Identification Capacity and Rate-Query Tradeoffs in Classification Systems

We study zero-error class identification under constrained observations with three resources: tag rate L (bits per entity), identification cost W (attribute queries), and distortion D (misidentification probability). We prove an information barrier: if the attribute-profile map pi is not injective on classes, then attribute-only observation cannot identify class identity with zero error. Let A_pi = max_u |{c : pi(c)=u}| be collision multiplicity. Any D=0 scheme must satisfy L >= log2(A_pi), and this bound is tight. In maximal-barrier domains (A_pi = k), the nominal point (L, W, D) = (ceil(log2(k)), O(1), 0) is the unique Pareto-optimal zero-error point. Without tags (L = 0), zero-error identification requires W = Omega(d) queries, where d is the distinguishing dimension (worst case d = n, so W = Omega(n)). Minimal sufficient query sets form the bases of a matroid, making d well-defined and linking the model to zero-error source coding via graph entropy. We also state fixed-axis incompleteness: a fixed observation axis is complete only for axis-measurable properties. Results instantiate to databases, biology, typed software systems, and model registries, and are machine-checked in Lean 4. 14 pages; 296 machine-verified theorem/lemma statements; 6,707 lines of Lean 4 across 14 files; 0 sorry.