The website is unlocked and in off-season mode. The 2023 season is displayed, and you can explore all the site features freely!

Subvertadown Overview

Nick Folk PHILADELPHIA, PA - JANUARY 01: Philadelphia Eagles wide receiver DeVonta Smith (6) is upended by New Orleans Saints safety Daniel Sorensen (25) during the game between the New Orleans Saints and the Philadelphia Eagles on January 1, 2023 at Lincoln Financial Field in Philadelphia, PA. (Photo by Andy Lewis/Icon Sportswire)

Subvertadown uses a scientific approach to provide fantasy football players with forecasts and strategic insights.

Subvertadown is especially geared towards players who appreciate:

  • High transparency: openness about the methodology used in the predictive modeling, so you understand what’s behind the outputs.

  • A scientific approach: analytical studies are conducted rigorously, in order to test whether hard data validates assumptions about forecasting.

  • Accuracy reporting: Comparing accuracy results to top-performing sources, to validate the approach and bolster confidence.

There are several articles about the background to the project. Just select the Articles tagged “Background” to learn more.

What I try to do

For the last 6 years, I have been on a mission to spread understanding of fantasy football statistics. I grind out serious stats to try and improve insights. For many of you, the studies and forecasting simply make the fantasy experience more fun. For others, it makes you feel less apprehensive in making your picks, when you know what projections are supported by my methodology.

I think it's comforting to make lineup choices based on tested and validated stats. I think its exciting to see odds improved from predictive models driven by scientific, data-backed methodology. When I publish studies of my analytical insights, the goal is to convey a deeper understanding of how much you can control game… versus how much you cannot! Fantasy football has a lot of inherent randomness, and I try to report very openly about that. It's less frustrating if we understand how much of the game is out of our control.

The following features are what I found lacking in other rankings— and they are the elements my followers have found valuable:

  • Predictive models for streaming, with high relative accuracy. I found many models were based on guess work.

    • Projections for future weeks, to assist streaming strategy

  • Transparency: regarding methodology for development, and through self-reported accuracy evaluations. I was not able to find rankers who were interested in openly describing their calculations— or interested in numerically comparing their accuracy.

  • Statistical analysis: articles with data-backed insights, to impart understanding about the game.

Better insights can assist with better strategy. Future-week projections can give an advantage in waivers. Measuring accuracy and transparent methodology can give confidence in the approach— because things often go wrong in fantasy.

The above features are described in more detail, below:

Examples of statistical analysis

In my previous posts, I've previously covered topics like:

  • how to measure predictability and randomness of the different positions

  • when streaming/holding a certain position makes sense

  • what kinds of information is predictive or not

  • what kinds of accuracy measurements are trustworthy

  • how the different positions impact each other within a team.

You can find a summary of these in this post.

Streaming Posts

Most people know me through my weekly posts, especially for streaming Kicker and D/ST. While I think my D/ST projections have got more attention, I'm happy my Kicker rankings get almost as much interest-- because Kicker streaming is where (I can measure) I can make the most difference. Meanwhile my in-season QB model has been quietly distinguishing itself.

Examples of streaming posts: for D/ST, and for Kicker

Accuracy evaluations

For me these serve 3 purposes:

  1. I found myself wishing other sources would report their own accuracy,

  2. This is the primary means of self-correction, for the continual improvement that has always been part of the project, and

  3. I want to be sure I'm not putting out sub-par rankings. If my approach was worse than others, then I'd just be wasting a lot of time. My models have improved over the years, and it's only because of tracking accuracy that I feel convinced to trust my own models.

Here are two examples: (1) weekly accuracy report, (2) extra accuracy overview.

All the above are meant to foster greater understanding and insight, for those interested in this kind of scientific approach.