Formation Taxonomy: Pattern Emergence, Demographics, and Hierarchical Structure

Overview

This whitepaper synthesizes the internal evidence on the Icosa model’s formation system, drawing on three completed studies that examine whether higher-order personality formations emerge from grid geometry, whether topology metrics behave according to their geometric specification, and whether the dynamics subsystem produces a momentum signal coupled to personality integration. All evidence derives from synthetic profiles generated within the Icosa computational framework. No human-respondent data, external criterion measures, or clinical outcome data inform these findings. The synthesis is organized around what the studies reveal about architectural fidelity, emergent structural regularities, boundary conditions, nulls (or their absence), and the research priorities that follow.

Formations in the Icosa model are higher-order personality structures derived from a profile’s Coherence Band and grid variance pattern across the 4x5 capacity-by-domain matrix of 20 centers (Harmonies). The model defines 76 formations, each representing a distinct structural configuration of personality organization. Unlike prototype-matching classification systems, formations are computed from the intersection of grid-level properties — completion, pair density, topology metrics, and dynamics indices — rather than assigned by similarity to fixed templates. The three studies reviewed here test whether this computational derivation behaves as specified: whether grid properties predict formation structure, whether topology metrics track integration and dysfunction, and whether dynamics momentum correlates with the global Coherence metric. Together they constitute the first systematic internal-consistency benchmark of the formation pipeline.

Benchmark Scope and Evidence Classification

The synthesis draws on three studies, all classified as formula verification under the evidence taxonomy.

formation-emergence (“Emergence of Personality Formations From Grid Geometry in the Icosa Model”). Evidence type: formula verification. This study generated N = 1,000 synthetic profiles using the d40 engine (seed = 42) and tested 11 confirmatory hypotheses: whether pair density predicts Coherence (H2), whether grid completion inversely predicts resonance total (H1), whether each of five domain-specific column completion indices predicts its corresponding Domain Health score (H3, five tests), and whether each of four capacity-specific row completion indices predicts its corresponding Capacity Health score (H4, four tests). All tests used Pearson correlation with Holm-Bonferroni correction across the full 11-test family.

pattern-demographics (“Topological Properties of Personality Formations and Their Structural Correlates”). Evidence type: formula verification. This study generated N = 1,000 synthetic profiles using the Icosa framework (seed = 42) and tested two hypotheses: whether topology fulcrum health predicts Coherence (H1, flagged under circularity governance as an expected formula-behavior check), and whether topology mirror asymmetry predicts Trap Count (H2, an emergent hypothesis not prescribed by any single formula). H1 used Pearson correlation; H2 used Spearman rank correlation.

pattern-recognition (“Dimensional Structure of Personality Dynamics in the Icosa Formation System”). Evidence type: formula verification. This study generated N = 1,000 synthetic profiles using the Icosa framework (seed = 42) and tested one hypothesis: whether dynamics momentum predicts Coherence (H1). The test used Pearson correlation.

Under the evidence taxonomy, all three studies qualify as synthetic benchmarks showing simulated behavior and boundary conditions. The profiles are computationally generated with unconstrained random inputs. Distributional properties of these profiles may diverge from those produced by human respondents, particularly at extremes and in the base rates of different Coherence Bands and formation types. The studies establish properties of the computational model’s internal architecture; they do not and cannot speak to external criterion validity, diagnostic performance, or clinical utility.

Circularity Governance

The circularity landscape across these three studies is straightforward but not uniform.

formation-emergence returned a clean audit: no flags, no unresolved findings, zero allowed/disallowed overlaps. Grid completion, pair density, column completion, and row completion are structural inputs computed upstream of the Coherence formula, Domain Health, and Capacity Health. No shared-ancestry overlap was detected between predictors and outcomes. All 11 correlations can be interpreted as reflecting genuine propagation relationships through the computational pipeline rather than formula-sharing artifacts.

pattern-demographics was flagged for one governed overlap. Analysis H1-a1 (topology fulcrum health vs. Coherence) shares computational ancestors through cell_health_FM and cell_health_FR. The disposition is expected_formula_check: the overlap was retained as a representative topology-to-coherence formula-behavior check. As a result, the r = .82 result for H1 confirms implementation fidelity but cannot be upgraded to independent evidence of a structural relationship. The second hypothesis (H2, mirror asymmetry vs. Trap Count) was not flagged. Mirror asymmetry and Trap Count are computed through independent pathways, making H2 an emergent hypothesis.

pattern-recognition returned a clean audit. Dynamics momentum and Coherence are computed through algorithmically independent pathways — momentum tracks the velocity and consistency of state change while Coherence aggregates static deviations from capacity-specific targets. The r = .77 correlation reflects convergence between two distinct computational routes rather than shared-formula inflation.

The practical consequence: of the 14 total statistical tests across the three studies, 13 carry full evidential weight as either structural propagation tests (formation-emergence) or emergent-relationship tests (pattern-demographics H2, pattern-recognition H1). One (pattern-demographics H1) is a governed formula-behavior check whose large effect confirms implementation fidelity but cannot be cited as independent support for the topology layer’s structural validity.

Major Findings

Grid-Level Properties Propagate Faithfully to Formation Outputs

The formation-emergence study produced the most comprehensive result in this synthesis: all 11 confirmatory hypotheses were supported after Holm-Bonferroni correction, with zero null results and zero below-threshold findings. Effect sizes ranged from medium (r = -.50 for grid completion vs. resonance total) to large (r = .81 for pair density vs. Coherence), with the nine domain- and capacity-level tests clustering in the large-effect range (r = .67-.74).

The signal profile for this study is maximally clean: 11 reportable findings, zero below-practical-threshold, zero exploratory positives, zero FDR-only, zero raw-only, zero nulls, zero not-evaluable.

The single strongest effect was the pair density-Coherence correlation (r = .81, p < .001, 95% CI [.78, .83]), accounting for approximately 65% of Coherence variance. This establishes pair density — the proportion of co-activating center pairs in the grid — as the dominant structural predictor of personality integration within the model. Pair density’s association with Coherence substantially exceeds any individual row or column completion metric (r = .67-.74), indicating that pairwise interaction topology contributes to global integration above and beyond the simple count of centered Harmonies. What matters for system-level Coherence is not just how many centers are in their optimal state but how those centers interact across the grid.

Domain-level column completion indices predicted their corresponding Domain Health scores with large effects and narrow range (r = .68-.71 across Physical, Emotional, Mental, Relational, and Spiritual domains). Capacity-level row completion indices predicted their corresponding Capacity Health scores with slightly wider spread (r = .67 for Move to .74 for Open). The relative attenuation at Move Capacity is worth noting: Move encompasses expressive functions (Vitality, Passion, Agency, Voice, Service) that sit at the output end of the model’s processing cycle (Receive, Discern, Integrate, Express). Weaker propagation from row completion to Move Capacity Health may reflect the theoretical commitment that expressive centers depend on upstream capacities (Open, Focus, Bond) more than on intra-row completion alone. This interpretation, while consistent with the model’s architecture, remains speculative within synthetic data and would require mediation analysis to test directly.

The inverse relationship between grid completion and resonance total (r = -.50) deserves separate attention. A medium effect indicates that higher-completion profiles produce lower resonance values — the expected direction in a model where well-completed grids settle into stable, low-resonance formations while incomplete grids produce higher resonance as the system oscillates among unstable configurations. That the magnitude (compared to the large pair density-Coherence effect) is only moderate suggests that resonance total captures aspects of formation structure partially but not fully redundant with completion: resonance appears to index the stability of whatever formation emerges, while completion indexes the breadth of centered functioning that produced it.

Topology Metrics Track Both Prescribed Integration and Emergent Dysfunction

The pattern-demographics study tested the topology layer with two contrasting hypotheses, and the contrast between them is more informative than either result alone.

Topology fulcrum health correlated with Coherence at r = .82, p < .001 — a large effect exceeding its r = .50 practical threshold. However, this result operates under circularity governance as an expected formula-behavior check (disposition: expected_formula_check, shared ancestors: cell_health_FM and cell_health_FR). What it confirms is that the topology layer adds no distortion to the integration signal it inherits from center-level scores. The engine faithfully translates its geometric specification into computed outputs across 1,000 diverse profiles. This is implementation fidelity, not independent discovery.

The more consequential result is the emergent finding: topology mirror asymmetry correlated with Trap Count at r_s = .63, p < .001. This hypothesis was not flagged for circularity governance. Mirror asymmetry is a formation-level summary statistic comparing structural contributions across the profile’s topological halves; Trap Count operates at the individual-center level, identifying specific state configurations that lock centers into dysfunctional cycles. No single formula connects them. That r_s = .63 emerged across profiles spanning the full range of formations and Coherence bands indicates a structural regularity: as formation topology becomes more asymmetric, the conditions for trap emergence multiply.

The magnitude deserves scrutiny. At r_s = .63, approximately 40% of Trap Count variance is shared with formation asymmetry. This is large enough to suggest that formation topology captures meaningful structural information about dysfunction risk, but 60% of trap-count variance remains unexplained. Residual variance likely reflects center-specific state configurations that produce traps without requiring global asymmetry — a profile can be perfectly symmetric and still harbor traps if specific centers happen to occupy dysfunctional states. This interpretation aligns with the model’s design, where traps are defined at the center level while formations describe profile-level shape.

Signal profile for this study: 2 reportable, zero across all other categories.

Dynamics Momentum Is Tightly Coupled to Coherence

The pattern-recognition study tested a single hypothesis and confirmed it at r = .77, p < .001 — a large effect accounting for approximately 60% of shared variance between dynamics momentum and Coherence. The signal profile is minimal and clean: 1 reportable, zero across all other categories.

This result establishes that the engine’s momentum computation (tracking velocity and consistency of state change across the 20-center matrix) reliably corresponds to its integration metric (aggregating static deviations from capacity-specific targets). Two quantities defined through different mathematical operations on the same underlying state vector converge at r = .77, indicating the model encodes a coherent relationship between positional integration and directional progress. The system does not merely compute where a profile stands; it produces a momentum signal that tracks how well-organized that standing is.

The 40% of variance not shared between momentum and Coherence is equally informative. Momentum captures dynamic properties — the shape and direction of change — that Coherence alone does not reflect. Two profiles at the same Coherence level with different momentum scores represent structurally different situations: a Strained profile with high momentum is actively moving toward greater integration, while the same Coherence score with low momentum indicates stasis or oscillation. This distinction maps onto the formation system, where profiles at the same Coherence Band can occupy different formation categories depending on their dynamic properties.

Null and Below-Threshold Results

Across all three studies, every tested hypothesis was supported. The combined signal profile is: 14 reportable, 0 below-practical-threshold, 0 exploratory positives, 0 FDR-only, 0 raw-only, 0 nulls, 0 not-evaluable. There are no nulls to report.

This absence of nulls requires careful interpretation. In formula verification studies, a null result would signal implementation divergence — the engine failing to propagate grid-level information to a theoretically designated output. The consistent confirmation across 14 tests, covering all five domains, all four capacities, global Coherence, resonance total, topology metrics, and dynamics, indicates that the computational pipeline carries structural information faithfully to every designated target without leaking, attenuating, or inverting the signal.

However, the absence of nulls should not be mistaken for an absence of boundary conditions. The studies tested predictions that were theoretically well-motivated and computationally grounded. The informative questions were about magnitude, not direction. Where within the large-effect range does each propagation pathway land? Do some domains or capacities show attenuated effects? The Move Capacity’s relative attenuation (r = .67 vs. .74 for Open) is the closest the evidence comes to a stress point, and it falls well within the large-effect range.

The consistent ceiling effects in health outcomes across all nine Domain and Capacity Health variables (negative skew ranging from -2.06 to -2.52) are a distributional property of the d40 generator rather than a pipeline fidelity issue. Unconstrained random profiles tend to produce more high-health than low-health outcomes, compressing variance at the upper end. Whether the propagation patterns hold under more clinically realistic conditions — where Burdened and Severe formations are more prevalent — remains untested.

Architectural Implications

Pair Density as the Dominant Structural Driver

The single most consequential architectural finding across the three studies is that pair density accounts for roughly 65% of Coherence variance, establishing inter-center coupling as the dominant driver of personality integration within the model. This exceeds any individual row or column completion metric by a substantial margin. For the Centering Paths algorithm, the implication is direct: intervention steps that increase the number of co-activating center pairs should produce larger Coherence gains than steps that bring individual centers to centered status in isolation.

The Gateway system already partially embodies this logic — opening a Gateway unlocks multiple Traps and disrupts Basins simultaneously, affecting pairwise relationships rather than single centers. The pair-density finding suggests that the Centering Path optimizer could explicitly target pair-density increases as an intermediate objective, with the expectation that pair-density gains would translate to Coherence gains at the r = .81 conversion rate observed in synthetic data.

Topology as Verified Input to Downstream Computation

The combined topology results (fulcrum health implementation-verified at r = .82; mirror asymmetry emergently linked to Trap Count at r_s = .63) establish that topology metrics can be trusted as inputs to downstream computations, including Centering Path algorithms, with confidence that their behavior is internally consistent. The mirror-asymmetry finding adds a complementary optimization signal: profiles with high formation asymmetry accumulate more traps, which means that asymmetry reduction could serve as a proxy for trap disruption, potentially simplifying the search for effective intervention targets.

Momentum as Dynamic Companion to Static Coherence

The r = .77 momentum-Coherence coupling creates a built-in validation channel for path computation. A Centering Path that increases Coherence should also increase momentum; a divergence between the two signals would flag an inconsistency in the path-planning algorithm. Beyond validation, the 40% unshared variance justifies treating momentum as an independent information source for formation-level interpretation, not merely a derivative of the static Coherence snapshot.

Consistent Propagation Across Domains and Capacities

The near-uniform propagation pattern across all five domains (r = .68-.71) and all four capacities (r = .67-.74) indicates that the formation pipeline does not favor any particular domain or capacity over others. No domain or capacity “leaks” structural information, and none is disproportionately attenuated. This uniformity is an architecturally desirable property: it means formation classifications reflect balanced structural information rather than being driven by a subset of the grid.

Limitations of the Current Evidence Base

Synthetic data only. All three studies used N = 1,000 profiles with the same fixed seed (seed = 42), meaning they may analyze the same 1,000 profiles through different analytical lenses rather than three independent samples of 3,000 profiles in total. All profiles were computationally generated within the Icosa framework with unconstrained random inputs (formation-emergence used the d40 engine specifically; the generation pipeline for the other two studies is not separately documented). Human response data introduces patterns — response styles, acquiescence, social desirability, incomplete engagement — that synthetic generation does not model. Findings describe properties of the computational model, not properties of human personality.

Single-seed design. All three studies used the same fixed seed (seed = 42). While this ensures reproducibility, the stability of effect-size estimates across different seeds has not been established. The r = .81 pair density-Coherence correlation could be tighter or looser in other draws from the generative distribution.

Unconstrained profile distribution. The d40 generator’s unconstrained random sampling produces ceiling effects in health outcomes and may over-represent profile configurations that are rare or implausible in human populations. Replication with persona-constrained profiles (which shift the distribution toward lower Coherence bands) would test whether propagation patterns hold under more clinically realistic distributional conditions.

No external criteria. No external validator, clinical outcome, or human-judgment measure has been tested against any formation metric. Internal consistency, however thoroughly documented, is a necessary but not sufficient condition for the formation system’s utility in practice.

Narrow hypothesis coverage for topology and dynamics. The topology study tested only fulcrum health and mirror asymmetry, leaving core-periphery ratios and repeller dynamics unexamined. The dynamics study tested only momentum, leaving cascade, compensation, cycling, inertia, patterning, and trajectory unanalyzed. The factorial structure of the full seven-metric dynamics system remains unknown.

No band-stratified analysis. All correlations are global, computed across the full range of Coherence bands. Whether the propagation patterns are uniform across severity levels or concentrate in particular bands (e.g., whether the mirror-asymmetry-to-trap-count link strengthens in the Strained and Burdened ranges) has not been investigated.

No dyadic extension. All studies examined individual profiles only. Whether grid-level propagation patterns, topology-behavior relationships, and momentum-Coherence coupling hold, strengthen, or weaken in dyadic profiles has not been tested.

Linearity assumption. Pearson correlations assume linear relationships. The non-normality observed in several health variables, while mitigated by the large sample sizes, raises the possibility that monotonic but nonlinear relationships could alter effect-size estimates. Spearman rank correlations were used for one analysis (mirror asymmetry vs. Trap Count) but not applied systematically across the other tests.

Research Priorities

The following priorities are ordered by their expected contribution to advancing the formations evidence base from its current synthetic-verification state toward external validation readiness.

Priority 1: Multi-seed stability analysis. Repeat the formation-emergence study’s 11 tests across 100 random seeds of N = 1,000 each to establish confidence intervals for all effect-size estimates and determine whether the magnitude hierarchy (pair density > row/column completion > grid completion) is stable across draws from the synthetic profile space. This is the lowest-cost, highest-value next step: it converts single-point estimates into distributions.

Priority 2: Band-stratified replication. Stratify all correlations by Coherence Band (Thriving, Steady, Strained, Burdened, Severe) to test whether propagation patterns are uniform across severity levels. The mirror-asymmetry-to-trap-count correlation (r_s = .63) is the highest-priority candidate for stratification: if it concentrates in the Strained and Burdened ranges where traps are most prevalent, asymmetry is a severity-dependent phenomenon with different implications for path optimization than if it is uniform.

Priority 3: Persona-constrained replication. Repeat the benchmarks using persona-constrained profiles that shift the distribution toward lower Coherence bands and more clinically realistic configurations. The ceiling effects observed in health outcomes compress variance in ways that could either attenuate or inflate correlations depending on where human-like data falls in the distribution. Persona-constrained profiles provide the closest synthetic approximation to clinical populations.

Priority 4: Complete topology and dynamics verification. Extend the pattern-demographics protocol to core-periphery ratio and repeller count, completing the topology layer verification. Conduct a principal component analysis of all seven dynamics metrics (cascade, compensation, cycling, inertia, momentum, patterning, trajectory) to establish the dimensional structure of the dynamics system and determine how much independent information each metric contributes beyond what momentum alone captures.

Priority 5: Mediation analysis for pair density. Test whether pair density mediates the relationship between grid completion and Coherence, formalizing the distinction between breadth of centered functioning and inter-center coupling as contributors to global integration. This would resolve whether pair density is a partial proxy for completion or an independent structural pathway.

Priority 6: Gateway perturbation studies. Selectively degrade individual Gateway states and measure the downstream impact on both pair density and formation classification, testing whether the Gateway system modulates formation emergence as the model’s geometry predicts. This extends the formation-emergence findings from observational correlation to causal-structure probing within the synthetic framework.

Priority 7: Cross-engine comparison. Replicate the benchmark suite using profiles generated by the c135 and p180 engines to determine whether propagation patterns are stable across assessment engines or vary with engine-specific distributional properties. Differences across engines would indicate engine-dependent artifacts; consistency would strengthen the claim that the patterns reflect stable properties of the formation architecture.

Priority 8: Human-respondent replication. Once human-respondent profiles are available in sufficient numbers (minimum N = 200, ideally N = 500), replicate the internal consistency benchmarks to determine whether the computational model’s formation architecture holds under the distributional constraints of real assessment data. This is the critical bridge from synthetic evidence to operational evidence and cannot be bypassed or approximated.

Conclusion

The Icosa model’s formation system passes a rigorous internal-consistency benchmark across three studies and 14 statistical tests. Grid-level structural properties — pair density, column completion, row completion, grid completion — propagate faithfully to every theoretically designated output across all four capacities, all five domains, and the global Coherence metric without exception. Topology metrics track both prescribed integration relationships and emergent dysfunction markers. Dynamics momentum couples tightly to Coherence through an algorithmically independent pathway.

The single most consequential finding is that pair density accounts for roughly 65% of Coherence variance (r = .81), establishing inter-center coupling as the dominant structural driver of personality integration within the model. This finding, combined with the emergence of a formation-topology-to-dysfunction link (mirror asymmetry vs. Trap Count, r_s = .63) through computationally independent pathways, indicates that the formation system captures structural information beyond what center-level metrics alone provide.

What this evidence establishes is architectural fidelity and internal coherence within the synthetic benchmark. What it does not establish is external validity, clinical utility, cross-engine stability, or human-respondent generalizability. The formation system behaves as its mathematics specifies; whether those mathematics describe something true about human personality is a separate question that awaits empirical data.

The most productive immediate next steps are low-cost extensions of the existing synthetic evidence: multi-seed stability analysis to convert point estimates into distributions, band-stratified replication to test whether propagation patterns vary by severity, and persona-constrained benchmarks to approximate clinically realistic profile distributions. These require no new data collection and would substantially sharpen the current findings before human-respondent data becomes available. The longer-term priorities — completing topology and dynamics verification, gateway perturbation studies, cross-engine comparison, and human-respondent replication — trace the path from verified internal architecture to externally validated assessment infrastructure.