Accuracy Report #1: Weeks 1 - 4, 2026

October 5, 2026

Time to reflect on the predictability of the 2025 season so far!

If you’re new to these: Examining predictive accuracy has been a long-standing tradition, underpinning a key purpose of Subvertadown: To give us an understanding of how predictable things have been. 

• We want to know "Was this season more or less predictable than normal?"

• And then we want to know if Subvertadown models are “keeping up” with top sources out there. If there’s good comparative accuracy, then we reassure ourselves that the future week projections are valid considerations for planning and strategy.

Compared to Other Seasons

As usual, here’s a look at how each individual model is doing, compared to other seasons.

This part isn’t meant to tell "how good the models are".  It tells us how predictable the current season is, compared to the historical norm.

The results show that most models are performing at a “normal” predictability level.

In particular, D/ST, RB, and TE are at least as “easy” as the historical average. And even game scores (and Vegas lines) are close to normal levels of randomness, too. That surprised me, because we’ve seen a good share of upsets to Survivor pathways (see section below).

The main exceptions to predictability are WR and QB, which must be due to a number of injuries.

(Meanwhile Kickers always start out looking slower. But as you’ll see below, Kickers have been pretty horrendous for other rankings.)

Comparative Accuracy Assessment

Reminder / for newbies: I don’t expect to be “#1” all the time— that would be plain unrealistic. Especially since I’ve worked to find other experts for comparison who have been consistently good for at least a few years. (Whereas many sources look great one year but then poor the next.) My goal is rather to check that the models are still performing at a similar level to these. From year to year, I find we naturally trade places at the top of accuracy. Knowing that the models perform at least at a "similarly" to top sources lends confidence and gives reason to trust forecasting, when we extrapolate models to future weeks.

Defensive Maneuvers

There’s one success that gets buried if I don’t mention it: for D/STs by ESPN settings (or YDB settings), the Subvertadown models are clearly better optimized for better accuracy. That’s the usual case, and it hasn’t changed.

So to compare on a level playing field.

Here’s the comparison of accuracy across different rankings sources, for the industry standard D/ST settings (which I call the “Yahoo” setting):

Luckily, the Subvertadown model has done great this year. Despite slumping in week 3, due to that week’s Bengals/ Vikings fiasco, the other weeks (1, 2, and 4) saw the model performing as 1st or 2nd best.

So the overall accuracy is off to a great start, relative to the rankings produced elsewhere.

I want to avoid that this whole report is just looking like #1 all the time, so I thought of displaying some other accuracy metrics. I want to emphasize that personally, I’m always judging the performance across several different methods. It’s not good to get overconfident, and I’m always trying to improve them. So let’s look at different accuracy metrics— “rank differential” and the FantasyPros “Accuracy Gap” method:

I don’t trust these metrics as much, but at least they help to reveal that it’s a tighter race— and with slight changes, my models could find themselves at #4 or #5.

The charts still show a good outcome. They show that all ranking sources made lots of mistakes, and mine tended to make a bit less. Last year’s #3 and #5 placed rankers appear to be the toughest competition.

All in all, it this is a good start for the models!

Here’s the Kicker

Cutting to the chase: It looks embarrassingly good…. Nobody has come close to my Kicker models, this year.

I’m sorry, I hate that it sounds overtly self-promotional!

But frankly it’s hard to hide it this time, since the graphs look way more ridiculous than other years.

This is the actual result:

The results are clear: All other Kicker ranking sources have performed pretty bad— usually worse than a coin flip, meaning pure random kicker selection would have been better for most people than using their lists.

They all have rank differentials of near 10.5, compared to my 9.5.

And for additional context, this is what the “Accuracy Gap” evaluation looks like:

Fingers crossed for the remaining of the season!

Personally, I believe in the power of my new “Kicker Situational Adjustment” model— a hugely complex machine learning for this season. I measure its accuracy separately, and it’s clear that it helps to make a big difference. Which justifies my offseason of developing this new method!

Two Cents for a Quarterback

Someone, somewhere out there thinks I’m making this up. But the fact is, the QB model has gone exceptionally well, too— at least in relative terms.
As reported at the topmost chart, QBs have been less predictable than usual to start the season.
But it appears other rankers just made more small mistakes than I did.

We shouldn’t overinterpret. The competition for QB accuracy is fierce— All rankers have a similar 9.15 rank differential, and it can be very hard to find an edge in accuracy. So, as with D/ST, let’s look at the Accuracy Gap measurement, where I come out top-3.

Most of the big errors this year have been when expert rankers lost accuracy due to a QB injury.

QB injuries also affected my model in week 1 (I came in LAST to start the season), and it affected the others in week 2.

All we can do is monitor the rest of the season.

Survivor

Based on probability alone, we should normally expect more than half the pathways to lose once, during 4 weeks of Survivor.  Two or three losses total would be a normal amount, and it would mean that I’d need to display my “backup” pathways that I show each week (1b, 2b, 3b).

In fact, we have seen 4 total losses out of 12 (3 pathways x 4 weeks). Meaning a win prediction rate of 67%.

That means the current results are worse than normal. It’s a somewhat more chaotic season start than usual.

Here are the pathways, with the losses and their back-ups indicated:

• Pathway 1: Bengals (replaced Chargers loss), 49ers , Bills, Ravens

• Pathway 2: Jaguars, Bills (replaced Buccaneers loss), Chiefs, Seahawks

• Pathway 3: Lions, Chiefs (replaced Ravens loss), 49ers, Seahawks (replaced Bills loss)

Betting Lines

Our baseline expectations should normally be to lose about -20% of the weekly pot by this time (after 4 weeks). That means, a normal “coin flip” betting process would lose us about -5% per week, from the target weekly bet amount. Yes, that’s if we were just monkeys shooting darts.

This has been a lucky year so far, with weekly ROIs of +25%, -17%, +18%, and +33%.

This gives a net ROI of 13%, which is in excess of the +5% average target. And in excess of the +10% high target.

As always, temper your expectations. Things will swing down, at some point.

Here are the recorded bet picks and their W/L outcomes, for each of the first 4 weeks:

FLEX position Accuracy (RB / WR / TE)

This is the first year that I launched my RB/WR/TE projections for individual players. I launched the models at the end of week 2.

It’s been going pretty well!

I have mostly been measuring the accuracy by the FantasyPros Accuracy Gap, because it’s an accuracy metric that’s more conducive to the large pool of players (and their distribution of projections and results). But I haven’t developed dedicated charts yet.

Here’s what I can observe:

• RB: For weeks 2 and 3, all experts were very similar to each other in accuracy, myself included. So I at least haven’t lagged in accuracy, and then in week 4 I was noticeably better in accuracy (2 sources did equally well, 5 did worse).

• WR: Accuracy has steadily improved. I was just below average in week 2, and I was a little worried. Then I was exactly average in week 3. Finally, in week 4, it appears I was one of the accuracy leaders (as with RB, I saw 2 sources did equally well at WR, 5 did worse).

• TE: A surprisingly strong start. Weeks 2 and 3 showed my TE model was noticeably better than the 7 sources I compare against. In week 4, all sources included myself had almost identical accuracies.

I didn’t know what to expect, because this was entirely new territory I' got into.

So I’m just happy to see that I’m not falling behind the average of these good sources.

Looking forward to the next 4 weeks!

/Subvertadown