CAL-L3: The Tensor Volume Layer — Composition Operator C and the Governance Manifold Hypothesis

Working paper v0.2 (characterization phase closed, ~95%) for Layer 3 (L3) of the Cognitive Abstraction Layers (CAL) architecture. Main result: the composition operator C is characterized — causal conservation under compression is preservation of the observational invariants Ω₀ = (Reachability, Coverage, Consistency), not reconstruction fidelity. A low-rank Tucker compression alone fabricates spurious causal edges; a structural-mask prune recovers 75% of the causal-observability headroom (U: 0.441→0.862) with no ground truth required at deployment. The residual ~25% is declared as an open frontier (ΔU≈0.138). All hypotheses were pre-registered before data collection. Part of the CAL five-level (L0–L4) research agenda. See companion CAL pre-paper, concept DOI 10.5281/zenodo.20430343.