Most bettors place NFL futures before Week 1 and move on. Learn how weekly power ratings and season simulations can uncover futures value all season long.
Back to VideosEverybody bets NFL futures in August, but what if they're doing it completely backwards? What if the best time to bet futures is not when we know the least about a team, but after the season starts when new information is coming in every week? Most people fire a Super Bowl bet before week one. Maybe they'll sprinkle in a regular season win total, and then they move on. But that means missing months of value. Well, today we're going to capture that value. I'm vibe coding again. This time I'm going to build a tool to attack futures markets all season long. We'll start with power ratings. We'll figure out the best way to blend multiple sources together.
We'll test whether the blend has signal against the betting market, and then use those ratings to price futures through a season simulator. The big question will be whether this is something that has real heft or is just another spreadsheet toy. Can we build a ratings blender that turns weekly ratings into futures value? You see, when you have good power ratings, the ratings themselves aren't the destination.
They're a compass that points you in the direction of your destination, which is finding value in futures bets. You then utilize a simulator to take you there and reap all the benefits. Futures prices are born from a few different factors. It comes down to team strength, schedule, team variability, and the market itself. If your estimate of team strength is accurate, a simulator like the one we have at Unabated can turn those ratings into projected spreads, season outcomes, and fair futures prices. If your ratings are bad, the simulator is just going to be confidently wrong for you. That's why power ratings matter.
A good set of power ratings sees beyond results and keys in on what really dictates a team's relative strength. The problem is that every rating system has blind spots. One might overreact to a blowout win, another might be slow to recognize an injury. They might not be player-based at all. Another might simply have a flawed process. As statistician George Box famously put it, "All models are wrong. Some are useful." The idea behind a ratings blender is simple. Instead of trusting one opinion, we combine several sharp opinions and let them check each other's weaknesses. If that blend ends up closer to reality than any individual system, we've got something useful.
But that's also where things get dangerous. Because if we're allowed to blend anything together, we're also allowed to overfit anything together. The true test that matters is whether the blend can hold up against the market later. Let's start though with setting up the parameters of the app.
I asked Codex to build a tool that would combine multiple rating systems, optimize their weights, and update those blends every week as the season unfolded. Honestly, the first draft was better than I expected. Within a few prompts, we had something that could blend ratings on the fly, test different weight combinations, and track performance week-to-week. As a proof of concept, it worked surprisingly well. As a betting tool, well, not so fast. The problem wasn't getting the app to run. The problem was making sure it wasn't just fitting yesterday's data and pretending it could predict tomorrow. And that's where the whole project gets interesting. Yeah.
If all you're after is a simple ratings blender, well you can do that in a spreadsheet, no problem. Line up a column for each system, rows for every team, add in some weights for each one, and it's one simple math problem to blend them all together. Why go through all the trouble to build something more involved?
Because this opens up data automation that can save you tons of time. And the more automatic a tool is, the more likely you are to actually use it than doing all the manual data yourself every week. Plus, automated processes free up time that you can spend on other pursuits. You also establish a systematic approach, which will be much easier to quantify going forward. In our case, that means including a data science layer to find the optimal blends, so we have the best possible information available for simulation. And if we're going to the trouble of building that out, we can also consider using our power ratings for week-to-week single-game betting, too.
For this project, we're using a combination of the Massey-Peabody ratings that we have here at Unabated, ESPN's FPI rating, NFL ELO, and former Pro Football Focus analyst Kevin Cole's unexpected points, all from the 2025 season. Now, Codex's first stab at the optimizer was a bit janky. It was only looking at the previous week's performance to craft an optimal blend for the week moving forward. That's like a better basing all their handicapping on a one-game sample.
But after some back and forth with Codex, we settled into a more refined development where the systems reached relative stability around week eight and finally settled in at a consistent blend around week 12. Does that mean we found usable signal?
Well, to figure that out, I did a walk forward test where we'd use our blended projections to paper bet the upcoming week. In a walk forward, the model can only see whatever data it would have had to that point to make a decision treating each week as its own unknown. You can't bet week seven with week nine information and we didn't want the model to, either. But once we did that, we asked it to figure out what we would have bet according to the blend. For example, in week nine, we saw in the Browns Jets game that the line was Cleveland minus two, but our blend said the Jets should be favored by 1.7.
Ratings say the value was on New York, so that would have been a bet. If we can find a blend that gets us closer to reality, we'll have something that's both great for betting futures markets and for making week-to-week plays. From weeks two through 16, betting only when we had an edge of 1 and 1/2% or better, we basically would have broken even on the season. You know who you are? Even Steven.
Now, what about if we took even the thinnest edges? Well, that would have left us up eight units on the year. Not exactly crushing, but beating the market is beating the market. Except if we look a little closer, week 12 is a standout. That was an eight-unit week all on its own. The rest of the season was break even. So, are we really dealing with something that's ready for primetime or did we just happen to get a little lucky in a week? And that's when I discovered that the data we pulled for this was using the closing on Unabated line from the 2025 season.
Well, now you know why any better with a pulse and a few without one would tell you not to waste your time betting into the NFL at the closing line. No one's beating the closing number in the most efficient market in US sports. I was actually happy to see that there wasn't much of an edge here. In fact, this is exactly what we would prefer to see because it makes us ask, "Well, what else is there to try?" So, now I did the same thing using the Unabated line on Sunday nights when openers are released. This particular mix of systems improved against openers. It went 129-108-3 on the season for a 22 unit gain and a impressive 9.18% ROI.
Now, if the ratings blend can beat openers, then it has signal. But, signal by itself isn't quite enough. A tool like this only matters if it fits into a real betting workflow. Otherwise, it becomes what every better already has buried somewhere on their hard drive. A spreadsheet they were definitely going to update every week until life, work, and the Thursday injury report beat them into submission. So, the next step was turning this from a one-off test into something repeatable.
That meant pulling in game data automatically. Schedules, results, standings, weekly lines, so that we're not rebuilding the wheel every time the Titans remember that they're the Titans. It also meant creating a library for rating systems, so each upload gets tagged by source and week. That way, we can track how each system changes over time instead of just staring at one frozen snapshot. So, from there we added settings the simulator actually needs. Home field advantage, quarterback adjustments, global defaults, and team specific overrides. And that matters because the difference between the Chiefs are two points better than the market thinks and Patrick Mahomes is out is not exactly a rounding error. And finally, the big one.
We connected the Unabated API to the tool so we could pull in market prices automatically. And that's where this becomes something more useful than a sandbox. Now, we're not just asking, what do the ratings say? We're asking the question that actually matters. What do the ratings say compared to the market right this second? And now we've got a tool with real heft. This isn't just something we play around with in a sandbox where an entire season's results are already known. This is something we can use going forward for the entire football season. If you're doing this work over the summer, the big drawback you'll hit is that the public rating systems haven't started to drop.
There is an early version of FPI that you can use, but um not much else. To start pulling together a futures picture, I'm going to use week 16 ratings from Massey Peabody and unexpected points before the chaos of week 17 and 18 factored into team ratings. I'm also going to give a best approximation of their week one ratings by adjusting for quarterback changes. Now, I can load these into the blender and export ratings to use in the Unabated simulator. And here's what we see. According to the simulator, there's value on the Eagles to win the NFC East, the Chiefs to win the AFC, and the Seahawks to repeat as Super Bowl champions.
But the bigger story is what happens as the season goes on. Now, we've got a data-driven, repeatable process that lets us keep shopping all season long. And with the automation layers we've put in, it makes it easy enough that you don't get halfway through the season, realize you haven't updated ratings in a couple of weeks, and give up rather than having to backfill all that data. And here's why that's so valuable. Let's just look at last year's Super Bowl champion. The Seahawks started the season at 65 to 1 to win the Super Bowl.
Now, even after starting the season 5 and 2 heading into their bye, they weren't the oddsmakers' darlings. But, if you used the Massey-Peabody ratings in the NFL Futures Simulator at Unabated, that system was screaming that there was value on Seattle to win it all by week 8. And that's just one bit. The longer the season goes on, the sharper the math gets. Sharper ratings going into the simulator give you a clearer picture of the market and where the blind spots are. You can get ahead of division and conference winners. You can exploit mispriced alternate win totals. You could even take trusted numbers and hang win total props all on your own on prediction markets.
The cool thing about market making on futures markets is they move like glaciers. You don't have to constantly watch the market to protect against getting your pocket picked on news or sudden changes. These are systematically priced events. The only time you have to really sweat adverse selection is if there's breaking news about a quarterback. Now, this tool isn't a magic money printer, but it's a way to turn scattered ratings into a repeatable workflow. It's a system that lets you ask sharper questions about futures every week. What changed? What did the market miss? Where are the vulnerabilities?
And you can get there with a little basic logic, a couple of tools, and the patience to vibe code your way to something that fits your approach. If you can turn team strength into prices, NFL futures stop being a preseason prediction contest and become a weekly market-making logic puzzle that you can solve.
