gemma_noise_detector_prototype

To classify files as noise or signal in a codebase, use a rule-based detector that evaluates file paths, extensions, sizes, and directory context to automatically label clutter versus valuable data assets. This generates a per-file JSONL report for cleanup planning.

Rule-based noise detection is the process of identifying and labeling individual files as NOISE or SIGNAL using explicit patterns like extension, age, backups, and critical-path overrides. It helps teams distinguish clutter from valuable files to enable faster maintenance and reduced manual review.

Yes, the noise-vs-signal file classifier works across code, documentation, and data assets, explicitly supporting common extensions like .jsonl, .log, and .json. It applies directory context and critical-path overrides to accurately categorize these formats.

Run the detector on a repository to classify files and emit a JSONL labels file. This output produces a JSONL record for each file containing fields such as label, category, confidence, and execution_id for automated cleanup planning.

To identify noise files in a repository, the classifier requires inputs including the file path, file extension, file size in bytes, last modified days, and parent directory. These parameters drive the rule-based classification patterns.

Yes, you can use explicit pattern matching to separate clutter from valuable data assets by applying rules based on file extension, age, backups, and directory context. The classifier handles code, documentation, and data files while applying critical-path overrides.