We Tested Whether Engineering Decisions Are Recoverable From Git History

Canonical article: https://blog-archaeo.vercel.app/

Most code outlives the reasons it was written.

Months later someone opens a file and asks:

Why is this line here?

git blame tells you who last touched it.

Usually that's a formatter, a rename, or someone moving files around.

It rarely tells you why the behavior exists.

I built archaeo, an open-source CLI that traces a line back through moves, renames, refactors, squash merges, and cherry-picks to recover:

• the commit that introduced the behavior
• the merged PR
• the linked issue
• the review discussion behind the decision

The rule I wouldn't compromise on:

If the evidence doesn't exist, archaeo says "No recorded decision found."

No hallucinated explanations.

Before publishing I validated it on 150+ real queries across Kubernetes, React, Cognee, and other repositories.

Some highlights:

• 97.7% PR recovery on PR-driven repositories
• Real review discussions recovered years later
• Local-first
• MIT licensed
• Bring your own LLM key (or run completely offline)

The full validation, benchmarks, architecture, and implementation details are in the article.

👉 Read: https://blog-archaeo.vercel.app/

⭐ GitHub: https://github.com/vanshitahujaa/archaeo

bash
npm install -g git-archaeo