Check that first. An NBA board looks identical until you open Target and the option you came for is not there.
Scope is where a totals build gets specific
The second slot, Scope, decides which games a build learns from. Twelve options sit there, and three of them speak to scoring conditions.
Dome or outdoor filter splits games by venue type. Weather severity bands sorts them by how rough the conditions were. Season window filter holds training to a chosen range of seasons.
Narrowing scope costs sample, and sample is the check that turns red first. A totals build trained on cold outdoor games is a specialist, and a specialist still needs 200 graded picks before its record says anything.
Seven steps to a trained totals model
- Open Studio from the top nav. The board arrives with eight slots and nothing decided.
- Pick your sport. Choose NFL or MLB, the two that carry this target.
- Decide the first slot, Target. Choose Total over/under.
- Add the scoring inputs by hand, in the Features slot. The two that matter are named below.
- Fill whatever is left. Auto-finish drops its strongest prior into every slot you skipped, and Collapse best works through them singly.
- Press Train model. It fills any slot still open, then runs on your own machine at a cost of $0.
- Name it in the Name this model box, save it, then open the saved model and read its six trust checks.
Step 4 comes before step 5 on purpose. Auto-finish will happily fill Features on its own, and the two scoring inputs are worth choosing yourself.
The six templates on the board leave totals out entirely. They cover a spread build, two prop builds, a daily-fantasy build, a sealed ensemble and a flat-stakes baseline. Auto-finish is the only shortcut left.
Two inputs that are about scoring
The Features slot holds fourteen options. Two of them speak to how many points a game produces.
Pace and total environment feeds team pace and scoring-environment signals. Weather conditions feeds game-time weather. Neither one is exotic, and a totals board without them is guessing at the quantity it was built to predict.
Much of the rest of that catalog is aimed elsewhere. Snap usage rate and the defense-versus-position input exist for player props. Ballpark factors and pitcher mix go dark on an NFL board and carry real weight on a baseball total. That leaves a short list to work through, which is a mercy on a first build.
One option to leave alone at the start is Market price feed. It carries the book's pregame price, and a totals model that leans on it will track the posted total instead of pricing it.
Why a scoring board cannot be your totals model
Fantasy numbers look close enough to team scoring to tempt a shortcut. Take three of the highest-scoring names carrying no injury tag on the Week 1 projection board. Jahmyr Gibbs sits at 22.5 points, Bijan Robinson at 19.28 and Josh Allen at 19.26.
Add them and you get 61.04. That figure is not a game total and never will be.
Those are fantasy points, counted with a reception bonus, spread across three names on three different sides. A game total counts what a scoreboard shows for two teams. Two different questions, two different answers, and that gap is why a totals build has to be trained rather than summed.
What the trust report says about a totals build
The six checks do not change when the target changes. Sample size still wants 1,000 graded picks for green and turns red under 200. Correlation with the market still turns red over 0.97.
The rolling record line keeps its against-the-spread label whatever you built, so read it as a recent hit rate. The publish gate is unchanged too, and it waits for 20 forward picks to grade.
That gate is the reason a totals build takes a month to prove anything. Nothing about the target shortens it, and the public record is where the answer eventually shows up.
The Receipts Drawer
A totals build is the cheapest way to find out whether the board's guidance is doing your thinking. The spread template hands you a working board in one click, and a totals target hands you an empty one with eight decisions in it.
Take the second path once. Deciding Features by hand, for a target you chose yourself, teaches more in twenty minutes than a week of clicking Auto-finish.
Then leave it alone. Save it, watch the sample count, and check the trust card again after 200 graded picks. Everything on that card before then is noise wearing a number.
FAQ
Which Studio target builds a totals model? Total over/under, the eighth option in the Target slot. Studio describes it as predicting whether a game total lands over or under the posted number. The other seven cover spreads, moneylines, player props, teasers, daily fantasy, start-sit calls and closing-line value.
Can I build a totals model for basketball or hockey? Not today. Total over/under runs for NFL and MLB only. The board still offers four sport buttons, so check that your sport carries the target before you spend an evening on the other seven slots.
Is there a one-click template for totals? No. Six templates sit on the board, and totals is not among them. Decide Target yourself, then let Auto-finish drop its strongest prior into whatever you skipped.
Which inputs actually matter for a total? Two of the fourteen Features options speak to scoring directly. Pace and total environment feeds team pace and scoring-environment signals, and Weather conditions feeds game-time weather. The rest are worth testing later.
Why does my totals model still show an ATS line? The trust report grades the same six checks whatever target you pick, and its rolling record line keeps the against-the-spread label. Read it as your recent hit rate, and read the sample count above it first.
