Skip to content

Your Repo. Your Results.

A benchmark on someone else's code proves nothing to you. Pick a repo you know — and judge the findings yourself.

1. Enter a public GitHub repo

Paste the URL of any public Python repo on GitHub. The analysis runs entirely in your browser — code is fetched via GitHub's public API and never sent to any server.

🔒 Code is fetched via GitHub's public API and analyzed in your browser. No server, no analytics, no cookies.

Rate limits

GitHub allows 60 API requests per hour without authentication. A typical analysis uses 10–35 requests. If you hit the limit, wait a few minutes and try again.

2. Check the findings yourself

Don't take drift's word for it. Here are three things to verify against your own knowledge:

The finding you already knew about

Every codebase has that one module that grew too fast, that one file everyone avoids, that one pattern that got copy-pasted three times. Look at the top findings — do you recognize the files?

If drift flagged something you've been meaning to fix anyway, that's not a coincidence. That's signal.

The duplicate you didn't notice

Look for MDS (Mutant Duplicates) findings. These are functions with the same name defined across different files — the classic artifact of AI-assisted development where each prompt produces a self-contained solution.

Ask yourself: are those functions really different? Do they need to exist separately?

The import that shouldn't be there

Look for CIR (Circular Import) findings. These flag modules that import each other — creating hidden coupling that makes refactoring dangerous.

Check: is the circular dependency intentional, or did it creep in during a fast iteration?

3. Want the full analysis?

The browser preview analyzes a subset of Python files using lightweight pattern detection. For the full 23-signal analysis including git history, AST parsing, and co-change patterns:

uvx drift-analyzer analyze --repo . --format json --compact

pip install drift-analyzer
drift analyze --repo . --format json --compact

pipx run drift-analyzer analyze --repo . --format json --compact

docker run --rm -v "$(pwd):/repo" ghcr.io/mick-gsk/drift:latest analyze --repo /repo --format json --compact

Or drop a local result file here:

🗎 Drop drift-results.json here
·

4. What if drift found nothing?

That's fine. It happens when:

  • The repo has fewer than ~10 Python files
  • The codebase is genuinely well-structured
  • The analyzed subset (browser preview) didn't cover the most affected files

None of these invalidate the tool. Small repos simply have less surface for structural drift to develop. Try it on a larger project — or run the full local analysis.

5. What if drift found something real?

That's the proof. Not our proof — yours.

Next steps:

6. What if a finding is wrong?

False positives exist. Drift is at 77% strict / 95% lenient precision — which means some findings may need human judgment.

If you spot one:

A proof you construct can be disputed. A proof you initiate cannot be denied.