This paper presents a systematic reanalysis of experimental data on condensed matter nuclear reactions over the past three decades. Based on the physical fact that deuteron wavefunctions are confined in nanoscale potential wells in condensed matter environments, this paper argues that the conventional approach of interpreting data using free-space conceptual frameworks—such as “equivalent temperature,” “kinetic energy enhancement,” and “electron screening”—constitutes a hidden category error. Through a statistical comparison of the effective fusion rate constants Γ_eff between carbon-based systems (n = 31) and metal-based systems (n = 18), this paper reveals a robust difference of about 8-fold (p < 0.001), which can be independently explained by the geometric effect of the confinement scale, indicating that the baseline fusion probability is determined by the microscopic geometry of the material. Segmented scaling law analysis further shows that the system exhibits superlinear power growth (α = 1.28) below a critical volume and becomes linear (α = 0.89) above it, a behavior consistent with the spatial boundary condition of network coherence. Based on these analyses, this paper proposes a fundamental taxonomic difference between cold fusion and hot fusion: hot fusion relies on kinetic energy to overcome the Coulomb barrier and is a deterministic “Motion-Penetration” process that utilizes disordered thermal energy; cold fusion relies on spatial coincidence upon wavefunction collapse and is a non-deterministic “Existence-Refresh” process that can only utilize ordered energy, with the complete reaction equation D + D + 2e⁻ → ⁴He + 2e⁻ + coherent energy, in contrast to the bare‑nucleus reaction D + D → ⁴He + γ of hot fusion. This taxonomy redefines the theoretical foundation and engineering direction of condensed matter nuclear reactions.