Pattern Consistency Under the Hood: 10 Personas, Three Layers Each

Three Systems, One Answer

Three things were built separately. The Icosa engine’s pattern-matching logic — the rules that assign capacity patterns, domain patterns, and formations based on raw scores. The clinical theory behind each diagnostic category — what anxiety looks like structurally, what depression does to the grid, how trauma signatures differ from burnout. And the persona definitions themselves — hand-crafted input vectors designed to embody each clinical picture.

None of these systems consult each other. The engine doesn’t know it’s looking at a “depression persona.” The clinical theory doesn’t tell the engine which pattern name to assign. The persona definitions don’t contain any expected output — they’re just numbers fed into a pipeline.

When all three agree on the same structural picture, that convergence isn’t a design artifact. It’s evidence that the geometry is capturing something real about how these clinical presentations organize themselves. Icosa Atlas ran 10 clinical personas through the full computation pipeline, and the results tell a specific story about where structural geometry and clinical observation meet.

What the Test Actually Checks

Each persona enters the engine as a set of capacity targets and domain targets — nothing more. The engine computes three independent outputs: a capacity pattern (how the four capacities relate to each other), a domain pattern (how the five domains distribute), and a formation (the overall structural shape of the grid). It also computes Coherence with a corresponding band.

The test asks a straightforward question: does what the engine sees match what clinical theory predicts? A persona built to model generalized anxiety should produce patterns consistent with hypervigilance and cognitive overactivation. A persona built to model severe depression should produce patterns consistent with system-wide shutdown. If it does, the engine is finding the same structure that clinicians observe. If it doesn’t, something is wrong — either in the engine’s logic, the clinical theory, or the persona definition.

All three layers have to agree. Getting one right could be coincidence. Getting two right narrows the odds. Getting all three right, across 10 different clinical presentations, means the geometry is tracking clinical reality with consistency.

The 10 Personas, by Cluster

Anxiety and shame pull the system in related but distinct directions. The gad_cognitive persona — cognitive anxiety with hypervigilant activation — produces a Fixating capacity pattern and a Storming domain pattern, landing in a Holding formation with Strained coherence. That’s a system locked into scanning mode: Focus pushed over, Open pulled under, the mental domain running hot while everything else stays flat. The shame_active persona shares the withdrawal signature but goes quieter. Its capacity pattern is Buried, its domain pattern Nullified, and its formation Narrowed — Burdened coherence. Where anxiety grips and scans, shame closes the shutters. Both are constrictive, but shame’s version is more total, and the engine sees the difference without being told to look for it.

Mood collapse and late-stage burnout converge in a way that’s clinically precise and geometrically unsurprising. The mdd_severe persona produces Total Shutdown for its capacity pattern, Nullified for its domain pattern, and Contracted for its formation — Severe coherence, deep in the band’s lowest range. Every capacity is under. Every domain is negative. The burnout_late persona arrives at the same structural destination: Total Shutdown, Nullified, Contracted, also with Severe coherence. The engine can’t distinguish these two at the pattern level, which is itself a finding — late burnout and severe depression present identical structural wreckage. The difference lives in the pathway that got them there, not in the snapshot the engine captures.

Personality and relational instability look nothing like shutdown. The npd_rage persona — narcissistic activation with Voice and Bond over — produces a Fanatical capacity pattern, no domain pattern (the domains are too evenly elevated to name), and an Activated formation with Strained coherence. This is a system running hot in a specific direction: Voice and Bond both pushed over, a relational surge of explosive output without structural collapse. The bpd_emotional persona is wilder. Its capacity pattern is Spiraling, domain pattern Saturated, formation Peaking — Burdened coherence. Where narcissistic rage has a vector, emotional instability has turbulence. Open over, Bond over, Voice over, Focus pulled under. The emotional domain scores 2.4, dwarfing everything else. The engine names it accurately: a system saturated with emotional intensity and spiking rather than holding any stable shape.

Complex trauma doesn’t look like either of those clusters. The cptsd_full persona produces a Vacant capacity pattern, Nullified domain pattern, and Contracted formation — Severe coherence. What makes it distinct from the depression/burnout signatures is the paradox in its capacity states: Open is centered while Focus, Bond, and Voice are all under. The system is receiving input but can’t organize, connect, or act on it. That’s the clinical hallmark of complex trauma — not a uniform shutdown, but a system stuck between contradictory states — and the pattern name, Vacant, captures exactly that quality.

Explosive anger is activation without the narcissistic structure. The anger_explosive persona yields a Flailing capacity pattern, a Wretched domain pattern, and a Peaking formation at Burdened coherence. Voice is pushed to 2.4, Open is over, and the emotional domain scores 2.0. But unlike the npd_rage persona, there’s no centered foundation holding the activation in place. The system is spiking in multiple directions with nothing organizing the output — which is why the engine calls it Flailing rather than Fanatical.

Healthy baselines confirm the geometry works in both directions. The secure_continuous persona — balanced across all capacities — produces no capacity pattern and no domain pattern (all four capacities over but uniformly distributed, so no asymmetric distortion registers), with a Stirring formation and Steady coherence. Zero traps, zero basins. The integrated_contemplative persona has more texture: a Melding capacity pattern, a Cataclysmic domain pattern, and an Ascending formation with Steady coherence. This persona was designed with deliberate asymmetry — elevated Open and Bond, strong Mental and Spiritual domains — and the engine reads that asymmetry as structured growth rather than distortion. The formation name, Ascending, reflects a system moving upward with intention rather than sitting in comfortable equilibrium.

Where Convergence Is Tightest

The shutdown cluster — mdd_severe, burnout_late, and cptsd_full — shows the strongest agreement between theory and engine output. Depression and burnout both predict full under-cascade, and the engine delivers Total Shutdown / Nullified / Contracted for both. Complex trauma predicts a different kind of collapse, and the engine distinguishes it with Vacant instead of Total Shutdown. These are the most geometrically predictable presentations because their clinical signatures translate directly into capacity and domain positions: everything drops, everything goes under, and the pattern names follow from that arithmetic.

The activation cluster — npd_rage, bpd_emotional, anger_explosive — shows strong convergence at the formation level (all produce either Activated or Peaking) while diverging meaningfully at the pattern level. That divergence is the interesting part: the engine isn’t just detecting “something is over-activated” but distinguishing between Fanatical (directed), Spiraling (oscillating), and Flailing (uncontrolled). Those distinctions map cleanly onto the clinical differences between narcissistic rage, emotional instability, and explosive anger.

What the Coherence Bands Add

The coherence bands provide a severity check that runs independently of the pattern names. For the three Severe-level personas — mdd_severe, cptsd_full, and burnout_late — the engine’s coherence band matches the expected band exactly: Severe across the board. These are personas designed to represent severe clinical presentations, and the engine classifies them accordingly without being told the target.

The healthy personas land where you’d predict. Secure_continuous lands in the Steady band (expected: Thriving), and integrated_contemplative also reaches Steady (expected: higher within the band). Both land somewhat below their expected bands, which reflects a real property of the engine’s coherence formula: any asymmetry, even healthy asymmetry, incurs a small structural cost. A perfectly balanced system scores higher than a richly textured one, even when the texture represents growth. Coherence measures integration, not development, and the persona data makes that tradeoff visible.

The middle-range personas show more spread. Bpd_emotional lands at Burdened against a Distressed expectation. Shame_active hits Burdened against Distressed. Anger_explosive scores Burdened against a Strained expectation. Across this group, the engine consistently rates presentations as more structurally compromised than clinical theory alone would suggest — which may say something about how structural geometry amplifies severity when multiple systems are affected simultaneously.

Ten personas, three layers each, built from three independent systems. When the results converge — Total Shutdown for depression, Vacant for complex trauma, Spiraling for emotional instability, Flailing for explosive anger, no pattern at all for secure attachment — that’s the geometry finding what clinicians observe through a completely different path.