Natural versus Manufactured Scoring

Phase 9 evaluates binary, three-class, five-class, and hierarchical mappings against approved human labels. Transparent fusion reports contributions, missing evidence, scores, thresholds, uncertainty, coverage, and abstention; it does not turn experimental scoring into objective truth.

The experimental rules engine returns naturalness, manufacturedness, and unknown values normalized to sum to one. Semantic, geometry, texture, frequency, shape, and motion evidence each carries scores, uncertainty, and an explanation. Missing evidence increases unknown rather than counting as zero.

Object crop → semantic evidence ─┐

            → geometry/texture ──┼→ versioned weights → normalized scores + explanations

            → frequency/shape ───┤

            → motion (optional) ─┘

These scores are research hypotheses, not proof of an object’s origin. Frequency patterns, line geometry, or apparent fractal complexity cannot independently establish whether something is natural or manufactured. Human review remains necessary.