Beyond Symptom Counts: A Theory-Driven, Computation-Based Assessment Framework Evaluated in Two Independent Cohorts

Preprint. Background. Reflective self-report totals can obscure differences in the organization of responses. The Holographic Coding Framework (HCF) instead maps situation–response items to three-element structural codes, composes a conflict graph, and computes carrier, process, and coherence readouts. Objective. We evaluate the readouts' psychometric behavior and whether coherence retains information beyond code counts and identical symptom totals under an explicit proprietary disclosure boundary. Method. Two cohorts (English N = 462; Chinese N = 504; gold-standard subsample n = 142) and a second wave (retest n = 164; dual-form n = 166; cross-batch n = 240) were analyzed. One finalized backend returned four α readouts, four carrier readouts, and H. Analyses covered wording similarity, validity, reliability, equivalence, H-versus-count comparisons, exact-total symptom-pattern permutations, and exploratory profiles. Results. No stem/option pair exceeded the 0.8 wording-similarity threshold (0/10,200), and score–similarity associations were nonsignificant; internal semantic independence remains unevaluated. Convergent associations were in predicted directions (α1–MAIA-2 .357; α2–AAQ-II −.581; α3–DCS −.416; α4–BADS .627), but 0/4 met the registered discriminant criterion. Retest ICC(3,1) was .331–.600 (H = .538). Dual-form equivalence held for α1/α4; registered cross-batch equivalence held for α3/α4 but not α1/α2. Against the theory-primary full-pool distinct-code count, H explained 5.96×/1.91× as much PHQ-9 variance and 5.13×/2.13× as much GAD-7 variance in the English/Chinese cohorts. A clinically valenced non-harmonious-code sensitivity count also favored H in every comparison. H's AUCs were .728–.828; all eight paired H-minus-count bootstrap CIs excluded zero. Within identical PHQ-9/GAD-7 totals, H-defined groups showed symptom-pattern distances 1.57–2.59 times their permutation null means (p ≤ .011), with continuous H–pattern partial R² = .0233–.0972. The α readouts were highly collinear (PC1 = 91.4%/92.7%); joint exploratory profile increments did not survive four-criterion correction, and disease associations neither consistently replicated nor survived symptom adjustment. A post-data latent-distress subgroup (all four α readouts in cohort-specific low tertiles with PHQ-9 < 10) showed higher chronic-pain prevalence in the Chinese cohort (42.4% vs 14.8%; OR = 4.24), an exploratory, hypothesis-generating contrast. Conclusion. HCF provides a testable structure-aware alternative to pure counts: H outperformed the specified count baseline, and H-defined groups retained pattern differences within identical totals. Nevertheless, evidence supports bounded, outcome-specific evaluation rather than comprehensive validation. Public reproduction remains limited to exported readouts and downstream analyses; internal mechanisms, diagnostic utility, causal prediction, and treatment allocation remain unestablished.