Action Network's NFL Luck Rankings are back in 2026 for its fifth season.
In the four-year history of the Luck Rankings, games that met luck-based thresholds for either sides or totals had a positive ROI in three of the four years when looking at side and totals combined.
Unfortunately, that one miss was last year, with sides and totals going a combined 50-48, with sides going 26-24 and totals 24-24.
The good news in that: with an extra year of data we can add a few small refinements to improve the process, and that's exactly the direction we're taking this year.
Instead of a major overhaul, it's minor tweaks to both the model itself and to the thresholds.
So since last year was Version 3.0 of the Luck Rankings, we'll call this Version 3.1 — a slight refinement to Version 3.0 instead of a larger update.
NFL Luck Rankings — Methodology Update
Expected Scores
The driving force behind Action Network's NFL Luck Rankings is something we call Expected Score.
For 2026, nothing about this has changed other than slight adjustments to the parameters to account for a new year's worth of data. It's the same model, just fit to more data.
Loosely, the Expected Score is the final score we'd expect two teams to achieve if they replayed the same game situations an infinite number of times against each other on that day.
These Expected Scores adjust to offensive and defensive strengths, account for injuries, and look at every single play in the play-by-play data, including penalties, special teams, and more.
Normalizing Luck
This is where the biggest tweak occurred, and it's also the most mathematically complicated.
Essentially, for any given week, the gap between two opposing teams' Luck% is approximately normally distributed and centered around zero. As more games accumulate, the standard deviation of that distribution tightens, which is how we can normalize it to a single Luck Score across weeks.
The tweak here was twofold.
First, we moved from week-based to game-based to account for byes.
Second, we slightly changed the game-based fit to smooth out some rough edges the old week-based normalization used.
In the end, all it does is slightly reorder the Luck Scores. Some B-graded games might move to a C, or some C-graded games might move to a B or a D, and so on.
But the idea is to get an even better ordering so we're more confident in the best games to use as our criteria, which is what we'll discuss next.
Luck Rankings Criteria
Luck Sides
Luck Sides last year went 26-24, and most of the struggle game from the bottom half of "bettable" games under the new Luck Score.
Last year, we had five tiers for Luck Sides: A, B, C, D, and F. An F-rated game was a no-play, but we considered everything D and up to be playable under the Version 3.0 Luck Scores, with the best bets coming at the top tiers.
But, if we use this new reordering and look at just the new top 23 sides, they would have gone 15-8 (65.2%) against the closing spread.
Scrutinizing further, going back to 2018, where our Luck Rankings data starts, games meeting this new Luck Score threshold of 1.9 or above have gone 82-43 (65.6%) against the closing spread for +34.7 units.
And for the numerically inclined, this isn't just a backtest.
There's multiple validation methods that show these games should hit at a high rate, and intuitively it makes sense, the Luck Rankings have largely been successful over the four years we've used them, and each evolution is mostly just a refinement of specific games and thresholds.
There's several somewhat similar ways to validate Version 3.1, so I'll just briefly cover them.
- Leave One Season Out: Build this Version 3.1 model, but only use seven of the eight seasons and then test it on the eighth season. Do this for each of the eight seasons, refitting so the model never sees the season it's grading against. This produces a record of 75-46 (62.0%) for +24.4 units.
- Era Split: Fit once on 2018-2020 data, test against 2021-2025. Since we're only testing on five seasons now, the sample size is smaller but it produces a 42-24 record (63.6%) for +15.6 units.
- Walk Forward: Update the model each year and test on the next season of data. So 2021 would use 2018-2020 data then 2022 would use 2018-2021 and so on. This produces a record of 50-29 (63.3%) for +18.1 units.
In other words, all three out-of-sample methods produce a hit rate of 62% or higher on data that wasn't seen by the model. And it makes sense, that's about the success rate Luck Sides have historically had in our past versions.
So, the new rule for a game to be a Luck Side is as follows: A Luck Score of 1.9 or higher.
It's a nice, clean simple rule, and as a result we drop the 1.454 Luck Score threshold for a D grade, and we drop the whole letter system in general, which was a bit too granular.
This means we're a bit more strict on what it takes to be a Luck Side, so there will be fewer Luck Sides over the course of the season, but they'll hopefully be positioned better for success.
Luck Totals
Like Luck Sides, we've simplified Luck Totals as well.
We're using this same model and just simplifying the criteria as well. We're dropping the "strong" and "weak" definitions of Luck Totals and just using a play/no-play system like we're moving to with sides.
And just like Luck Sides, the same out-of-sample testing was done to validate the new threshold, and the leave-one-season-out method actually shows an improvement over a pure backtest, potentially giving hope that the numbers I'll discuss below have been a little … unlucky.
The new rule for a game to be a Luck Total is as follows: A Luck Total above +1.5 is a Luck Over. A Luck Total below -1.5 is a Luck Under.
That's as simple as can be, and applying it to the past eight seasons Luck Totals would be profitable in six of the eight years with an overall record of 155-117 (57%) against closing totals.
That breaks down into 78-54 (59.1%) for overs and 77-63 (55%) for unders, which matches everything we've seen with past versions. Luck Overs tend to be the stronger bet of the two Luck Total options.
Week 18
One thing that fell out of these tweaks and stronger thresholds is that the final week of the regular season was actually positive on both Luck Sides and Luck Totals.
Luck Sides have gone 6-2 in the last week of the regular season, while Luck Unders have gone 7-4 to the under and Luck Overs have gone 9-4 to the over.
In the past, the looser thresholds and slightly worse ordering made the last week of the regular season unplayable from a luck perspective. But a total record of 22-10 across the Luck Sides and Luck Totals suggests the last week of the regular season is back in play.
Luck Rankings 2026
We hope these updates improve, simplify, and enhance the experience for you all going forward.
The Luck Rankings and its associated games will once again be available to Action PRO subscribers with weekly Luck Rankings starting in Week 2, as well as a midweek PRO piece by our Predictive Analyst, Dr. Nick Giffen, highlighting each week's luck-based sides and totals.
Good … luck … in 2026!
