ReWeaver Introduces DriftDetector for AI Code Drift

Overview

ReWeaver has launched DriftDetector, a new tool designed to address the growing concern of AI-induced code drift in modern software repositories. The utility promises to scan entire codebases and identify lines of code that may appear to pass review but are silently failing to meet production standards.

Unlike many current solutions that rely on large language models to interpret code quality, DriftDetector positions itself as a deterministic alternative. It utilizes rule-based assessments across nine distinct production-readiness dimensions to provide consistent, reproducible results without the variability inherent in probabilistic AI systems.

How DriftDetector Works

The platform operates by accepting a GitHub repository URL, which can be either public or private. Upon submission, the tool scans the codebase line by line, evaluating it against a standardized framework of nine dimensions. These dimensions are designed to capture subtle deviations from intended code standards that often occur when AI-generated patches are merged without thorough scrutiny.

A central feature of the tool is the Production Drift Ratio, a single score ranging from 0 to 1. This ratio is weighted by estimated fix hours, providing teams with a clear metric that reflects both the severity and the effort required to resolve identified issues. Because the scoring system is standardized, it allows for meaningful comparisons between different repositories or across different points in time for the same project.

Using probabilistic AI to find a problem created by a probabilistic AI guarantees more drift. Applying more agents doesn't solve the gap in a handoff.

The tool also offers historical context by scoring the full commit history of a repository. This enables developers to pinpoint exactly which commit introduced a gap in standards and track whether the drift is closing or widening over time. Users can view these reports directly in the browser or download them in JSON, PDF, or Markdown formats for deeper analysis.

Deterministic vs. Probabilistic Approaches

A key differentiator for DriftDetector is its rejection of token-based, LLM-driven analysis. The company argues that relying on generative models to detect errors introduced by those same models creates a circular dependency that risks amplifying rather than resolving quality issues. By stripping away prompts, tokens, and context windows, the tool eliminates the unpredictability associated with LLM outputs.

This deterministic approach ensures that the same repo, scanned twice, yields the same score. Every finding is anchored to a specific file and line number rather than being buried in a paragraph of AI-generated prose. This verifiability allows engineers to click through to GitHub and inspect the flagged code personally, removing the need to trust a "black box" recommendation.

Because the system does not call external LLM APIs, there are no per-scan costs or risk of context window exhaustion. ReWeaver states that this architecture is precisely why the core scanning capability can remain free for users, avoiding the metered pricing models common in the AI code-assistance market.

Privacy and Data Security

Given that developers may scan proprietary or internal code, ReWeaver emphasizes a strict privacy-first architecture. Code from private repositories is transferred directly from GitHub to the scanner, bypassing the user's browser entirely. The data streams into temporary memory and is never written to disk during the initial scan.

For the charting component, the system performs a clone that fetches only commit metadata, excluding file contents. The architecture includes a fail-safe mechanism: if a single file blob were to reach disk, the scan would stop immediately rather than continue.

Nothing about a private scan is kept or left behind: no cached clone, no cached scores, no name, no score, no report.

The only persistent data associated with a user is their encrypted GitHub token, which remains stored only until the user disconnects. ReWeaver clarifies that disconnecting or revoking access on GitHub instantly terminates the application's ability to read any repositories. There are no long-term caches of scans or results, ensuring that proprietary code does not linger in the system after the assessment is complete.

Efficiency and Time Savings

DriftDetector aims to quantify the efficiency gains from its scans. The report generated after each assessment includes an estimate of the time saved by the user. It breaks down identified issues into two categories: those that can be auto-fixed via future ReWeaver features and those that require human judgment.

This distinction helps engineering leads allocate resources more effectively, separating mechanical fixes from complex architectural decisions. By providing a standardized score and pinpointing exact locations of drift, the tool reduces the time traditionally spent hunting for subtle quality regressions in large codebases.

Users interested in staying informed about updates and new features can join the ReWeaver mailing list. As the tool continues to evolve, it seeks to provide a reliable, non-probabilistic layer of quality assurance in an era where AI-generated code is becoming increasingly prevalent in professional development workflows.