(Collaborated with ChatGPT/Gemini)Semantic Kaleidoscope Engine is an executable research prototype for comparing multiple partial representations of the same text corpus without treating any single embedding space as raw meaning. The package preserves source pointers, constructs view-specific neighborhood graphs, applies a minimal fixed-scale signed cellular-sheaf diagnostic layer, ranks passages that behave inconsistently across views, and returns the user to exact source locations for inspection. The included self-contained demonstration uses 24 synthetic passages, four observation views, inspectable signed-graph unit cases, a toy recursive-contraction experiment, five figures, CSV outputs, a Markdown anomaly report, unit tests, and an English LaTeX paper. The project explicitly separates implemented components from future work. It does not claim benchmark validation, raw-meaning recovery, production-ready model-collapse prevention, or completed persistent sheaf cohomology.