Blog
Comedy Analytics
Deep dives into what makes TV comedy work — backed by data, not opinions.
5
fan theories, tested
May 25, 2026 · 5 min read
We Re-Cut Five Sitcoms by the Numbers — and Busted Two Fan Theories
Our new Explorer lets you rebuild any sitcom and see how funny that cut scores. The Arrested Development revival really was worse — but The Office didn't fall off after Steve Carell, and the Schitt's Creek glow-up doesn't show up in the jokes.
77.9
Community Humor Index (#5)
May 24, 2026 · 8 min read
You Can See Exactly When Dan Harmon Left Community
We scored all 110 episodes of Community. It lands at Humor Index 77.9 — fifth of nine shows. But the real story is the season curve: it tracks the show's behind-the-scenes history almost to the decimal. The gas-leak year is real, the Harmon rebound is real, and the season everyone defends is quietly the weakest.
6 / 80
HI ∩ IMDb top-10 across catalog
May 16, 2026 · 8 min read
We Recomputed Our IMDb Correlation Three Ways. At the Show Level, It's Negative.
Our April post reported r = -0.005 between the Humor Index and IMDb. We dug back in. Within-show correlations span -0.115 to +0.392. The show-level correlation is -0.287. HI's top 10 and IMDb's top 10 overlap 6 of 80 across the catalog — exactly chance.
77.4
Taxi Humor Index (#7)
May 16, 2026 · 9 min read
Taxi Lands at 77.4 — A 1978 Show Inside the Same Tier as Seinfeld
We just scored all 114 episodes of Taxi. It lands at Humor Index 77.4 — statistically indistinguishable from Seinfeld, Friends, and Schitt's Creek. The format we used to penalize for is the format every modern character comedy descends from. And Louie De Palma cracks the top 10 cross-show.
8 of 10
AD characters in the craft top 10
May 15, 2026 · 8 min read
Arrested Development Has 8 of the Top 10 Best-Crafted Characters in TV Comedy
Of the ten characters with the highest per-joke Craft scores on our index, eight come from one show. The Bluth family owns the craft leaderboard. The other two outliers — Ron Swanson and Creed Bratton — are revealing in their own right.
1,181.9
George Costanza's career WAR
May 12, 2026 · 8 min read
George Costanza Just Beat Jerry Seinfeld for the Funniest Character in TV Comedy. Here's the Full Cross-Show Leaderboard.
We finally have six shows scored under v2 consensus. George Costanza tops the cross-show WAR leaderboard at 1,181.9 — narrowly edging Jerry Seinfeld. Ron Swanson has the highest per-joke Craft score of any character on the index. Here's the top 25.
69.1%
Schitt's character comedy share
May 3, 2026 · 7 min read
Modern Sitcoms Are More Character-Driven Than the Classics
Across 6 fully-scored shows, character_comedy is the most variable axis in our taxonomy — a 45-point spread. Schitt's Creek tells more character-driven jokes than Seinfeld. By a factor of nearly three.
78.3
Schitt's Creek HI (#5)
May 2, 2026 · 7 min read
Schitt's Creek: Last on the Board, First on Impact
Schitt's Creek scored the lowest of the five published shows when it debuted — but ranks #1 on Impact and #2 on Craft. One of the cleanest demonstrations our methodology has of why joke count alone is the wrong question.
85.2
AD Humor Index (#1)
May 4, 2026 · 8 min read
Arrested Development Just Took the #1 Spot. The Gap to #2 Is the Biggest on the Board.
Arrested Development debuts at 85.2 — 4.65 points clear of Parks. That's twice the size of the gaps between #2 and #6 combined. Here's what's actually inside the gap.
80.55
Parks Humor Index (#1)
April 30, 2026 · 7 min read
Parks and Recreation Just Took the #1 Spot from The Office.
After scoring all 126 Parks and Rec episodes, Pawnee edges Scranton 80.55 to 80.22. The margin is inside our noise floor — but every secondary metric points the same direction, and Ron Swanson is now the highest-quality lead character on the site.
ICC 0.28
Single-run reliability
April 17, 2026 · 9 min read
We Rescored 30 Episodes Twice. Our Single-Run Humor Index Has an ICC of 0.28.
A test-retest study on 30 episodes. Show-identity bias is tiny (not significant). But the scorer is noisier than we thought — individual episode Humor Indexes are only 28% signal, 72% run-to-run variance. Here's what we're doing about it.
7.9%
Variance between shows
April 17, 2026 · 8 min read
We Fitted a Bayesian Model to 15,000 Jokes. Every Show Ranking Is Within Noise.
A hierarchical Bayesian model of joke impact on 15,000 jokes. Format effect: statistically zero. Top-tier shows' credible intervals all overlap. 64% of joke-level variance is unexplained within-joke noise.
1,181.9
George's current WAR (#1)
April 16, 2026 · 7 min read
Jerry Seinfeld Is the Most Valuable Comedy Character in Television History
Our April 16 analysis of Comedy WAR. Post includes an April 19 update: after fixing a standup-weighting bug, George Costanza overtook Jerry for #1.
78.6 vs 77.8
Office vs Seinfeld (current)
April 16, 2026 · 6 min read
Seinfeld Just Passed The Office on Our Humor Index. Here's Why.
Our April 16 analysis of removing the multi-cam penalty. Post includes an April 19 update: after a standup-aware rescore, Office retook the top spot at 80.2 vs Seinfeld 79.1.
r = −0.005
Pooled correlation
April 12, 2026 · 7 min read
IMDb Ratings vs. The Humor Index: Does "Funny" Mean "Good"?
We compared 591 episodes across The Office, Seinfeld, and Friends against IMDb. Pooled correlation: r = −0.005. Audience ratings and comedy craft are essentially unrelated.
186
Episodes Analyzed
April 10, 2026 · 8 min read
Is The Office Actually Funny? We Analyzed Every Joke to Find Out.
We ran all 186 episodes of The Office through our AI comedy analyst. The results surprised us — some "classic" episodes aren't as funny as you think.
5
Dimensions
April 10, 2026 · 6 min read
How We Score Comedy: The Math Behind the Humor Index
Peak density, effective craft, memorability bonus — here's the complete breakdown of how we turn thousands of joke scores into a single number.
Retracted
See update note in post
April 10, 2026 · 5 min read
Should Laugh Tracks Be Penalized? Our Data Says Yes.
Our original April 10 argument for a laugh-track penalty — and why we removed it a week later when a Bayesian audit showed the effect was indistinguishable from zero.