// Data Science

DFS SIMS
THE WORD EVERY TOOL USES. THE THING ALMOST NONE OF THEM DO.

"Powered by simulations" is on every DFS tool's landing page in 2026. But there are three completely different architectures hiding behind that word — and they produce completely different lineups.

Michael  ·  August 2026  ·  11 min read

Two players sit down on the same Sunday. Both subscribe to a tool that advertises simulations. Both build 20 lineups.

One of them gets lineups that were graded by sims after a projection optimizer built them. The other gets lineups that were constructed inside the sims from the start.

Same word on the sales page. Structurally different lineups in the contest.

"Sims" has become the most overloaded word in DFS. Every serious tool claims them. Almost no tool explains where in the pipeline the simulation actually sits — and that placement matters more than the sim itself.

This article is the breakdown I wish existed when I started building my own sim engine: what DFS sims actually are, the three architectures on the market, and how to tell which one a tool really uses before you pay for it.

// The Foundation

WHAT DFS SIMS
ACTUALLY ARE

A DFS sim is a Monte Carlo engine that plays out the slate thousands of times. Instead of one projection per player — Judge 9.2, Ohtani 10.1 — every player gets a full probability distribution: floor, ceiling, and everything between, shaped by matchup, park, weather, Vegas lines, and game script. The engine then draws random outcomes from those distributions, over and over, producing thousands of complete versions of the night.

The critical ingredient is correlation. In real baseball and football, scoring clusters. When a team hangs 9 runs, four hitters go off together. When a game shoots out, the QB, his receivers, and the opposing passing attack all eat at once. A sim that models this — teammates rising and falling together, both sides of a game linked — reproduces the environments that actually win tournaments. A sim without it is a random number generator with extra steps.

If you want the full mechanics — distributions, correlation math, sim win rates — I wrote a deeper piece on how Monte Carlo simulation works in DFS. This article is about the part that piece doesn't cover: what tools actually do with those simulations.

THE QUESTION ISN'T WHETHER A TOOL RUNS SIMS. IT'S WHERE IN THE PIPELINE THE SIMS SIT.

// Architecture 1

SIMS AS A
REPORT CARD

// Type 1 — Sim-Graded Lineups

BUILD FIRST, SIMULATE AFTER

In this architecture, a traditional projection-based optimizer builds your lineups first. Then the simulation runs afterward, scoring each finished lineup across thousands of outcomes and reporting metrics like win rate, ROI, and duplication risk. This is the approach Stokastic is best known for.

To be fair: this is genuinely useful. Knowing that Lineup A wins 0.8% of sims while Lineup B wins 2.1% is real information, and grading is a big step up from flying blind.

But notice what the sim never did: it never chose anything. The lineups were already built by a points-maximizing optimizer before the simulation saw them. If the projection optimizer never assembles a particular correlated stack, the sim can never tell you it was the best construction on the slate — it can only grade what it's handed. The ceiling of your lineup pool is set before the simulation ever runs.

Sims Grade Optimizer Builds Post-Hoc Scoring
// Architecture 2

SIMS AS A
BLACK BOX

// Type 2 — Sim-Built, Machine-Controlled

THE SIM BUILDS. YOU WATCH.

The second architecture builds lineups inside the simulations — genuinely better. SaberSim pioneered this approach, and credit where due: constructing lineups from sim outcomes instead of static projections is the right core idea.

The trade-off is control. The philosophy of these tools is that the machine knows best: press the button, get lineups, don't touch the dials. Manual stack rules, exposure caps, and construction constraints are treated as user error — friction the tool tries to talk you out of. For some players that's a feature. Set it, forget it, fire the entries.

But if you have fifteen years of reads — you know this pitcher's velocity is quietly down, you know this offense punishes a specific defensive scheme, you know this game sets up for a bring-back — a black box has no slot for that knowledge. You either accept the machine's lineups or fight the tool to override them. The sim is doing the building, but you've been removed from the process.

Sim-Built Automated Limited Control
// Architecture 3

BUILT INSIDE THE SIMS.
CONTROLLED BY YOU.

// Type 3 — Sim-Built, User-Controlled

THE ARCHITECTURE I BUILT DFS ONLY AROUND

The third architecture keeps the good half of both worlds. DFS Only runs 100,000 correlated simulations per slate, producing complete game scripts — not just player scores, but entire correlated versions of every game. The optimizer then samples those simulated worlds directly, so every lineup is assembled from scenarios where its players actually went off together. Lineups are born inside the sims, not graded by them after the fact.

The difference from the black box: every construction lever stays in your hands. Stack shapes, primary and secondary stack sizes, bring-backs, exposure caps, player locks and bans, ownership and leverage targets — you set them, and the optimizer finds the best sim-backed lineups within your rules. Your reads get expressed through the engine instead of being overridden by it.

This is the whole reason I built the tool. I didn't want a report card, and I didn't want a machine that treats my fifteen years of reps as noise. I wanted lineups constructed from correlated simulations with my judgment still in the loop. The same engine runs MLB stacks and NFL stacks, with NBA on the way for the 2026-27 season.

100K Correlated Sims Sim-Built Lineups Full User Control Leverage Scoring
1
Sims grade lineups
2
Sims build, you watch
3
Sims build, you control
// Buyer's Checklist

HOW TO VET A SIM TOOL
BEFORE YOU PAY

Marketing pages won't tell you which architecture you're buying. These five questions will:

// Five Questions

ASK THESE BEFORE SUBSCRIBING

1. Do the sims build the lineups, or grade them? If the sim output only appears after lineups exist — as win rates and ROI on finished builds — it's Type 1. If lineups come out of the sim itself, it's Type 2 or 3.

2. Is the correlation real? Ask whether teammates' outcomes are linked and whether both sides of a game move together. Uncorrelated sims produce stacks that look diverse but never model the clustered scoring that wins GPPs.

3. How many simulations, and of what? Simulating final player scores is not the same as simulating game scripts. Full game-script sims preserve who scored with whom — that's what makes stack construction meaningful.

4. Can you enforce your own construction rules? Try to set a specific stack shape, cap an exposure, or force a bring-back. If the tool fights you or the settings are buried, you're buying a black box.

5. Does it sim the field, not just your lineups? Your win rate depends on what the other entries look like. A tool that models the tournament field — projected ownership, duplication, leverage — is answering the question that actually pays.

I've written honest, no-affiliate-link comparisons of how the major tools stack up on exactly these questions — pricing included — for both MLB optimizer tools and NFL optimizer tools. Where competitors are genuinely strong, those pages say so.

// The Bottom Line

SIMS ARE THE EDGE.
ARCHITECTURE IS THE DIFFERENCE.

Simulation is no longer optional for tournament DFS — the field got too sharp, and static projections lose to correlated variance every large-field Sunday. The real decision in 2026 isn't whether to use sims. It's which of the three architectures you want doing the work: a report card, a black box, or an engine that builds inside the sims with your hands still on the wheel.

I spent fifteen years playing DFS and years building the third option because the first two frustrated me as a player. If that architecture matches how you think about this game, the whole engine — 100,000 correlated sims, ML projections, leverage scoring, field simulation — is $39 a month, less than a single GPP entry fee for most regulars.

LINEUPS BORN INSIDE THE SIMS.
RULES SET BY YOU.

100,000 correlated simulations per slate. Full control over stacks, exposures, and construction. MLB and NFL live now.

Run Your First Simulation →