The State of Comprehension Debt: I measured 22 major repos

Grasp, a new tool that measures comprehension debt from git metadata alone, has benchmarked 22 major repositories and found that AI writes an increasing share of code while human understanding of codebases declines. The tool produces a dark-code ranking in 60 seconds without reading source code, using only commit history and PR events to calculate a deterministic Grasp Score.

AI writes more of your code every quarter, authors leave, and reviews keep getting thinner. Grasp measures the number no dashboard tracks: how much of your codebase any currently-active human still genuinely understands. It works from git metadata alone, without reading a line of your source. Sixty seconds from repo URL to a dark-code ranking. No signup, no code access. Prefer to look before you scan? See a full sample report /sample , our 22-repo benchmark. Curious how the number is made? The methodology is published in full /methodology . Read the full report: The State of Comprehension Debt 2026 /state-of-comprehension-debt — 22 repositories measured, what the numbers mean, and the method behind them. Acquiring a codebase, auditing your own, or losing a key engineer? Fixed-fee diligence reports are available /diligence , with deeper analysis delivered in five business days. Metadata only. Grasp reads commit history and PR events, never file contents. Hard evidence only. AI attribution counts co-author trailers and bot accounts; we report it as a floor, not a guess. Honest limits. The Grasp Score is a potential indicator. It tells you where to look, and we say so on every report. No AI inside. Grasp itself uses no AI: every score is deterministic, reproducible arithmetic over git metadata. Run it twice and you get the same number, and the raw data lets you re-derive it yourself.