Some predictions. Several assumptions. A season available to invalidate both.

The NHL season starts on September 29, and Fantasy Hockey Intel has produced its first locked preseason forecast.

This is not a ranking and it is not an attempt to predict the final scoring table with unnecessary precision. The purpose is narrower: to identify players where the available data, role, trajectory and team context suggest that 2026/27 may develop differently from what the previous season alone would imply.

Some of these predictions will be correct. Some will not. That is partly the point.

A projection is not an event

A projection describes what appears reasonably likely under a set of assumptions. An NHL season has a habit of interfering with those assumptions.

Players get injured. Coaches change lines. Power-play units that looked settled in September can be reorganized in October. Teammates are traded, other players improve unexpectedly and roles change for reasons that were impossible to include in a preseason model.

There are also things that have very little to do with hockey analysis. A player can become a parent, get sick, miss time for personal reasons or play through an injury nobody outside the organization knows about. Someone can also simply shoot 6% for two months while doing most other things correctly.

None of this makes projections useless. It means they should be treated for what they are: estimates based on the information available at a particular point in time.

Fantasy Hockey Intel therefore treats this forecast as a baseline to test, not a result to defend.

The model starts with current production and Expected P/G, then adds information about offensive process, deployment, sustainability, trajectory, role stability, organizational commitment, dependency and team context. Expected P/G is not the final forecast. It is one input among several.

That distinction is important.

Preseason forecast

These are rate forecasts. Point totals should be derived from actual availability rather than mechanically multiplying every number by 84 games.

The strongest positive cases

Andrei Svechnikov

Forecast: 0.95–1.05 P/G

Svechnikov is probably the cleanest process-driven upside call in the first version of the model.

His actual production last season was 0.89 P/G and his Expected P/G was also 0.89. On the surface, that is not especially interesting. The more useful information sits underneath those numbers.

His Offensive Driver score of 19.59 exceeded his Deployment score of 17.90. In simple terms, his individual offensive process looked slightly stronger than the opportunity he was receiving. That is generally preferable to the reverse situation, where a player is producing mainly because premium deployment is doing much of the work.

Svechnikov’s trajectory is Improving and his process signal is Elite Stable. The forecast is therefore not primarily a regression call. It assumes that an already strong individual process can support somewhat more production if deployment and availability cooperate.

The last part matters. A 1.00 P/G player who appears in 62 games is a very different fantasy asset from a 0.95 P/G player who plays 80. The model is more confident about Svechnikov’s scoring rate than about his final point total.

Nico Hischier

Forecast: 0.90–1.00 P/G

Hischier is a more conventional regression case. He produced 0.80 P/G last season against an Expected P/G of 0.99, which is a substantial gap.

Unlike some breakout candidates, he does not need a new role for the prediction to work. His existing deployment is already strong enough. The basic expectation is simply that results move closer to the underlying performance.

That makes Hischier a useful test of Expected P/G. If he receives broadly similar deployment, generates similar underlying offense and again finishes around 0.80 P/G, the model will have learned something. If production moves back toward 0.95–1.00, the simpler interpretation was probably correct: he deserved more points than he actually received.

Not every forecast requires a complicated story.

Adam Fantilli

Forecast: 0.80–0.90 P/G

Fantilli is the most traditional breakout prediction in the group.

He finished last season at 0.72 P/G with an Expected P/G of 0.80. His role expanded, Columbus’ offensive environment has improved and he remains at an age where normal player development should still matter.

None of those factors is especially remarkable on its own. Together they create a stronger case. There is some positive regression available from the gap between actual and expected production, there is more opportunity, and there is a reasonable expectation that the player himself is still improving.

The current forecast places him around 0.80–0.90 P/G, roughly upper-60s to mid-70s production over a mostly healthy season. That does not require Fantilli to become a different type of player. It requires several things that were already moving in the right direction to keep moving.

Connor Bedard and the problem with season totals

Forecast: 1.10–1.20 P/G after returning

Bedard illustrates why scoring rate and total fantasy value need to be separated.

His underlying forecast is relatively straightforward. He produced 1.09 P/G last season against an Expected P/G of 1.08, while carrying an Elite Driver and a Rebound trajectory. There is little in those numbers suggesting that his scoring rate was artificially high.

The complication is external to that process: he is injured.

Shoulder surgery cannot be averaged away by adding more decimal places to a projection. Fantasy Hockey Intel therefore expects an elite scoring rate after his return while also expecting his full-season totals to be reduced by missed games.

Both things can be true at the same time. It is also why mechanically multiplying a projected rate by 84 games often creates a number that looks more precise than the underlying information deserves.

Leo Carlsson: being right does not necessarily create value

Forecast: 0.95–1.05 P/G

Carlsson is one of the strongest profiles in the model. He produced 0.96 P/G last season against an Expected P/G of 0.98, with a Strong Up trajectory and a Breakout Confirmed signal.

There is therefore little reason in the current model to treat last season as a fluke.

That does not automatically make him an attractive acquisition. If the fantasy market already values Carlsson as approximately a point-per-game player, correctly predicting that he will remain approximately a point-per-game player creates limited excess value.

Fantasy analysis often mixes up two different statements: this player will probably be very good and this player is undervalued.

They are not the same thing. Carlsson can have an excellent season and still be priced fairly.

Beckett Sennecke: year-two progression

Forecast: 0.75–0.85 P/G

Sennecke is no longer a prospect projection. He played all 82 games in 2025/26 and produced 60 points, or 0.73 P/G. That gives us a complete NHL season to work from.

The question for 2026/27 is therefore not whether his junior production will translate. It already has. The more relevant question is how much progression should reasonably be expected in year two.

The starting point is strong. A 20-year-old producing 60 points over a full NHL season has already established a meaningful offensive baseline. The uncertainty now concerns role growth, power-play development, shot generation and whether his individual process improves as he gains experience.

A fairly normal second-year progression would move him from roughly 0.73 P/G toward the 0.75–0.85 range. A larger breakout is possible, particularly if his usage, individual process and power-play opportunity all increase together, but the current model does not need to assume that outcome.

Sennecke is therefore useful as a year-two progression test rather than a prospect watch.

Two regression calls

Clayton Keller

Forecast: 0.90–1.00 P/G

Keller produced 1.07 P/G last season against an Expected P/G of 0.94. His Deployment score exceeded his Driver score, while his trajectory was Declining despite a Stable process classification.

None of this means Keller is suddenly a poor fantasy player. It means the evidence for another clean jump, particularly into sustained 100-point territory, is weaker than the raw point total might suggest.

Utah can improve as a team. Keller can retain premium usage and benefit from better teammates. He can also regress individually from 1.07 P/G. Those outcomes are not contradictory.

This is one of the more useful disagreements to follow because external expectations are more aggressive. If Keller pushes toward 100 points while his Driver and Expected P/G rise with him, the current FHI forecast will simply have been too conservative.

That would also be useful information.

Nick Suzuki

Forecast: 0.98–1.10 P/G

Suzuki is a different kind of regression call.

His 1.23 P/G season is unlikely to be the most useful single baseline for 2026/27. Expected P/G was lower at 1.12, and his production remains relatively dependent on premium deployment. At the same time, the underlying evidence does not suggest that the breakout itself was fictitious.

The result is a fairly unexciting conclusion: Suzuki can regress and still be excellent.

A decline from 1.23 to around 1.05 P/G would be statistically meaningful without constituting anything close to a collapse. Regression is often discussed as if something has gone wrong. Usually it just means an unusually high or low result has moved closer to the underlying level.

Suzuki is a good example of that distinction.

Brandt Clarke: opportunity versus process

Forecast: 0.45–0.55 P/G initially

Clarke may be the most informative player in the entire preseason exercise.

The opportunity case is straightforward. Los Angeles needs more offense from the defense, the power play needs improvement and Clarke has a realistic path to more valuable deployment.

The underlying evidence is less enthusiastic. He produced 0.49 P/G last season against an Expected P/G of 0.46, while carrying a Declining trajectory and a Regression signal.

That gives us two competing explanations.

One is that opportunity is the missing ingredient. Give Clarke PP1 and greater offensive responsibility, and production rises with it.

The other is that deployment alone is not enough and that the individual process also needs to improve before a major fantasy breakout becomes sustainable.

Fantasy Hockey Intel currently leans toward the second interpretation. That does not mean the forecast should be stubborn. If PP1 share rises together with Driver and Expected P/G, the evidence has changed and the forecast should change with it.

Clarke is a useful reminder that deployment creates possibilities. It does not guarantee production.

Pavel Dorofeyev and environmental dependency

Forecast: 0.75–0.85 P/G

Dorofeyev changes teams after producing 0.78 P/G against an Expected P/G of 0.81. The starting assumption is therefore not that last season was artificial. His production was reasonably well supported.

The question is whether his new environment can recreate the conditions behind that production.

Dorofeyev’s profile was both Shot Driven and Role Driven. The first characteristic should travel reasonably well. The second may not.

That means his first few weeks should be evaluated through usage as much as through points. Shot volume, shooting locations, power-play time and linemate quality will tell us more about whether the role has survived the move.

Ten games with seven points but declining shot generation could be less encouraging than ten games with five points and an expanding offensive role.

This is one reason early-season fantasy analysis should involve more than sorting the scoring table.

Will Smith: opportunity ahead of process

Forecast: 0.85–0.95 P/G

Smith has one of the better environments available to an emerging NHL forward. Playing around Macklin Celebrini is useful, and San Jose can provide premium offensive deployment.

The remaining question is how much of the next step Smith creates himself.

Fantasy Hockey Intel is not rejecting a breakout. The model is attaching a condition to it. For a genuine point-per-game season to become the central expectation, the individual process should begin catching up with the deployment.

That means more evidence from shot generation, individual expected goals and Driver rather than simply more minutes beside elite teammates.

A player can benefit from a strong environment for a very long time, and those fantasy points still count. Environmental dependency matters mainly when assessing how stable and transferable that production is likely to be.

Prospects and transition cases

This section is deliberately narrower. It covers players for whom Fantasy Hockey Intel does not yet have a mature NHL baseline.

That is a data distinction, not an age distinction.

Anton Frondell — Chicago

Frondell is closer to an NHL transition case than a conventional prospect projection. Immediate opportunity may exist, particularly while Chicago deals with injuries, but the NHL sample remains limited.

His projected role is useful information, but it is not yet a reliable production baseline.

Status: Early NHL opportunity watch.

Ivar Stenberg — San Jose

Stenberg is the clearer prospect case. There is no mature NHL baseline, so the model has to rely more heavily on development history, prospect production, organizational context and projected opportunity.

San Jose is an attractive environment, but an attractive environment is not the same thing as an established NHL role.

Status: Prospect / dynasty watch.

The relevant trigger is when projected opportunity becomes actual NHL usage.

What we actually want to learn

The obvious way to judge these forecasts would be to return in April and count how many players finished inside their projected range. We will do that, but it is not the most interesting part.

The more useful question is why a forecast was right or wrong.

With Svechnikov, we want to know whether Driver identified additional scoring potential before the box score did. With Hischier, the question is whether the gap between actual and Expected P/G closes. Fantilli tests whether trajectory, role growth and positive regression can combine into a genuine breakout.

Sennecke gives us a different test: whether a strong full-season rookie baseline leads to relatively normal year-two progression or something larger.

Keller tests whether the model is right to resist a popular external breakout narrative. Clarke tests whether opportunity alone can produce the next step or whether individual process must improve with it. Dorofeyev tests how much of a role-driven scoring profile survives a change of environment.

These questions tell us considerably more about the model than successfully predicting that Nathan MacKinnon will score many points.

The latter is probably true. It is also not especially informative.

Probabilities should not be taken too literally

The forecast includes probabilities, but they currently represent confidence in the direction of the forecast, not statistically calibrated probabilities.

When Fantasy Hockey Intel assigns roughly 75% to Svechnikov or Hischier and 70% to Fantilli, we are not yet claiming that events labeled 70% have historically occurred seven times out of ten. There is not enough forecast history to support that claim.

Calibration has to be earned.

The 2026/27 season is part of that process. Over time, we should be able to test whether High Confidence forecasts actually outperform Medium Confidence ones, which signals create the most false positives, whether Deployment tends to move before production and whether Trajectory provides useful information beyond Expected P/G.

At present those are research questions.

Calling them anything else would make the model sound more certain without actually making it better.

The forecast will change

The preseason snapshot is locked, but that does not mean every forecast remains fixed until April.

New evidence should change a forecast. A sustained PP-role change matters. So does a meaningful line change, injury, trade, coaching change or sustained movement in Driver, Expected P/G or trajectory.

When that happens, Fantasy Hockey Intel can publish a revised forecast while keeping the original prediction intact.

That distinction is important. Otherwise it becomes very easy to gradually edit a prediction until it resembles the eventual outcome and then call the exercise successful.

The sequence should remain visible:

Preseason forecast → new evidence → revised forecast → outcome.

Final preseason view

The strongest positive cases entering the season are Andrei Svechnikov, Nico Hischier and Adam Fantilli. Connor Bedard remains an elite rate forecast complicated by availability, while Leo Carlsson’s breakout appears real even if much of that information is already reflected in his market value.

Beckett Sennecke is now a year-two progression case rather than a prospect projection. Sixty points over a full rookie season gives the model a useful NHL baseline; the remaining question is how much his role and individual process move from there.

Clayton Keller is the clearest case where Fantasy Hockey Intel currently disagrees with a more aggressive breakout narrative. Nick Suzuki is expected to regress without ceasing to be a high-end fantasy player. Brandt Clarke may tell us more about the model than almost anyone else because his opportunity and underlying process currently point in different directions.

There will still be injuries, line changes and outcomes nobody modeled correctly. Someone who currently appears irrelevant will probably score 25 goals. Someone projected for 75 points may shoot 6.8%. A coach will make a decision that appears statistically unhelpful. A player will miss games because something happened in his actual life rather than inside a hockey model.

That is normal.

Projections do not predict events. They describe what the available information suggests before those events occur.

The useful question at the end of the season is therefore not simply whether the numbers were correct. It is whether Fantasy Hockey Intel identified meaningful changes early enough for that information to have been useful before everyone could see them in the standings.

That is what the 2026/27 forecast is intended to test.