EditorialsDraft strategy
Does the best draft strategy change from year to year?
The best draft strategy does change from year to year, by about as much as three other things nobody puts in the same sentence. Draft advice is written for one league whether it says so or not: 10-team PPR redraft, one quarterback. Take one ordering and draft it in that league and in 38 others, across 2 real seasons, 5 sizes and 4 formats, scoring every roster against what actually happened. Its advantage over a field with no model runs from -5 to 12.1 points of a perfect board, and is negative in 11 of the 40. Changing only the season moves it 5.6 on average. Changing only whether the league starts a superflex moves it 5.5. The season is one term of four, not the term.
Every figure on this page is HexValue, the cross-position number HexMetrics assigns a player once it knows a league's shape. Two boards are built from the same 553-player pool for the 2026 season, identical except for the one setting under discussion, and the difference is what you see. Both are 12-team leagues. Percentages are suppressed for players below a HexValue of 5, because a ratio against a player worth almost nothing reports the size of the denominator rather than the size of the change.
The league nobody names
Every August the same sentence gets written a few thousand times. This is a good year to wait on running back. This is a year the tight ends are thin. Nobody says which league they mean.
They do not have to, because everybody already knows. 10-team PPR redraft, one quarterback. It is the default on the big platforms and it is the shape a positional ranking is built against unless it announces otherwise. Call it the assumed league. Almost every piece of strategy advice you will read this month is written for it, and almost none of it says so.
In that league, over the two seasons that can be measured, a board that prices value over replacement was worth 3.2 points of a perfect board against a field with no model in 2024, and -1.6 in 2025. One year it worked. The next year the same approach, in the same league, was slightly worse than doing nothing clever at all.
That is the whole problem in two numbers, and it is not the interesting part. The interesting part is what happens when you leave the assumed league, which most people already have.
What is being measured, and in what unit
A board is an ordering. A draft off that ordering is a roster. A roster under a league's real lineup rules scores real points against a season that actually happened. That is a falsifiable thing to measure, and it is what every number here is.
EDGE is the one to hold on to: what a value-over-replacement board scored as a percentage of a perfect board, minus what the field scored on the same measure. The field drafts off prior-season points per game, which is the ordering a reasonable person without a model would use. So the edge is how much better than nothing the approach was. 40 drafts, each averaged across every seat so a draft slot cannot flatter or bury a result.
Percentage of a perfect board rather than raw points, and that choice does real work. An 8-team roster is deeper than a 16-team roster and scores more, so a point difference would read as a league-size effect when it is really an arithmetic one. Measuring in points would have manufactured the finding this piece set out to test.
The four variables, side by side
Hold everything still, change one thing, see how far the edge moves. Averaged over every combination of the others, that is what each of the four is worth. The last column is the harder claim: how often changing that one thing took the approach from beating the field to losing to it.
| Change only this | Levels tested | Mean swing in edge | Sign flips |
|---|---|---|---|
| The number of teams | 5 | 11.1 | 6 of 8 |
| The season | 2 | 5.6 | 9 of 20 |
| Whether a superflex starts | 2 | 5.5 | 7 of 20 |
| Whether tight ends get a per-catch bonus | 2 | 2.7 | 4 of 20 |
Read that as it stands and the answer looks settled. League size moves the edge about twice as far as the season does, so ask how many teams you have and stop asking about the year. That is the article I expected to write. It is wrong, and the reason it is wrong is in the second column.
The control, and what it actually shows
League size was tested at 5 levels and the season at 2. A range across 5 samples is wider than a range across 2 for no reason except the counting. The two rows are not comparable and the ordering between them is a fact about the grid, not about football.
Cut league size to two levels and it becomes a like for like. Take its extremes, 8 and 16 teams, and the ranking reverses: the season moves the edge 6.1 and league size moves it 3.5. That is the correction most people would stop at, and it is one pair out of 10.
The pair is a choice, and the edge is not a straight line in team count, so the choice decides the answer. Across all 10 two-level pairs, league size moves the edge between 2.8 and 8.8, and the season between 3.6 and 8.2. League size is the larger of the two in 5 of the 10 and the smaller in 5. Compare 10 and 14 teams and league size wins at 8.8; compare 10 and 12 and it collapses to 2.8. Same grid, same season axis.
So the finding is not that your league matters more than the year. It is that the season, the size and the flex rules all move the answer by a similar amount, and every one of them is large next to the edge itself, which averages 3.1 across the grid. An approach worth 3.1 points that swings 5.6 with the year and 5.5 with a single lineup slot is not something to describe with one adjective.
Both the uncorrected table and the single-pair reversal are published rather than quietly replaced, because each is what this measurement looks like when you stop one step early, and the second one is a step this piece had already taken and called a control. Two axes whose ranges overlap almost completely is a duller sentence than the ranking reversed, and it is the one the grid supports.
What the same board actually ends up with
An edge is a summary. The roster is the thing itself, and the same ordering builds a different one depending on the league it is dropped into. Below is what it came away with in 2025, in a superflex league, at each size, out of 14 picks.
| Teams | QB | RB | WR | TE | Edge |
|---|---|---|---|---|---|
| 8 | 3 | 2.25 | 6.75 | 2 | -2.3 |
| 10 | 3 | 5.9 | 3.1 | 2 | 11 |
| 12 | 2.58 | 6 | 3.42 | 2 | 3.5 |
| 14 | 0.79 | 6 | 5.21 | 2 | -5 |
| 16 | 0.69 | 6 | 5.31 | 2 | 1.7 |
The quarterback column is the story. In the small leagues the board fills its superflex seat and its bench behind it, 3 quarterbacks out of 14 picks. By 16 teams it comes away with 0.69, which in a format that starts two of them is a roster that cannot field a legal lineup without help from waivers. Nothing about the board changed. The league got big enough that 32 starting quarterback seats were chasing a pool that runs out, and a board that will not pay above its own valuation simply got outbid. The receivers are where the picks went instead.
Read the other three columns with the caveat, not without it. Every drafter here plays under positional caps, and they bind: tight end sits exactly at its cap of 2 in all 40 cells, and running back at its cap of 6 in 28 of them. Where a column does not move, that is roster construction holding it still rather than the league declining to move it. Quarterback and receiver are the two with room to breathe, and they are the two that breathe.
What to do with this on draft day
Stop treating the year as the qualifier. A recommendation that opens with this season and never mentions how many teams you play with, or whether you start a second quarterback, has named one term of four and hidden the rest inside the word this.
Worth seeing once: in the widest single case in the grid, holding the season and the format still and changing only the number of teams moved the edge 16 points, from -5 to 11. Same board. Same year. Same rules. It went from clearly behind a field with no model to clearly ahead of it.
The practical version is short. Before you take any positional advice at all, know your league's size, its flex rules and its scoring, because each is worth about as much as the year the advice was written for. Where a piece does not say which league it is for, it is describing an average across leagues, and the spread around that average is wider than the effect it is recommending. If your league is 10-team PPR redraft, one quarterback, the advice is probably aimed at you. If it is not, you are reading somebody else's mail.
What this does not show
There are only 2 target seasons, and that is the real limit. A board here has to be built from 2 strictly prior seasons, which leaves 2024 and 2025 as the only two that can be scored without building one out of a single year. Two levels can show that the season moves the answer. They cannot describe the shape of that movement, and a trend read into two points is not there.
These are also not HexMetrics boards. Reconstructing our valuation as of a past preseason would need the player pool, the market data and the advanced stats as they stood then, and running today's versions against an earlier year is look-ahead dressed up as a backtest. The orderings here are ones buildable from strictly prior seasons, so the claim is about approaches rather than about our board.
The positional caps are a modelling choice and they are stated rather than buried: RB 6, WR 7, TE 2, in 14 rounds. Quarterback is the one that moves with the format, 2 where the league starts one and 3 where it starts a superflex, which is why the roster table above can show 3 of them. They exist so a board cannot draft nine quarterbacks because its metric happens to like them, which is a fair constraint and still a constraint. A different set would move the roster columns and would not move the edge columns much, since a capped board and its field are capped alike.
Lineups are set by each board's own ranking, which is what a manager can actually do. The alternative, setting each week's best legal lineup with hindsight, was measured too and is not used, because it pays for bodies: every extra healthy player is another guaranteed lottery ticket, and under that rule a board that simply drafts durable players cannot lose. Reporting the flattering rule would have been a choice about which answer to show.