Leverage is a difference, not a low number

Most people use "leverage" to mean "not many people have him." That definition fails on its own terms, and the failure is easy to demonstrate: under it, the fourth-string tight end projected for three points is the highest-leverage player on the slate, every week. Nobody owns him. He is also not a play.

Leverage is a difference between two rankings — how good a player is, and how much the field wants him. It is large when the gap between those two is large and positive. Low ownership is one half of the measurement, and on its own it is the less informative half.

Ownership is a conserved quantity

Start with arithmetic nobody has to take on faith. Ownership percentage is the share of entries containing a player. A DraftKings Classic entry holds nine distinct players. So if you add up the ownership of every player on the slate — all of them, not just the ones you like — the total is exactly 900%, no matter how the field behaves, what the projections say, or who gets hurt on Sunday morning.

That single fact reframes most ownership talk. The field's attention is a fixed budget. It is not "high this week" or "low this week"; it is 900% every week, redistributed. Fading a 35%-owned running back does not remove 35 points of ownership from the slate. It moves them onto other players, and the interesting question is always which ones.

It also gives you a rough consistency check on any ownership projection you buy or build. Sum the column. A source that scores players independently, rather than as competitors for a fixed number of roster spots, misses 900% by a wide margin, and it will cheerfully tell you four running backs are 40% owned on a slate that starts about two and a half of them. Every leverage call made from those numbers is then wrong in the same direction.

GameScript's own house model is built to satisfy that identity: one softmax per position group, each group's shares scaled to sum to the roster slots that group fills. Ownership is competitive by construction, so the model has to be too. The FLEX is the awkward slot, because no single group owns it, and the model spreads it rather than assigning it — more to running backs and receivers than to tight ends, since that is how the field actually fills it. What matters is that the spread adds to the one FLEX that exists. Nine slots, 900%.

It is worth saying that this check found a real error here, because the version of this article published on 2026-09-04 said so. The FLEX weights were written as 0.5 to running backs, 0.5 to receivers and 0.25 to tight ends — 1.25 slots for a slot that only exists once — which put the group targets at 9.25 and every published ownership figure about 2.8% high. The model's own validation could not see it, because it derived what to expect from the same table it was checking. The weights now carry the same relative split across one slot, and the totals are asserted against the roster rather than against the table.

The reason to run the check on your own source is that it is built to catch a source wrong by a lot — a column totalling 1,400% is scoring players in isolation — rather than to audit the last few percent. A uniform scale error inflates each group together instead of rearranging it, so the within-position ordering that leverage actually consumes survives it. That is what makes it survivable, not what makes it fine.

The ratio everyone uses ranks the wrong players first

The common leverage metric is projection divided by ownership. It is intuitive, and it is dominated by its denominator. Watch what it does to four hypothetical receivers — invented numbers, chosen to make the mechanism visible:

The ratio ranks D first by a factor of five, and cannot separate B from C at all. D is a player with no route to a usable score whom the field is correctly ignoring. Putting him at the top of a leverage list is not a quirk of the metric; it is what the metric measures. Any unbounded ratio with ownership on the bottom will hand its highest scores to the players nobody has a reason to roster.

The right-hand column is the alternative: projection percentile minus ownership percentile. It ranks B first, C second, D near zero, and A slightly negative — which is the ordering a person would give if they thought about it for a minute.

Why a bounded measure is the point

The difference of percentiles has a property the ratio does not, and the property is the whole argument for it: a player's leverage score can never exceed his projection percentile. Both terms live on the same 0-to-100 scale, and ownership percentile is never negative, so the largest score available to receiver D above is 8 — the exact number his projection earns him — and that is only if literally no one in the field rosters him.

Read that as a design constraint rather than a curiosity. It means the measure cannot be gamed by finding someone more obscure. A bottom-decile projection cannot post a big leverage number no matter how thoroughly the field ignores him, because the metric refuses to pay for obscurity on its own. What it pays for is disagreement: a good player the field has looked at and declined.

The scale is -100 to +100, in percentile points. GameScript's contrarian playbook treats +25 as the floor for calling a player a leverage buy, on the reasoning that twenty-five percentile points is about two deciles of gap — wide enough to survive swapping one projection source for another, and narrow enough that a real slate produces several. Below roughly fifteen you are inside the noise of whoever wrote the ownership projection, and you are measuring their opinion rather than the field.

Both halves belong within position

Raw ownership is not comparable across positions and neither is raw projection. Twenty percent is crowded for a tight end and unremarkable for a quarterback, whose position has a third as many rosterable bodies competing for the same share of the field. Twelve projected points is a strong receiver and a weak quarterback.

So both percentiles are computed within position. A cross-position projection percentile mostly re-sorts the pool by position, and a cross-position ownership percentile tells you a quarterback is chalk when he is simply a quarterback. The one place a slate-wide ownership percentile earns its keep is the different question of whether a player is crowded on this slate at all, which is worth knowing but is not the input to leverage.

A related trap: percentiles need a population. Over four players a percentile can only take four values, and rounding makes 12.5 and 37.5 look like meaningfully different standings when they are the same coin flip. Six is roughly the smallest group where the deciles a reader assumes are the deciles they get. This bites hardest on single-game slates, where a position group is often two players — and the honest answer there is to report no percentile at all.

What the field-relative edge actually buys

Ownership does not change a player's points. It changes how many other entries hold the same outcome you do. That effect is multiplicative across the roster, which is why it compounds much faster than intuition suggests.

Suppose you swap one 35%-owned player for an equally projected 12%-owned one. If lineup construction were independent across slots — it is not — the chance that another entry matches yours on that slot falls by a factor of about 0.35 / 0.12, roughly three. Do it twice and the factor is about nine. Do it three times and it is about twenty-five.

Treat that as a statement about shape, not a duplicate count. Real fields build lineups with correlated processes: everyone reads the same articles, the same optimizer defaults ship in the same tools, and the same three stacks show up thousands of times. Actual duplication runs well above the independent product. What survives the imprecision is the exponent: a roster's distinctness is the product of nine decisions, so a single well-chosen swap moves it more than a long argument about the ninth-best flex.

When the chalk is correct

Chalk is not a mistake. It is the field agreeing that a player is underpriced, and most of the time the field is looking at the same true thing you are.

In a contest paying a large fraction of entries — double-ups, 50/50s, head-to-heads — being different is a cost with no prize attached. Get the sign right, though, because the measure is a difference and the informative point is its zero. Agreement between the field and the projection is a score near zero — the two percentiles land in the same place. A strongly negative score is the opposite of agreement: it says the field ranks a player materially higher than your projection does.

What cash licenses is not a taste for negative scores but indifference to the sign. The player you want is the one with the high projection percentile, and if the field is crowded onto him too — a score near zero, near the top of both scales — that is the market corroborating your number, and the overlap that would cost you in a tournament costs you nothing here. A strongly negative score still earns a second look: it means the field is paying for something your projection is not pricing, which is either information you are missing or a player who is genuinely over-owned. In a 50/50 neither is disqualifying. Both are worth knowing before you decide.

In a top-heavy tournament, the useful question is not whether to eat chalk but which chalk survives contact. Two kinds show up every week and they behave differently:

• Chalk built on a role change. A back who just inherited the entire backfield, a receiver promoted to the slot. The field is on him because his usage genuinely changed, and the reason holds whatever else happens in the game. Eating this is usually correct, once.
• Chalk built on an injury that also caps the replacement. The starter is out, so the backup is 60% owned at minimum price — but the offense that made the starter good is the same offense now handing the ball to a worse player against the same defense. The field is buying the vacancy and ignoring the ceiling.

Zero chalk is its own error. A lineup with nine contrarian pieces has usually spent its entire projection budget on being different, and being different is not a scoring category.

Leverage you cannot spend twice

One more failure worth naming, because it looks like diversification and is not. Two 8%-owned receivers in the same game are not two independent leverage bets. They are one bet on that game scripting a particular way, held twice. If the game turns into a 13-10 slog, both die together, and your roster has paid two slots for a single outcome.

Independence is what makes low ownership worth owning. Prefer leverage spread across games — or, deliberately, one game you have decided to concentrate in, with the concentration stated rather than stumbled into.

The number under the number

Every sentence above assumes you have a projected ownership figure. Most weeks, most people do not — not a published one. What they have is a model's estimate of what the field will do, usually driven by salary and value, and a percentile computed against an estimate is a guess with a percentile on it.

That does not make it useless. An ordering is what these methods actually consume, and a decent model's ordering is far better than nothing. It does mean the claim you are entitled to make changes. "He is the sixth-best leverage play on this slate" is a defensible reading of a model's ranking. "He will be 6% owned" is a different sentence, and the model has not earned it.

GameScript publishes this distinction rather than hiding it: every pool's ownership carries a basis saying whether the numbers came from the account's own upload, from the house model's estimate, or from a mix — and the contrarian playbook refuses to run at all on a slate carrying no projected ownership, on the grounds that guessing at chalk is worse than saying you cannot measure it. That is the whole product claim here. The arithmetic above is yours whether or not you ever use it.