iiinova — where your digital self keeps evolving

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

Here's what emerges: everything builds on everything else. A team with elite passing but no finishing will lose. A team with mediocre passing but excellent finishing will win more than their components suggest. Everything is in service of ending the rally in your favor. More in the post.

000

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

Setting is the most volatile skill in the model. A perfect set adds 2%. A poor set costs 12%. An error costs 52%. A bad set kills your whole offensive system. Everything downstream depends on it.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

Serving produces the biggest non-terminal swings. An ace adds 57%. An error costs 43%. An overpass adds 18%. Attacking surprises: a kill adds 40%, but a good attack adds basically zero. Good attacks keep you in the rally—they don't improve your odds on their own.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

The findings: non-terminal contacts have modest individual value. A good pass adds 1%. A perfect dig adds 3%. They matter because they pile up, not because they individually swing rallies.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

We split the data three ways: 2017-2022 train, 2023 validation, 2024-2025 test. Accuracy stays at 60%. AUC stays at 64%. Metrics barely move between splits. The model doesn't learn patterns it can't generalize.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

Before the findings: is the model reliable? At 45% predicted, home teams win 45% of the time. At 65%, they win 65%. The calibration tracks closely to the diagonal across test and validation sets.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 07/07/2026

New post: Introducing CONTACT—rally-level win probability Given everything happening right now in this rally, how likely is this team to win the next point? Took longer than expected to build something we could trust. But now we have it. open.substack.com/pub/nerdabov...

open.substack.com

Where Winning Lives: What Actually Matters

What the model found about which contacts actually shape rally outcomes

111

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

The model shows that fifth sets behave differently than sets 1-4; deficits become terminal faster for underdogs; aggression pays off more late. All of that suggests we should be training different situations different to capture the leverage we know exists. More in the post.

000

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

Conventional wisdom says errors hurt more late. Play conservative. Don't give away free points. But the data shows: positive outcomes increase in value even more. A kill early is worth +3.4%. A kill late is worth +5.3%. Teams should be more aggressive late, not less.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

When deficits become terminal depends on who's on the other side of the net. A heavy favorite down five points still has ~30% chance to win the set. A heavy underdog down three is already under 25%. The talent gap buys margin for error.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

That changes how you should think about timeouts. In a 25-point set, you can afford to work your way back slowly. In a 15-point set, you don't have that luxury. A timeout down three in Set 1 might be premature. A timeout down two at 10-12 in Set 5? That might be necessary.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

One point in a set to 25 is ~4% of the journey. One point in a set to 15 is nearly 7%. Being down 3 points in the first set gives you roughly the same comeback probability as being down 3 in the second, third, or fourth. But being down 3 in the fifth? The margin tightens significantly faster.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

Set 5 carries significantly higher average leverage than sets 1-4. The mean win probability swing per rally jumps from around 4% in the first four sets to 6% in the deciding set. That's a 50% increase in how much each point matters.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 28/04/2026

New post: Training for Leverage—Fifth Sets, Risk-Reward, and How to Prepare for What Actually Matters Win probability models are only useful if they change how you prepare. If they don't change practice, they're just expensive scorekeeping. open.substack.com/pub/nerdabov...

open.substack.com

Where Winning Lives: Training the Leverage

Working backwards from results to process

100

Nerd Above the Net @nerdabovethenet.bsky.social · 25/04/2026

bruh how are you so far ahead of this stream 😂

100

Nerd Above the Net @nerdabovethenet.bsky.social · 25/04/2026

once more i say: brett veach is so lucky he took over a team led by andy reid and patrick mahomes. this dude is so bad at his job 😂

000

Nerd Above the Net @nerdabovethenet.bsky.social · 25/04/2026

man, thank god i decided to stop caring about the chiefs after that stadium deal. i'd be crashing out rn with them going def -> def -> def AND shipping out one of their four day 1 or 2 picks lmao

100

Nerd Above the Net @nerdabovethenet.bsky.social · 25/04/2026

mccoy's medicals have gotta be nasty for him to keep falling and get jumped by his teammate

010

Nerd Above the Net @nerdabovethenet.bsky.social · 24/04/2026

fwiw you've got the texans logo in here twice for 26 and 27

110

Nerd Above the Net @nerdabovethenet.bsky.social · 24/04/2026

jefe living in the future with these announcements lmao

000

Nerd Above the Net @nerdabovethenet.bsky.social · 24/04/2026

every april it becomes even clearer how lucky veach is that the team he took over had andy reid and patrick mahomes 💀💀

010

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

The infrastructure tax wasn't about making numbers prettier. It was about making them actionable. The model works now. The mirror is finally clean. And now that the foundation is solid, we can start asking better questions. More in the post. Good to be back.

000

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

That's not luck. That's A&M identifying where the pressure points were and executing when it mattered. They didn't just win the big points. They found the big points and took them. The model quantifies which moments carry the most weight.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

The highest leverage moment in the match wasn't the final point. It was 3-4 in the fifth set. When A&M took that point to go up 5-3, their win probability swung 10 percentage points in a single rally. Seven of the ten highest-leverage moments went to A&M.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

To test if the new SWEEP was ready, I ran it through the ultimate stress test: Nebraska vs. Texas A&M, 2025 Elite Eight. This isn't a back-and-forth battle. It's a controlled demolition. Nebraska's structural advantages were evaporating point by point.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

The result? A model that was technically right about the parts but fundamentally wrong about the whole. I moved to single-stream architecture. Cleaned the IDs. Fixed the Elo joins. Stopped the model from lying to itself. But a model is only as good as the chaos it can explain.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

The problem wasn't the math. The math was sound. The problem was organization. I was running independent predictions for home and away teams. Battling inconsistent naming conventions where 'University of Nebraska-Lincoln' and 'Nebraska' lived in different silos.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 21/04/2026

New post: The Geometry of Leverage I've been quiet for a few weeks. I was paying an infrastructure tax. SWEEP had a hallucination problem. It would tell you Nebraska had a 40% chance of winning a match they'd literally just lost - so I tore it down. nerdabovethenet.substack.com/p/where-winn...

nerdabovethenet.substack.com

Where Winning Lives: The Geometry of Leverage

How rebuilding SWEEP revealed where Nebraska's advantage collapsed.

210

Nerd Above the Net @nerdabovethenet.bsky.social · 10/03/2026

SWEEP builds on Elo. Elo tells us who's stronger over a season. SWEEP tells us what's happening in the moment. At 19–19 in the fourth set, with everything hanging in the balance, we finally have a way to investigate what's really happening when the scales tip.

000

Nerd Above the Net @nerdabovethenet.bsky.social · 10/03/2026

That's why SWEEP exists. Given the current score, rotation matchup, and team strength—how likely is each team to win the set? A 10–8 lead doesn't mean the same thing in every rotation. Serving into their weakest rotation while in your strongest? That's leverage.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/03/2026

One coach watches A&M leading 18–14 and feels comfortable. Another sees the same score and calls timeout immediately. Both experienced. Both watching the same match. Who's right? Frustratingly: probably both, depending on what happens next.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/03/2026

Think about how we talk about matches in real time: 'They're in trouble now.' 'This is the rotation where they have to hold.' 'If they can just get to 20 first...' These are intuitions about leverage. But without a shared framework, we're just guessing.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/03/2026

Here's what we know: Texas A&M won 15–13 in the fifth. We don't know when the advantage actually shifted. We don't know if that 10-point run in the first set was the moment A&M seized control—or if Nebraska was still in it. The box score is great at who won. Terrible at when.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/03/2026

New post: Introducing SWEEP—Set-Level Win Probability The 2025 Elite Eight: Texas A&M vs Nebraska. Leads didn't stick. Momentum swung wildly. Nothing felt secure. The box score tells us who won. It can't tell us when. That gap matters. open.substack.com/pub/nerdabov...

open.substack.com

Where Winning Lives: The Living Set

Introducing SWEEP: Translating Score, Rotation, and Leverage into Probability

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

Penn State 2024 sits alone in the 2500 club at Elo 2508. Most top-tier programs live between 2300-2400. Crossing 2500 requires perfect convergence of roster and discipline. The Big 12 might be a meat grinder, but it hasn't produced a masterpiece yet. More in the post.

110

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

On the flip side: Creighton and Marquette pay a 'Narrative Tax.' Creighton went undefeated in the Big East. Polls dismiss it as 'Empty Calories.' But Elo sees a team so clinical they refuse to drop matches to inferior opponents. No leaks in the system.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

There's almost zero evidence teams can consistently 'win' five-set matches based on repeatable skill. A five-set match is a tie after four. It's a sprint to 15 decided by service pressure and occasional luck. Elo views close wins as high-variance noise. Polls view them as character.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

The 'clutch' delusion: Colorado 2025 sat 19 spots higher in polls than in Elo. Why? Their 6-1 record in five-setters. To human voters, that looks like grit. Like clutch. To Elo, it looks like a coin flipped seven times that came up heads six times.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

The reality check: inter-conference matchups. When the Big 12 steps outside its silo, the gauntlet narrative flickers. Below 50% vs ACC, Big Ten, SEC. Flat 50% vs Big East. Either they forget how to play in non-conference—or they're mid-tier powers in an echo chamber.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

Is the Big 12 actually inflated? In sports modeling, 'circular logic' is a constant danger. If a conference is full of solidly 'good' teams that only play each other, they trade the same pool of Elo points back and forth. High floors. But maybe mathematical stagnation.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

The Big 12 has the thickest 'middle' in the country. No dead weight. High floors across the board. But notice what's missing: the right tail. The Big Ten, SEC, and ACC push further into elite territory. Their best programs have reached peaks the Big 12 hasn't touched.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 03/03/2026

New post: The Big 12 Mystery—Poll Bias, Conference Strength, and What the Ratings Actually Reveal The Big 12 is either a gauntlet of elite programs—or a thick middle with no peak. In Elo's math, that's a double-edged sword. nerdabovethenet.substack.com/p/ncaa-volle...

nerdabovethenet.substack.com

Where Winning Lives: Math, Myth, and the Modern Game

Conference Gauntlets, 5-Set Coin Flips, and the Anatomy of an Ascent

101

Nerd Above the Net @nerdabovethenet.bsky.social · 24/02/2026

New post: The Texas A&M Story—The Anatomy of a S atistical Heist Nebraska was 33-0. They'd lost seven sets all year. When they met Texas A&M in the Elite Eight, the math was dismissive. Nebraska Elo: 2433 A&M Elo: 2218 A&M had a 22% chance. Then they won. open.substack.com/pub/nerdabov...

open.substack.com

Where Winning Lives: The Texas A&M Story

The Anatomy of a Statistical Heist

010

Nerd Above the Net @nerdabovethenet.bsky.social · 10/02/2026

Elo isn't a ranking. It's not an absolute measure of quality. It's a moving belief—updating as the season unfolds. Always provisional. Always relative. Always open to revision. That's the foundation for everything that follows: set WP, rally-level decisions, isolating leverage.

000

Nerd Above the Net @nerdabovethenet.bsky.social · 10/02/2026

Schedules are wildly uneven. A Big Ten team might play five ranked opponents in three weeks. A mid-major might not face a single one all season. Elo adjusts organically. Every match updates not just the teams that played, but reshapes our understanding of everyone they've faced.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/02/2026

Elo is a way of estimating relative ability. Teams gain or lose rating points based on wins and losses, with adjustments shaped by opponent strength. Beating a strong team matters more than beating a weaker one. Losing to a much weaker team costs more than losing to a peer.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/02/2026

RPI accounts for opponent strength but treats a close loss the same as a blowout. What we need is a system that treats every result as new information and adjusts accordingly. That's where Elo comes in.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/02/2026

The instinct is to let results speak. Wins matter. Losses matter. Titles matter most. That instinct isn't wrong—it's just incomplete. Polls are sticky. They anchor on preseason expectations and struggle to distinguish between genuinely good teams and teams that haven't played anyone yet.

100

Nerd Above the Net @nerdabovethenet.bsky.social · 10/02/2026

New post: Building the foundation for win probability. Nebraska sweeps Penn State. Penn State beats Pitt in four. Pitt takes a set off Nebraska. How good are these teams, really? nerdabovethenet.substack.com/p/where-winn...

nerdabovethenet.substack.com

Where Winning Lives: The Foundation

Using Elo ratings to quantify team strength across uneven schedules

122

Nerd Above the Net @nerdabovethenet.bsky.social · 04/02/2026

It won't replace film study. It won't tell you which rotation to target first. What it does: translate observable performance into leverage. It removes one of coaching's most expensive mistakes—spending a week preparing for the wrong fight. More in the post.

000