Future algorithm adjustments

[bvankat]

• Consider WAB as an additional resume metric?
  • Does Torvik calculate this early in the season?
• Bonus/penalty for Quad 1 wins and Quad 3/4 losses.
  • 3-4 Q1 wins seems like the high end for 2022 bubble teams.
  • Teams with 3 Q3 losses are very bubbly.
  • Only one team in NET top 70 has more than 1 Q4 loss
• Injury adjustments
  • Could base on Sports Reference Win Shares (#)

[bvankat]

Consider: Adjust for victory/defeat margin within Q1/Q2 games. Nebraska didn't play many good teams close in 2023, lots of blowouts. The predictive metrics did not like that. SOR/KPI gave them credit for playing good teams, so there was a 50+ rank difference between the resume and predictive metrics.

UPDATE: Here's some detailed analysis of a similar erratic team: 2023 Missouri — https://www.rockmnation.com/2023/3/2/23621266/missouri-tigers-basketball-analysis-2022-23-net-rankings-explainer-ncaa-tournament-dennisgats

[bvankat]

Related to the margin comment above: Perhaps Kenpom Luck or Torvik FUN or Haslametrics Consistency metrics can be a proxy for explaining that variance? High-luck teams likely have wide gaps in metrics.

[bvankat]

Kenpom has good thoughts on how the quadrants should be broken down: https://theathletic.com/1465470/2019/12/17/kenpom-an-idea-of-a-better-quadrant-system-to-reward-bubble-teams/

• Argues for smaller Quad 1
• Eliminate the games against the lowest opponents.
• Undefeated in Quad 3-4 is a sign of a good team

[bvankat]

Related to the margin comment above: Perhaps Kenpom Luck or Torvik FUN or Haslametrics Consistency metrics can be a proxy for explaining that variance? High-luck teams likely have wide gaps in metrics.

What about looking at margin in Quad 1 and 2 losses? Close losses indicate higher quality — especially road losses. Blowouts would maybe indicate they're less reliable/lower quality.

Nebraska's current losses in 2023-24 🫣:

• Creighton by 26 (Home, Quad 1)
• Minnesota by 11 (A, II)
• Wisconsin by 16 (A, I)
• Iowa by 18 (A, I)
• Rutgers by 6 (A, II)
• Maryland by 22 (A, II)

[bvankat]

Non-conference SOS should be a factor. Really poor ratings definitely affect how the committee views a team. Probably need some kind of penalty for a NCSOS above 250.

[bvankat]

Hoop Explorer lets you build and save a leaderboard with weighted inputs for resume vs efficiency, quality and dominance. Could be a nice way to calculate a bonus/penalty for the quad and margin adjustments. Look especially close at Wins Above Elite. (LINK)

[bvankat]

Might want to add an intangibles bonus for being undefeated in Q3 and Q4. In reality this might offset perceived weakness of playing a soft schedule. In practice this would offset the penalty applied for having NCSOS above 250.

[bvankat]

Game score might be another proxy for game performance vs good and bad teams. You can get a high game score and lose — moral victory. You can get a low game score and win — moral defeat. So what about checking game score in Q1 games and game score in Q4 games and see if they're under- or overperforming?

A better metric would be to calculate expected margin vs actual margin in those games. (LINK)

[bvankat]

Expected margin vs actual margin is Kenpom Luck and Torvik FUN. No need to calculate that separately. And the caveat is that they might not be luck anyway. If the team has a specific skill that causes them to win more games than expected — ex: Good at crunch time free throws — that skill isn't necessarily showing up in any stats other than Luck.

[bvankat]

Idea: Use a large gap in resume vs quality metrics to add intangibles. Torvik's teamsheet page calculates average resume rank and average quality rank. You could find the difference and add some intangibles based on that. Say, +5 intangibles for a high resume rank and -5 intangibles for a high quality rank, since resume is a better judge of who gets in?

Torvik Teamsheets

[bvankat]

Is it possible to game the NET by being really good in Q3? Away wins in Quad 3 might lead to higher-than-expected NET. (Link)

[bvankat]

Some analysis of the 2025 tournament showing WAB is the best predictor of bubble teams. (link)

[bvankat]

Used Torvik's "Compare Tourney Resumes" tool to sort for the top 100 teams to miss the tournament since 2008 (link), sorted by his "Score" column, which reads like a ranking of the resumes. Here's what Claude says about those 100 teams, theoretically the "best teams to miss"

Key Patterns

The Resume Paradox: Teams ranked highly by Torvik's model often have excellent computer metrics (NET, ELO, Power ratings) but lack the traditional "resume" markers that selection committees value. The highest-scoring misses typically excel in predictive metrics while underperforming in results-based categories.
Quality Wins Matter Most: The Q1 W + Q2 W columns are revealing. The best misses often have 11-14 quality wins total, but this still wasn't enough. Having fewer than 10 combined Q1+Q2 wins almost guarantees exclusion, regardless of other metrics.
Bad Losses Kill: The Q3Q4 L column shows that even 2-3 bad losses can be devastating. The highest-scoring misses typically have 0-2 bad losses. Once you accumulate 3+ losses to inferior opponents, you need an exceptional Q1 record to compensate.
The "Above .500" Threshold: The +500 column (record against teams ranked in the top 500) is particularly telling. Teams need at least 6-7 games above .500 against quality competition. Several strong teams missed with only 2-3 games over .500.
Computer Metrics Can't Save You Alone: Teams can rank in the 30s-40s in NET and still miss if they lack wins. Middle Tennessee (2018) and Indiana State (2024) had strong NET rankings but very weak resumes.

Tournament Resume Maxims

Win 4+ Quad 1 games, period. Teams with fewer than 4 Q1 wins almost never overcome this deficit.
Keep bad losses to 2 or fewer. Three Q3/Q4 losses puts you in serious danger; four or more is nearly disqualifying.
Target 11+ combined Q1+Q2 wins. This appears to be the threshold where committees start seriously considering you.
Finish at least 6 games over .500 against the field. The +500 metric matters - you can't just beat up on bad teams.
Resume trumps metrics. A team ranked 30th in NET with a weak resume will lose out to a team ranked 65th with quality wins.
One dimension isn't enough. You need both quality wins AND clean losses AND decent metrics. Excellence in just one area leaves you vulnerable.
The eye test matters. Power ratings (which measure dominance) correlate weakly with selection - committees prefer teams that win close games over teams that dominate weak opponents.
Schedule strength is invisible insurance. Teams that play tough schedules get more Q1/Q2 opportunities and more forgiveness for losses.