What Wins DraftKings MLB GPP Contests: A 76,000-Lineup Study

We pulled the full opponent field from 24 DraftKings low-stakes, 20-max-entry MLB GPP contests played across 11 slate nights in the 2026 season — roughly 76,000 individual lineups — and joined every entry against actual box scores, DraftKings salaries, and final payout results. The goal was simple: stop guessing at what "good" GPP construction looks like and measure it directly from lineups that actually cashed.

The results reorder our own priority list. Team selection matters more than we expected, going contrarian on ownership showed no measurable edge, and punting pitching — a common low-stakes GPP habit — was actively costly. Below is what the data says, in the order it matters.

Why we built this dataset

Most GPP strategy advice is theoretical: correlation math, ownership models, or a handful of anecdotal winning lineups from a single big-field tournament. We wanted something closer to ground truth for the specific format we actually play — low-stakes, 20-max-entry classic contests — so we pulled full contest standings for 24 of them, reconstructed every opponent's roster from DraftKings' public lineup data, and joined each entry against that night's actual box scores, salaries, and Vegas lines. That gives us, for every one of ~76,000 lineups, exactly which players were rostered, what team(s) got stacked, how the lineup finished, and what it would have paid.

This is deliberately a low-stakes, multi-entry, classic-format sample — not single-entry, and not a large-field tournament with a different payout curve. Findings here should generalize well to other 20-max classic GPPs, and reasonably well to other multi-entry classic formats. They may not transfer cleanly to single-entry contests, where the practical takeaway around concentrating entries on top-total teams doesn't apply the same way, or to very large-field tournaments with a flatter, more top-heavy payout structure than what we describe below.

The payout structure only rewards the top few spots

Before construction strategy, the shape of the payout table matters. In a 20-max-entry field, 1st place pays roughly 15 times what a single entry costs across a full 20-entry block, and the payout curve falls off fast from there. By around 10th place, a cashing entry is already close to break-even relative to its share of a 20-entry outlay. Essentially all of the positive expected value in these contests lives in finishing rank 1 through 3.

That has a direct construction implication: a lineup built to "probably cash" in the min-cash sense is optimizing for the wrong outcome. The lineups worth building are the ones with a real shot at a top-3 finish, even at the cost of a higher bust rate on the rest of your entries.

This is a genuinely different objective from cash-game construction, and it's worth being explicit about the trade-off. A cash lineup should minimize the odds of an unusually bad outcome. A GPP lineup in a field shaped like this one should be willing to accept a higher chance of scoring near zero payout in exchange for a meaningfully higher chance of a rank-1-through-3 outcome, because that's the only place the format actually pays you for being right. Building one lineup that tries to do both jobs at once tends to do neither well.

Team selection dominates everything else

This was the single largest effect in the data. We ranked every team on the slate by actual runs scored (a proxy for the market's pre-game read on scoring environment) and checked how often a stack of that team appeared in a top-10 finish.

Lineups that stacked the slate's #1 run-scoring team hit the top 10 at a rate of about 2.2% per entry. Lineups stacking a team ranked 7th or worse in scoring hit the top 10 at about 0.06% per entry — a roughly 35x difference in top-10 hit rate, purely from which team got stacked. Sixty-seven percent of all top-10 finishers stacked one of the slate's two highest-scoring teams.

Put plainly: picking the right team to stack contributes more to a GPP result than almost any individual player decision inside that stack. Getting the team wrong is very hard to make up for with better player selection within it.

This also reframes what "player research" should spend its time on in this format. Debating the 4th-best hitter on a mediocre-total team versus the 5th-best hitter on that same team is a marginal decision relative to whether that team should be stacked at all. A disproportionate amount of GPP prep time in low-stakes MLB pools goes into exactly that kind of marginal, within-team player ranking, when the far larger lever — which team's offensive environment to attach your lineup to — gets comparatively little attention.

A real second stack roughly doubles your shot

Among lineups that had already stacked a top team, we compared those that also carried a genuine secondary stack — two or more hitters from a second team — against those that didn't. Adding a real second stack roughly doubled the conditional rate of a top-10 finish, holding primary stack quality constant.

This matches the batting-order correlation logic MLB DFS is built on: a single stack correlates its own outcome to one team's offensive game, but a second stack diversifies which specific game environment you need to run hot, without giving up the correlation benefit inside either stack.

It also explains why the modal top-100 stack shapes below cluster so heavily around two-team builds rather than a single mega-stack or a fully spread-out roster. A lineup with 5 hitters from one team and nothing else is fully dependent on that one offense having its night; if it doesn't, there's no secondary path to a big score. A lineup spread across four or five different teams gets almost no correlation benefit from any of them. The two-team shape sits at the point that captured most of the top-10 finishes in our sample — concentrated enough to benefit from real correlation, diversified enough not to be entirely dependent on one team's outcome.

Winners were not contrarian — they carried market ownership

A persistent low-stakes GPP myth is that differentiation from the field is what wins. Our data does not support that. The median top-3 finisher's summed lineup ownership (the sum of each roster spot's field-ownership percentage) was 74%, essentially identical to the field median of 72%. Winners weren't avoiding chalk — they were on the same players as everyone else, just more often on the right team.

We also isolated "leverage" plays — hitters rostered by under 5% of the field — and compared how many sub-5%-owned bats appeared in winning lineups versus the field. Winners averaged 2.44 such players; the field averaged 2.33. That gap is small enough to be noise, not signal. Deliberately rostering unpopular players purely for differentiation did not produce a measurable edge in this sample.

We think this result gets misread in DFS strategy discussion generally. "Leverage" is real when it's a byproduct of being right about something the field is wrong about — a team the market is under-pricing, a hitter about to move up in the batting order, a platoon advantage the field hasn't priced in. It stops being an edge the moment it's manufactured for its own sake, by rostering a worse player simply because he's owned less. Our data lines up with the first kind of leverage mattering and the second kind not paying off at all, at least in this sample.

Winners bought pitching

A common low-stakes-GPP habit is punting the pitcher slot to jam another bat into the lineup. The data argues against it. Median pitcher-slot output among top-10 finishers was 45.5 DraftKings points, rising to 51.2 points among top-3 finishers specifically — well above the field median pitcher output of 34.7 points. Winning lineups were not surviving despite a cheap, low-output arm; they were winning partly because of a real one.

The gap between top-10 and top-3 pitcher output (45.5 versus 51.2) is itself informative — it's not just that a decent pitcher is table stakes to compete, it's that pitcher quality kept separating winners even among lineups that were already good enough to finish in the top 10. Salary spent on a real starter, in this sample, kept paying off at the margin rather than hitting diminishing returns quickly.

Modal winning lineup shapes

We classified every top-100 finisher's stack composition by how many hitters came from each team represented in the lineup. Three shapes dominated:

Together, those three two-team stack shapes accounted for over half of all top-100 finishes. Diffuse, one-off hitter selection across three or more teams was comparatively rare among top finishers.

Duplication was a non-issue in this format

In 20-max-entry classic fields, we worried duplicate lineups might be capping upside — if many entries land on an identical build, they split whatever payout that rank earns. In practice, this wasn't a meaningful factor: winning lineups in our sample averaged only 1.2 copies each. Classic MLB fields at this size are diverse enough that near-consensus builds rarely collide in a way that matters. This is a real difference from small-field showdown formats, where duplication can be severe.

Combined with the ownership finding above, this is a genuinely useful piece of news: it means you can build the "obvious," market-consensus version of the best team stack on the slate without much fear of splitting a top prize with a dozen identical opponents. In this format, that risk is small enough not to justify forcing artificial differences into an otherwise correct lineup.

What this means for entry allocation

Put the payout-structure finding and the team-selection finding together and a practical allocation rule falls out: since almost all the expected value sits in ranks 1 through 3, and team selection is the biggest lever on getting there, multiple entries are better spent concentrated on a small number of the slate's highest-total teams than spread thin trying to "cover" every plausible outcome. A 20-entry set that hedges across six or seven different team stacks is diluting exposure to the two or three environments that our data says actually produce winners.

This is one season's sample from low-stakes DraftKings MLB GPP contests — 24 contests across 11 slate nights. Variance dominates any single-slate or single-contest outcome, and low-stakes field composition may not generalize perfectly to higher-stakes contests. We'll continue expanding this dataset and will revise these findings if a larger sample changes the picture.