2025-26 Fantasy Hockey Projection Report Card

The short version

Who to trust for your banger draft, in three lines.

• The crowd beat everyone this year. The simple average of all the lists finished first under both scoring methods, ahead of every individual projector. Beating the consensus is famously hard, and this season nobody managed it.
• Of the individual projectors, Lineup Experts and Bangers Fantasy Hockey led the pack, though the whole top of the field is close enough to be a statistical tie. The Athletic and Kubota trailed it.
• Nobody can predict hits. Last season's hit total forecasts this season as well as any model does, so if you draft for hits you can skip the projections and just look up last year.

How we scored it, two lenses

The method, briefly. Every projection and every actual result is first converted to an 82-game pace, so we grade the quality of each rate projection, not whether a player stayed healthy. A star who missed 20 games to injury isn't docked for it, because in a real league you just replace him. Then we grade two ways, simplest first. Both run on the 10 entries that project all 7 categories (9 individual lists plus the crowd); the two lists that skip one (Dobber, no power-play points; Laidlaw, no penalty minutes) get fair same-category boards further down.

* Dobber doesn't project power-play points; Laidlaw doesn't project penalty minutes. Lens 1 ranks them on the 7 categories they do project. Neither appears in Lens 2, which needs all 7 to build one value score, so they race on fair 6-category boards further down instead.

The two lenses agree on the podium, the crowd then Lineup Experts and Bangers, and reshuffle the middle, which is exactly where getting the value order right pulls apart from nailing the raw numbers. Everything below goes deeper: position scarcity (VORP), consistency, the optimism tax, the bust-avoidance test, and a full per-category breakdown. The whole approach is built in the spirit of Apples & Ginos and Hockey Pool-Aid, who held the field to account first.

The stats we use, in plain English

Five terms show up throughout the report. Here's what each one means before you hit the charts.

Draft utility: whose ranking recovered real value best?

Spearman rank correlation between each list's projected order and the real finishing order · 1.0 = a perfect draft board

This is the headline. How closely each list's ranking matched the order players actually finished in, scored as a Spearman rank correlation where 1.0 is a perfect board and higher is better (0.80 means it recovered most of the real order). Bars show the 90% margin of error, so overlapping bars mean a statistical tie. The table adds precision@25/50/100 and nDCG; all four terms are defined in the glossary just above.

The Wisdom of the Crowd, the simple average of every projection, tops the board. That's the classic result: no individual reliably beats the consensus. The whole field is tight up here, with every margin-of-error bar overlapping, so the leaders are statistically inseparable and each is a sharp, well-built set. Dobber and Laidlaw each skip one category, so they race on the 6-category boards below, and hold their own there.

Position-adjusted: value over replacement (VORP)

Spearman rank correlation, scored on value over the last starter instead of over zero · 1.0 = a perfect draft board

Plain value is position-blind, and drafting is not. A scarce elite defenseman is worth more than his raw totals, because the drop-off to a replacement D is steep. This re-scores everyone as value over the last starter in a standard Yahoo 12-team league (6 F / 4 D per team, so replacement is the 72nd forward and 48th defenseman), which lifts scarce positions up the board. If the leaderboard survives this, the original result was measuring real value and not a scoring illusion. It survives.

The top order holds, so the original result was not an artifact of ignoring position. The reshuffle in the middle is the story here. A source that rises valued scarce defensemen closer to the truth; one that fell leaned too hard on forward scoring. (Replacement is set at the last starter; a deeper waiver baseline that counts bench spots is a one-parameter change.)

Same race without PPP, so Dobber can play

Spearman rank correlation on 6 categories, power-play points excluded

Dobber doesn't project power-play points, so it can't get a 7-category value. Here every source is scored on the same 6 categories (PPP excluded), the fair way to still rank Dobber head-to-head. Pro-rated properly, it lands right in the mix. 274 shared players.

Same race without PIM, so Laidlaw can play

Spearman rank correlation on 6 categories, penalty minutes excluded

Laidlaw's league doesn't score penalty minutes, so he never projects them, which keeps him off the 7-category board. Here everyone is scored on the same 6 categories (PIM excluded) so Laidlaw races head-to-head. 272 shared players.

Why the order? Consistency over peaks

Each projector's average category rank (0 = best in the field, 1 = worst) and how much it swings between categories. The leaders win by having no weak category. They rarely top any single stat, and they never finish last in one. Bangers Fantasy Hockey and Lineup Experts have nearly identical averages, but Lineup Experts is steadier, while Bangers is elite on scoring and softer on peripherals.

Blind spots: the players the field ranked way too cheap

The season's biggest under-rankings, measured in the unit you actually pay in: draft slots. These are actual top-150 finishers the field's consensus buried, sorted by how many times earlier they should have gone. Darren Raddysh leads it: the field had him around 555th and he finished 14th, a 39.6× steal. Each dot is one projector's draft slot for that player, and the gold line is where he actually landed. Dots far right of the line mean the whole field slept on a breakout. Green = Bangers.

These are players everyone had on their board but priced far too low, the juicy under-valued breakouts rather than off-radar depth. You can see exactly which projector came closest to the gold line.

The other half: the players the field ranked way too expensive

The mirror of the chart above, measured in the unit you actually pay in: draft slots. These are players the consensus ranked inside its own top 50 who finished furthest below it. Each bar runs from where the field's numbers had the player to where he actually finished, so the length of the bar is the size of the mistake. Victor Hedman leads it, drafted around 47th and finishing 240th. Falling out of the top 50 or the top 100 is weighted extra, because that's where a player stops being startable. Green = where Bangers had him.

Each cause is read off the player's own underlying numbers, checked in this order: goals finishing well under the chances he generated is "chances held, finishing didn't"; a drop of 10% or more in expected goals per 60 is "chance quality fell away"; a drop of 10% or more in shots per 60 is "stopped shooting as much"; then a loss of ice time, then a shooting percentage below his own career mark. Age and games played are only a fallback for players whose process data says nothing either way. This is the expensive error: you draft an over-projected player in every single league, immediately.

Bangers Fantasy Hockey: where we broke from the field

Players where Bangers landed near reality while the field was wrong (left), and the reverse, where the field was right and Bangers was off (right). Chips show the driving category as Bangers / field / actual (per 82), and the reasoning comes from whatever moved the call.

Bangers Fantasy Hockey: category accuracy

Placing within each category by weighted absolute percentage error, with that error shown beneath · lower is better

Where Bangers finished on per-game accuracy in each category (rank of 12). The scoring categories are the strength: best of the individual projectors at Points, and top 3 in Goals, Assists and PIM.

The actual misses: percentage error by category

Weighted absolute percentage error · total miss divided by total production · lower is better

The raw numbers the placings are built from. Each cell is a list's percentage error in that category, its total misses as a share of what players actually produced, per game. Lower is better. We aggregate the error across players rather than averaging each player's own percentage, so a zero or near-zero actual (common in power-play points and penalty minutes) can't blow the number up. This is the layer that lines up with an Apples & Ginos style accuracy report. Colored down each column, so darker = lower error within that category. The Bangers row is highlighted.

Shading is per column, so it ranks the field within each category. Empty cells are categories a source doesn't project (Dobber skips power-play points, Laidlaw penalty minutes). Compare down a column, since percentage error runs naturally higher in low-count categories like power-play points, where a small raw miss is a big percentage.

Category by category, the whole field

Placing within each category by weighted absolute percentage error · 1 = smallest error · a dot marks a placing statistically tied with the category leader

The same eight categories as placings now, 1 to 12, with 1 the smallest percentage error and darker = better. A ● dot marks a placing statistically tied with the category leader, meaning that race is inside the noise. Hover any cell for the percentage error, bias and skill. The Bangers row is highlighted.

Accuracy spread, category by category

Weighted absolute percentage error across the field in each category

Each dot is one projector's average miss for that category, where left is more accurate. Tight clusters mean the field agrees and nobody has an edge; wide spreads mean real separation. Green = Bangers Fantasy Hockey. Hover any dot.

Does projecting even beat "last year's rate"?

Bangers' accuracy gain over simply carrying forward each player's prior-season rate. Positive means the projection added real value. Notice HITS ≈ 0 for everyone in the field. Nobody beats last year's number on hits.

Does the result survive a hostile reader?

The Bangers Fantasy Hockey rank across value systems (neutral z-score vs the league's own formula) and games-played floors. A finish that only held under one setting would be cherry-picked; a stable top-2 is the real claim.

The players the field got most wrong, by category

For each category, the draft-relevant skaters (top-200 value or top-100 points) where the field consensus missed by the most. Shown as consensus → actual per 82; ▲ green = the field was too low (a breakout), ▼ red = too high (a bust).

Every projector on the biggest points surprises

Each of the season's biggest point swings vs consensus: one dot per projector's projected points (dodged so all are visible), and the gold line is the actual total. Dots piled far from the line = the whole field missed together. Green = Bangers.

The 1st-overall stress test: did the top of the board hold?

Overall accuracy hides how fragile a board is. A list built around a consensus star who busts can turn the best draft pick into a liability. So we rank each projector's players by its own board and check where those top picks actually finished, both position-adjusted (VORP), so a scarce elite defenseman gets credited at its real draft slot instead of being buried by raw scoring. A banger-league wrinkle: the consensus #1 was Brady Tkachuk, a hits-and-PIM monster rather than a pure scorer, who finished #4. The actual VORP leader was Connor McDavid. Colors: top-15, 16 to 40, cratered. Almost every board's biggest top-10 bust was the same player, Auston Matthews (finished #44), the season's universal lock that cratered, so the "biggest crack" column shows each list's next-worst hold instead. To be clear about what “their #1” means: it is the highest-VORP player implied by that list's projected numbers, run through the same neutral formula as everyone else. It is not the player the source published at first overall, and several would tell you they'd have drafted somebody else.

2025-26 had an unusually predictable elite tier. Every board's top 10 mostly held, so no one suffered a McDavid-in-2023-24 style collapse. The separation is in the averages and the biggest cracks.

Where Bangers over-reached: a self-autopsy

Bangers Fantasy Hockey's biggest points over-projections on draft-relevant players, each tagged with the cause its own underlying numbers point to: expected goals per 60 for chance quality, shots per 60 for volume, ice time for role, and goals against expected goals for finishing. This is the kind of accountability the peer field mostly doesn't publish.

These split into 2 different mistakes, and only 1 of them is our fault. Matthews, Bratt and Weegar genuinely generated worse chances than the year before, which our engine should have caught and did not. Miller and Pettersson kept creating at the same rate and stopped converting, which is finishing luck, and re-rating them down now would be chasing noise. Telling those 2 apart is the cheapest honest gain available to us next season.

The optimism check: who runs hot on their own board

Over-projection is the expensive error. You draft the bust immediately, in every draft. Under-projection only costs you a late gem in deep leagues. So we split each projector's error on its own top-150 board (the players it would actually draft) into over versus under, per 82. % over is how often a projector's own pick came in below what it projected, meaning it ran too hot. Field average: 43%.

The aggressive sites (Dobber, Daily Faceoff around 51%) over-project their own boards most, Bangers sits on the conservative side, and Kubota is the most conservative of all. That's the mild edge the asymmetry rewards: fewer over-reaches. (On raw totals the whole industry looks far more optimistic, but that's mostly unpredicted injuries pulling actuals down. We grade rate, so this is the pure rate-only lean.)

Does avoiding busts actually win? Testing the hypothesis

The pool-aid creators suspected the strongest lists won by drafting fewer over-projected players rather than by making more correct calls, but flagged it as unproven. We can test it directly. Across the 9 full-category projectors, does over-projection error predict draft utility more than total error does? (More negative means that kind of error hurts drafting more, and the hypothesis predicts over-projection should be the strongest negative.)

In 2025-26 the data leaned the other way. Over-projecting players was not what hurt draft utility most. Its correlation with draft utility came out positive, +0.16, while under-projection ran to -0.27 and total error to -0.36, the strongest signal of the three. So overall accuracy was the best predictor of a good draft, and under-projecting players hurt at least as much as over-projecting them. The season's strongest lists got there by being more accurate across the board. Playing it safe, on its own, bought nothing. There are two honest reasons it came out this way. First, our per-82 framework strips out the injury-driven busts that made over-projection so costly in 2023-24. Second, this season's elite tier was predictable (the stress test above), so over-reaching on a consensus star rarely got punished. With n=9 this is directional rather than definitive, and it's still a real, testable answer to a question the original analysts could only guess at.

Missed opportunity: points left on the table at each draft slot

Actual scoring points (goals plus assists, per 82 games) left unclaimed per draft slot, averaged within each stretch of the board · lower is better · sorted by the all-200 average

A rank-accuracy metric (from the pool-aid missed-op video). At each draft slot, how many points did a list's pick leave on the table versus the best player still available on its own board? Points here means actual scoring points, goals plus assists, on an 82-game pace, not the z-score value the main leaderboard uses. A lower number means a sharper board. Each list is graded against the perfect ordering of its own players (so a deeper list isn't penalized), banded by depth. Darker means fewer points lost. It rewards getting the raw scoring order right rather than the all-category value order.

These are disjoint stretches, not running totals: 11-50 covers only picks 11 through 50, so a strong top 10 no longer flatters every deeper column. Bangers Fantasy Hockey orders the top of its board best, around 8.3 actual points left per top-10 slot, and is best again from 51-100 and 101-200, but its weakest stretch by far is picks 11-50. The Athletic is the mirror image: soft through the first 10, then the best list in the field from 11-50.

The bottom line

What to actually take from all this.

The consensus of the field graded out best this season, and a small group of individual lists, Bangers among them, sat a hair behind it in a statistical tie. Hold that lightly, though, because these results move year to year. A list that hugs a sharp consensus wins in a season like this one, and it can just as easily trail in a season where the crowd runs soft.

The bigger truth is that no projection, ours or anyone else's, wins you a championship on its own. A draft board gets you to the starting line, and the season is run from there. The managers who win are the active ones. They study the schedule, they work the waiver wire, and they keep feeding on great in-season content all year, Dobber's Ramblings and the Apples & Ginos podcast chief among it. Draft off the numbers, then go outmanage the person next to you.