How much of this code has no human on its recent record?
A read-only comprehension-debt Map of famous open-source repos. Every debtpercentage here is the same number in the same direction: the share of scored code, weighted by bytes, that no human has recently written, reviewed, or explained — higher is worse. The one other number on a card is coverage (how much of the code we scored), where higher is better. Open any repo for the full decomposition.
750 KB of code sits below 0.30 — that is the 23% above, as an absolute.
Human review depth scores zero on 32% of this code, by bytes — that share has never had a substantive human review. It is one of four factors; the 23% above combines them all.
as of 2026-07-28coverage: scored 100% of the codegithub.com/bellard/quickjs ↗
248 KB of code sits below 0.30 — that is the 21% above, as an absolute.
Human author recency scores zero on 100% of this code, by bytes — that share has no recent human author. It is one of four factors; the 21% above combines them all.
as of 2026-07-28coverage: scored 100% of the codegithub.com/karpathy/llm.c ↗
86.6 KB of code sits below 0.30 — that is the 19% above, as an absolute.
Human review depth scores zero on 47% of this code, by bytes — that share has never had a substantive human review. It is one of four factors; the 19% above combines them all.
as of 2026-07-28coverage: scored 100% of the codegithub.com/karpathy/nanochat ↗
170 KB of code sits below 0.30 — that is the 17% above, as an absolute.
Human review depth scores zero on 28% of this code, by bytes — that share has never had a substantive human review. It is one of four factors; the 17% above combines them all.
as of 2026-07-28coverage: scored 100% of the codegithub.com/simonw/llm ↗
62.1 KB of code sits below 0.30 — that is the 13% above, as an absolute.
Human author recency scores zero on 71% of this code, by bytes — that share has no recent human author. It is one of four factors; the 13% above combines them all.
as of 2026-07-28coverage: scored 83% of the codegithub.com/expressjs/express ↗
149 KB of code sits below 0.30 — that is the 6% above, as an absolute.
Human author recency scores zero on 19% of this code, by bytes — that share has no recent human author. It is one of four factors; the 6% above combines them all.
as of 2026-07-28coverage: scored 71% of the codegithub.com/honojs/hono ↗
1 MB of code sits below 0.30 — that is the 6% above, as an absolute.
Human author recency scores zero on 92% of this code, by bytes — that share has no recent human author. It is one of four factors; the 6% above combines them all.
as of 2026-08-19coverage: scored 100% of the codegithub.com/getzep/graphiti ↗
Notice: the two healthiest repositories here are getzep/graphiti (6%), where an agent's signature sits on the latest commit for 7% of the code, and honojs/hono (6%), where almost none of it does. The most indebted, bellard/quickjs (23%), has almost none either. Agent authorship does not predict the number — debt tracks how deeply humans review, not who or what wrote the code. A commit counts as agent-signed when an agent appears on it, alone or beside a human.