Hacker News front pageAug 27, 2026, 16:59 UTC+8

The load-bearing vocabulary of Claude: a word-frequency project finds a concentrated set of terms in Claude-authored PRs in 2026

Show HN: The load-bearing vocabulary of Claude

The project scraped 47,464 GitHub PRs over 595 days and clustered them into 8 vocabulary groups using KL-divergence k-means. One cluster emerged in 2026 and accounted for 45% of human-attributed PRs last month. Its top words—load-bearing, latent, genuine, seam, ladder—match terms reported by Claude Code users. The author interprets this as a fingerprint of Claude’s writing style in code, not natural human usage. The post doesn’t spell out how “human-attributed” is defined or what the mislabeling rate might be.

Why it matters: Solid methodology (KL-divergence k-means on 47k PRs over 595 days) with a striking finding: Claude Code's vocabulary cluster now appears in 45% of human-attributed PRs. Observational rather than a product release, so capped at 78.

Read the original ↗Export Markdown

Take

The number that makes this worth clicking: 45% of human-attributed PRs last month fell into a vocabulary cluster that didn't exist before 2026. The author scraped 47,464 PRs over 595 days, ran KL-divergence k-means to get 8 clusters, and one cluster's top words—load-bearing, latent, genuine, seam, ladder—match exactly what Claude Code users have been reporting in issue threads.

I'd discount the 45% a bit. The post doesn't define "human-attributed"—if it just means the commit author isn't a bot account, then a PR Claude wrote but a human committed would count as human. That makes 45% an upper bound on Claude penetration, not a precise measurement. Clustering also tends to pull borderline cases into the dominant group.

But the direction holds. Load-bearing appears 123× more often in this cluster than in the background corpus. That's not random drift. This isn't "AI is polluting GitHub"—it's a measurable fingerprint of Claude Code's writing style showing up in open-source collaboration at scale.

Source: Hacker News front page · louisabraham.github.io
Published: Aug 27, 2026, 16:59 UTC+8 · Aug 27, 2026