Analyze a repository

Which files are most likely to need a bug fix next, and why.

EvoMind reads a repository's commit history, learns which files went on to need bug fixes, and ranks today's files by the chance that they will need one next. Paste a path or Git URL above and press Analyze.

Try a public repository

How it works

Mine the history

Clones the repository and reads every commit: which files changed, how often and by how much.

Trace bug fixes

Finds fix commits, then uses SZZ (git blame) to mark the files that held the bug before the cutoff.

Train and compare

Builds code and history metrics per file and trains Random Forest against Logistic Regression; the higher ROC-AUC wins.

Rank and explain

Scores today’s files, shows what drives each score and how the risk has moved over time.

Highest risk

—

Files with rising risk

—

Bug-prone in training

—

Model ROC-AUC

—

Risk ranking

Chance of a bug-fixing commit within the horizon, scored on the latest commit. Each file lists the measurements that raise its score most.

#		Band			Actions

Risk is a predicted probability, not a certainty. Fix commits are found by commit-message keywords, then traced back with SZZ (git blame on each fix's parent) to confirm the defect already existed at the training cutoff. Treat scores as a directional signal. "Ask AI" sends that file's source and metrics to Groq's API (a third party) to write a recommendation.

Coupling

Import graph parsed from source: 0 edges inside the repository. Instability is efferent ÷ (efferent + afferent); near 0 means many files depend on it, so changes ripple.

Circular dependencies

Files that import each other in a cycle. Usually a sign they should be merged or share an extracted piece.

Evaluation

Held-out files from the training cutoff. The model with the higher ROC-AUC is the one that scores the repository.

Model	Precision	Recall	F1	ROC-AUC	Avg. precision

Top predictors

Feature importances from the selected model.

Stability across time

Trains at 50%, 65% and 80% through the history and tests only on the period after each cutoff. Repeats the pipeline three times, so it runs on request.

Starting the scan…

What's going on inside your code?