Go wider.
Look deeper.
A strong lineup can only make your shortlist if you find it. A promising ROI estimate becomes more useful when you measure it with less simulation noise. SimSlate gives you room to do both: up to 100,000 generated field lineups and up to 100,000 simulated versions of the slate.
Candidate lineups
The roster combinations you evaluate and can choose from. A broader pool gives the search more chances to find useful stacks, pivots, and salary combinations.
Modeled field entries
The crowd your lineup competes against. Ownership, repeated builds, and payouts matter. Set the field to represent your target contest.
Simulation draws
How many possible slates you play out. Each draw scores the lineups together. More draws help you estimate the model’s average return with less sampling noise.
How this works in SimSlate: the generated field also supplies the lineup pool you rank and select from. Increasing Field size expands that pool and changes the modeled competition; Sim mode sets the number of slate draws separately. Repeated rosters can occur, so 100,000 entries does not promise 100,000 unique builds.
Find possibilities a small pool can miss.
GPP lineups have many routes to a ceiling: a different stack, a lower-owned teammate, a bring-back, or a salary pivot. A lineup with attractive model-estimated ROI cannot reach your shortlist if it never enters the pool.
A broader search can uncover better estimated opportunities. It still samples the available combinations; it does not prove you found the best possible lineup.
One more route to the top.
A small pool might contain only the first build. A larger pool might include the alternative. The simulator can then compare their modeled payouts and duplication.
Illustration only. The alternative may score better or worse; no performance result is implied.
Measure the opportunity more clearly.
One simulated Sunday can swing a lineup’s return. Thousands of additional Sundays give those swings more room to average out. For a fixed model and contest, more independent draws reduce the sampling noise in estimated average payout and ROI.
Moving from 2,000 to 100,000 draws gives you 50 times as many observations. Under the assumptions shown here, the standard error is about 7.1 times lower. That describes precision within the model.
Free runs use 1,000 field entries and 2,000 draws. Paid supports up to 100,000 of each. Because changing the field changes the question being measured, the illustration is not a measured Free-versus-Paid accuracy improvement.
| Slate draws | Relative standard error |
|---|---|
| 2,000 | 100% (baseline) |
| 10,000 | 44.7% |
| 50,000 | 20.0% |
| 100,000 | 14.1% |
Spend more time in the tail.
Top-heavy GPPs pay heavily for unusual combinations of big performances. If a modeled outcome is rare, a short run may barely observe it. More draws give you more chances to see how those modeled scenarios affect payouts.
That matters for evaluating upside, but even 100,000 draws can leave meaningful uncertainty around very rare events. More simulations cannot create a scenario that your model leaves out.
Same rare event. More chances to observe it.
- 2,000 draws
- 2 expected observations
- 100,000 draws
- 100 expected observations
At 2,000 independent draws, there is about a 13.5% chance of seeing none. At 100,000, the event-rate estimate still has about 10% relative standard error.
Scale helps when the contest makes sense.
Estimated ROI comes from modeled payouts relative to the entry fee. It depends on projections, player outcome ranges and correlation, opponent ownership, duplicate prize sharing, and the payout shape.
Match the field to the contest you want to evaluate. A 100,000-entry model can be useful for a large GPP. Inflating a smaller contest to 100,000 changes the competition and payout economics. A larger field is not automatically a more realistic one.
More draws sharpen estimates under those assumptions. They do not repair biased projections or guarantee real returns. A wider search can also select a lineup that looks good partly through sampling luck; reevaluating finalists with fresh draws is a sensible way to check them.
Start smaller. Scale with purpose.
- Explore quicklyUse a sensible smaller field and Quick mode to learn the workflow or investigate an idea. Paid Quick mode uses 10,000 draws.
- Look deeperUse Full (50,000 draws) or Max (100,000) when sampling noise could affect a close decision, especially in a top-heavy contest.
- Expect diminishing returnsFour times as many independent draws halves standard error. Doubling from 50,000 to 100,000 reduces it by about 29%, for the same modeled quantity.
Choose field size for the contest, then choose draws for the precision you need. 100,000 is a useful option, not a requirement for every decision.
A little more about the math and its limits
The standard error of a sample mean is σ/√N for independent draws with finite variance. Correlated player scores within one simulated slate are compatible with that assumption; dependence between successive slate draws changes the effective sample size. See the Stan reference on effective sample size.
At full requested scale, 100,000 lineups × 100,000 slate draws represents up to 10 billion lineup-score evaluations. The lineups share the same simulated slate in each draw; those evaluations are not 10 billion independent worlds.
Choosing the largest estimate from many noisy candidates can favor sampling luck. Fresh-draw reevaluation is a recommendation here, not a claim that SimSlate automatically performs an independent validation step. See Cawley and Talbot on selection bias.
Give your shortlist room to improve.
Start with a real 1,000-lineup field simulation, no card required. Use SimSlate’s built-in projections or upload your own. Upgrade when you want broader exploration and more simulation draws.