Contains FrameTimecode and Timecode classes that handle frame-to-time conversion, exact rational arithmetic via Python's Fraction, and PTS-based timing for variable frame rate video. This module is self-contained and could be extracted as a standalone timecode utility for any video project requiring precise VFR-aware time manipulation.

Implements StatsManager, which caches per-frame detector metrics (e.g., content_val, delta_hue) to a CSV file. This enables incremental re-analysis without re-decoding, and the CSV export allows users to inspect or tune thresholds. The component is isolated, depends only on FrameTimecode, and demonstrates a reusable pattern for two-pass metric storage in video analysis pipelines.

Provides FanOutVideoStream, which fans a single VideoStream decode into N consumer streams, allowing multiple detectors or benchmarks to process the same frame sequence concurrently without redundant decoding. This is a non-obvious optimization pattern originally used for parameter sweeps; it is isolated and could be reused in any multi-consumer video analysis workflow.

Implements greedy 1-to-1 matching for hard cuts and point-in-interval matching for fades, producing precision/recall/F1 per video and aggregate metrics. The score_video() and evaluate() functions are dataset-agnostic and could be extracted for any shot-boundary evaluation task, requiring only ground truth and detected scene lists in a standard format.