repository · survival report #5 · 2026-10-12 · method v4 · unranked

vllm-project/llm-compressor

In Survival Report #5 (2026-10-12, method v4): 18,061 of 21,533 lines introduced by 160 AI-tagged commits are still at HEAD, 83.9 %. These are counts, not grades. No rank, no colour, no verdict, and no comparison with any other repository on this page. Every number links to the audit bytes behind it and the method and its limits are public.

GitHub · latest.json · badge.svg · Survival Report #5 · all repositories · Atom feed

160AI-tagged commits · of 3,252 21,533lines introduced by them 18,061still attributed to them at HEAD 83.9 %line-weighted 85.7 %median per commit 24 %of introduced lines in the largest commit

Baseline: this repository's untagged lines

Commits with no machine-readable AI signal, human-written, inline-completed or untagged-agent code alike: 3,086 commits, 923,791 lines introduced, 67,056 still at HEAD (7.3 % line-weighted, 0.0 % median per commit). Age is the time from a commit to HEAD. The baseline was added in method v3 and is unchanged later; definition at causari.dev/method.

Age-matched: AI-tagged lines 83.9 % against untagged lines in the matched age windows 78.0 %, a gap of +5.9 pts, over 5 age windows holding 100 % of the AI-tagged lines.

Line ageAI-taggedcommitsUntaggedcommits
0–30 d98.4 %2284.6 %57
30–90 d74.0 %3178.7 %85
90–180 d65.5 %5179.8 %124
180–365 d80.2 %4158.8 %286
365–730 d43.9 %1533.5 %533
730+ d—01.2 %2,001

By agent, in Survival Report #5

Alphabetical. A commit is attributed to the agent its metadata names; one agent per commit. VERIFIED commits only.

AgentCommitsLines introducedStill at HEADLine-weightedCapped
claude-code7916,45913,97484.9 %74.6 %
cursor11027977.5 %77.5 %
gemini784,6673,71379.6 %78.4 %
github-copilot230529596.7 %96.7 %

History across reports

One row per report this repository was measured in, oldest first; each report measured HEAD as of its own date. The line is the line-weighted ratio on a fixed 0 – 100 % axis; each point is one report.

Line-weighted ratio across reports, 0 to 100 %: #2 80.4 %; #3 80.3 %; #4 80.1 %; #5 83.9 %
ReportDateAI-tagged commitsLines introducedStill at HEADLine-weighted
Survival Report #22026-09-2315417,21013,83880.4 %
Survival Report #32026-09-2815417,21013,81680.3 %
Survival Report #42026-10-0515817,40013,93580.1 %
Survival Report #52026-10-1216021,53318,06183.9 %

Reproduce

One command, any machine, no account: the same bytes the report was built from. The exact bytes behind this page are repos/vllm-project__llm-compressor.json of Survival Report #5. Open an issue with your JSON if it differs.

re audit vllm-project/llm-compressor --json   # method v4, full clone

Badge

Text only, one colour, no red and no green. It follows the latest report and is cached for one hour. badge-dark.svg is the same badge as paper on ink.

AI code survival 83.9 % · causari · #5 AI code survival 83.9 % · causari · #5, paper on ink

README · Markdown
[![AI code survival](https://causari.dev/r/vllm-project/llm-compressor/badge.svg)](https://causari.dev/r/vllm-project/llm-compressor/)

Coverage

  • Method v4, causari 0.4.1. Detection from commit metadata only; survival from git blame -w -M -C at HEAD.
  • Full clone: yes. Method v4 refuses shallow clones; this row comes from a complete history.
  • Sample: 160 AI-tagged commits; the sample floor is 5. At or above the floor, so the ratio is aggregated into the report.
  • PROBABLE: 6 PROBABLE commits (358 lines introduced, 339 still at HEAD) are listed here and never summed into the VERIFIED counts.
  • Cap: 432 lines per commit in this repository; the capped ratio is 75.9 %.
  • .git-blame-ignore-revs: not present in this repository.