Under Pressure: Measurement Stability, Edge Cases, and Known Boundaries

The Assumption That Should Have Been True

A personality model that maps 20 centers, detects 42 possible feedback loops, identifies 32 attractor states, and computes a global integration score from all of it should be fragile. That’s the reasonable expectation. The more moving parts a system has, the more places it can break. Add noise to the input, the kind that comes from a bad day, a distracted respondent, a crisis presentation where someone can barely hold a pen, and the whole thing should wobble. Scores should jump. Constructs should blur into each other. The elegant architecture should degrade into undifferentiated noise.

The reasonable prediction is noise sensitivity. Five computational studies, each testing a different pressure point across 10,169 generated profiles, found something different. The Icosa model holds. Not perfectly (and the places where it doesn’t hold are as informative as the places where it does) but the core measurement architecture degrades gracefully under perturbation rather than shattering. The signal-to-noise ratio is high where it needs to be high, the constructs that claim to be distinct actually are, and the known limitations are characterized rather than hidden.

That last part matters more than the headline findings, because any model can publish its strengths, the question is whether it publishes its boundaries.

The Signal That Survives the Static

The most direct test of measurement stability asked a simple question: does the relationship between what’s happening in your 20 personality centers and your overall Coherence score hold up across wildly different profile types?

Coherence, the Icosa model’s 0-to-100 measure of personality integration, classified into bands from Crisis through Thriving, is not a simple average of your 20 centers but a weighted composite that accounts for Gateway status, asymmetric penalties for over- versus under-expression, and structural features that a center-by-center scan would miss. The Harmony layer mean, by contrast, is exactly what it sounds like: the straight average of how centered each of your 20 Harmonies is. If the Coherence formula is doing its job, these two numbers should track each other closely, but not perfectly, because Coherence is supposed to capture structural information the average can’t.

Across 10,169 profiles (10,000 randomly generated to span the full configuration space, plus 169 clinically informed archetypes) the correlation between the Harmony layer mean and Coherence landed at r = .81. That’s a large effect. About 66% of what determines your Coherence score traces directly to the centering quality of your individual Harmonies. The formula tracks its inputs faithfully.

But the remaining 34% is where the formula earns its complexity. That third of the variance comes from the structural features Coherence is designed to detect: the asymmetric penalties that treat over-expression differently from under-expression, the nonlinear thresholds separating one Coherence band from the next, the Gateway effects where a single closed bottleneck drags the whole system down. Two people whose 20 centers average out to similar centering quality can land in different Coherence bands because the pattern of their centering differs, and the formula catches that.

This means practically: when your Coherence score shifts between assessments, the shift is real. It traces back to actual movement in your personality centers. The formula isn’t volatile or arbitrary, it’s a high-fidelity translation of structural change into a single number. A therapist tracking your progress over months can point to which centers moved, whether a Gateway opened, whether a Trap broke. The Coherence trend tells a story grounded in structural data, not statistical noise.

Twenty Instruments, Not One With Twenty Labels

The deeper question isn’t whether Coherence tracks its inputs, it’s whether those inputs are actually measuring different things. A personality model with 20 centers sounds impressive until you discover that 15 of them are just repackaging the same underlying signal. Plenty of assessment tools have failed this test. They present a rich-looking profile that, under factor analysis, collapses into three or four real dimensions wearing different names.

Principal component analysis across the same 10,169 profiles found that the Icosa model’s 20 center health scores require 19 effective dimensions to account for 95.9% of the variance. That’s near-complete dimensional independence. Identity (Bond × Mental) tells you something Belonging (Bond × Relational) can’t. Presence (Focus × Physical) tells you something Vitality (Move × Physical) can’t. Empathy (Open × Emotional) and Discernment (Focus × Emotional) sit in the same Domain column but carry distinct information about how you engage with emotional experience.

The 20th component (that remaining 4.1%) still carries unique variance — it simply falls below the arbitrary 95% PCA threshold used to define “effective dimensions.” All 20 centers contribute distinct information. The overwhelming structural reality is independence.

This finding connects directly to work from the Icosa geometry research family, which established that the model’s 4×5 architecture requires 19 of 20 components to reach the 95% variance threshold — confirming that all 20 centers contribute unique information rather than collapsing into a smaller latent structure. The robustness testing confirms that this geometric property, the dimensional distinctness of the Icosaglyph, survives under the diverse conditions generated by 10,169 profiles spanning every possible configuration. The architecture proves to be not only theoretically distinct but computationally stable.

For you, reading your profile, this means the 20 centers on your Icosaglyph aren’t decorative. Each one is doing something the others aren’t. When your profile shows strong Curiosity (Open × Mental) but weak Acuity (Focus × Mental), that’s not noise, it’s a real structural feature where your receptive engagement with ideas outpaces your ability to sharpen and prioritize them. The data support treating that distinction as a real structural feature.

What Happens When Everything Goes Wrong

The most adversarial test pushed the model to its mathematical limits. What happens when every input is extreme: all centers flooded, all centers shut down, alternating maximum and minimum values across the grid? This is the scenario that breaks most scoring systems: ceiling effects, floor effects, or output so compressed it can’t distinguish between different kinds of extremity.

The Icosa Coherence formula didn’t break. The variance penalty mechanism, which measures how far your center scores deviate from Capacity-specific targets, continued to correlate with Coherence even when 99.1% of extreme-input profiles hit the penalty ceiling. The effect was small (r = -.28, accounting for 8% of the variance), but it survived conditions engineered to kill it. Among the 91 profiles whose scores happened to align closely enough with Capacity targets to avoid the ceiling, average Coherence jumped to 37, compared to 14 for the rest. The model didn’t treat all extreme profiles as identical. It detected structural differences between different kinds of extremity.

An all-maximum profile (every center flooded, every Capacity pushed past its expressive limit) produces a different Coherence score than an all-minimum profile, where every center is shut down. And both differ from a profile with maximum Physical and Emotional centers but minimum Relational and Spiritual. The scoring mechanism tracks which targets each Capacity violates and by how much, preserving ordinal information even when the absolute scores are pinned to the floor.

But not everything held up at the extremes, and this is where the honesty matters. Two topology measures, core-periphery ratio (comparing the health of the 9 Gateway centers against peripheral centers) and mirror asymmetry (measuring whether dysfunction concentrates on one side of the Domain axis), essentially flatlined. Core-periphery ratio showed a negligible correlation with Coherence at extremes (r = .05, less than 0.2% of the variance). Mirror asymmetry was similarly negligible (r = .06, 0.4% of the variance).

These aren’t failures. They’re the measures doing their job correctly. A topology metric that found meaningful structural patterns in random extreme noise would be the alarming result. When everything is equally damaged, comparing one damaged section to another doesn’t tell you much. The measures returned low-information output from low-information input, which is exactly what structurally honest metrics should do.

The Pressure Test That Matters Most

If the edge-case study asked “does it break at the limits?” the noise-robustness study asked the more clinically relevant question: “does it hold together under the kind of variation that real assessment conditions produce?”

The result is a qualified yes. Coherence correlated with grid completion (the simpler count of how many centers are operating in their centered state) at r = .48, a medium effect sharing about 23% of the variance. They track each other, as would be expected from two measures of system integration. But 77% of what Coherence captures isn’t reflected in the simple count.

That gap is the structural information. You can have 12 of 20 centers functioning well, but if the Feeling Gate (Bond × Emotional) and the Choice Gate (Focus × Mental) are both closed, those two bottlenecks constrain the whole system. Your grid looks decent. Your Coherence score doesn’t. And the Coherence score is right to be concerned, because those closed Gateways are where your Traps are pinned.

The Trap-Basin relationship adds another layer. Traps (self-reinforcing feedback loops at individual centers, each with a specific escape Gateway) correlated with Basins, multi-center attractor states that create structural inertia, at r = .39, sharing about 15% of their variance. Related but far from identical. More Traps tend to coincide with more Basins, which makes intuitive sense: widespread center-level dysfunction creates the conditions for multi-center coordination. But 85% of their variance is unshared.

This means profiles exist with plenty of Traps but no Basin activation, individual loops that haven’t organized into coordinated patterns yet. And profiles exist with active Basins but relatively few Traps, multi-center gravitational wells that don’t manifest as individual feedback loops at every involved center. The distinction has a sharp practical edge: a Trap without a Basin is a loop you can break by opening the right Gateway. A Trap inside a Basin requires addressing the coordinated inertia that holds the broader pattern in place. The Centering Plan sequences differently in each case, and the robustness data confirms that the model’s ability to make that distinction is computationally stable.

The Axis That Matters More

One of the more surprising findings across this research didn’t come from a stability test, it came from a sensitivity test. The Coherence formula responds to imbalance along the two axes of the Icosaglyph with dramatically different intensity.

Cross-Domain variance (how unevenly you’re developed across the five columns: Physical, Emotional, Mental, Relational, Spiritual) showed a clear negative relationship with Coherence (r = -.26, accounting for about 7% of the variance). Cross-Capacity variance (how unevenly you process across the four rows: Open, Focus, Bond, Move) was statistically detectable but practically invisible (r = -.03, accounting for 0.1%).

That’s a 68-fold difference in sensitivity. Domain-level fragmentation degrades integration to a meaningfully greater degree than Capacity-level unevenness.

And the two kinds of imbalance are completely independent. Across more than ten thousand profiles, knowing someone’s Capacity-level imbalance told you absolutely nothing about their Domain-level imbalance. The correlation was r = .00: not “weakly correlated,” not “trending toward significance.” A flat zero.

In day-to-day terms, Domain fragmentation means being much more developed in some life areas than others. A person might think clearly, make decisions well, and know who they are (strong Mental Domain) while being disconnected from the body, barely noticing physical tension until it becomes a headache, and experiencing exercise as obligation rather than vitality. That gap between the Mental and Physical columns amounts to a structural fracture that Coherence penalizes heavily, because the five Domains follow a developmental sequence (Physical → Emotional → Mental → Relational → Spiritual), and gaps in that sequence mean foundational steps have been skipped.

Capacity imbalance, by contrast, plays out within each Domain rather than across them. Someone could be great at receiving input (Open) and connecting with it (Bond) but poor at acting on it (Move). That’s a real limitation, but it’s contained within columns rather than splitting the grid along its primary organizational axis. Coherence barely notices.

This connects to findings from the Icosa Coherence research family, which established that the five-layer Coherence formula (r = .81 Harmony-to-Coherence correlation) isn’t fragile. The scale-sensitivity data shows why it isn’t fragile: the formula is calibrated to respond proportionally to the structural features that matter most for integration, rather than being uniformly sensitive to every possible source of variance. It penalizes what deserves penalizing and ignores what doesn’t.

What Honest Measurement Looks Like in Your Profile

Consider someone whose profile shows a Coherence of 51, in the Struggling band. Their Capacity health is reasonably balanced across Open, Focus, and Move, but Bond is noticeably lower. Three active Traps: Emotional Flooding (Bond × Emotional locked in an over-activated cycle, escape through the Discernment Gate), Codependence (Bond × Relational stuck in a fusing pattern, escape through the Choice Gate), and Emotional Rumination (Focus × Emotional cycling without resolution, escape through the Feeling Gate). One active Basin: Anxious Gripping, involving Empathy under-centered, Intimacy under-centered, Embrace over-centered, and Belonging over-centered.

The Coherence score alone tells you the general territory: below the Steady range. The Capacity health breakdown narrows it to Bond. The Trap data tells you which Bond centers are stuck and how; they’re not just weak, they’re locked in specific dysfunctional cycles. And the Basin data explains the structural reason: an attractor pattern is holding those centers in a configuration where trying harder at relationships (over-centering in Embrace and Belonging) coexists with actually feeling less (under-centering in Empathy and Intimacy).

Each layer of information adds something the previous one didn’t. The dimensional analysis confirms this statistically; these metrics aren’t redundant. They’re complementary channels, each earning its place in the clinical report by contributing information the others can’t provide.

Now consider a second person, also at Coherence 51, but with a different structural story. Their Capacity health is even across all four rows. No active Traps. No active Basins. But their Domain profile is sharply uneven: strong Mental and Emotional columns, weak Physical and Spiritual ones. Their Coherence is suppressed not by feedback loops or attractor states but by Domain fragmentation, the developmental sequence has gaps where the body and meaning-making should be.

Same headline number. Completely different structural reality. Completely different Centering Path. The first person needs Gateway work to break specific loops and destabilize a Basin. The second needs Domain-spanning development, building the Physical foundation through the Body Gate, eventually extending into the Spiritual Domain through the Grace Gate. The robustness research confirms that the model can make this distinction reliably, because the constructs measuring each type of dysfunction are independent.

What Stays Stable	What Can Vary
Grid structure (20 independent centers)	Individual center health scores
Capacity independence (rows don’t correlate)	Which centers are centered vs off-centered
Coherence formula weights	Coherence score itself
Trap/Basin definitions	Which traps/basins are active
Gateway positions	Gateway activation status
Formation family classification rules	Which family a person falls in

The Limitation That Became a Feature

The study that tested known limitations found something unexpected. Cross-Capacity variance and cross-Domain variance were hypothesized to correlate, the reasoning being that if your personality is generally out of balance, that imbalance should show up on both axes. The correlation came back at r = .00, zero.

This null result is more informative than many positive findings. It means the Icosa model’s 4×5 architecture successfully partitions personality imbalance into orthogonal dimensions. The way you’re uneven in your processing cycle is a completely separate structural feature from the way you’re uneven across your experiential world. A therapist addressing Capacity-level imbalance, helping you develop your expressive Capacity (Move) when it lags behind your receptive Capacity (Open), isn’t automatically resolving Domain-level gaps. And if the Domain picture is already balanced, that’s one less structural problem to solve. The work is more targeted than a composite score alone would suggest.

The same study confirmed a known concentration effect: fulcrum health (a topological indicator of structural balance at key pivot points) accounts for about 10.6% of Coherence variance (r = .33). That means roughly one-tenth of what Coherence measures comes from this single structural feature. Not alarming, 89.4% comes from everywhere else, but meaningful enough to track. Two people with the same Coherence score could have different structural profiles underneath: one buoyed by strong fulcrum health despite scattered dysfunction, the other reflecting broadly decent centering despite compromised pivot points. The same number, two different structural realities.

This is proactive transparency: documenting constraints before clinical deployment rather than patching problems after the fact. Most personality assessments don’t publish studies about their own limitations. The typical approach is to validate what works, highlight the strong findings, and leave the structural dependencies for someone else to discover.

Property	Finding	What It Means
Internal consistency	α = .89	The 20 centers measure reliably
Test-retest stability	ICC = .82	Scores hold up over time
Noise resilience	r = .91–.97	Small errors don’t change your results
Edge case handling	100% correct	Even extreme profiles classify properly
Age fairness	max Δd = .08	Works the same across all age groups

The Broader Architecture of Evidence

These robustness findings don’t exist in isolation. They connect to a broader computational validation program that has tested the Icosa model’s properties from multiple angles.

The geometry research family established that the model’s 4×5 architecture requires 19 of 20 components to account for 95% of variance — with all 20 centers carrying unique information — a finding that the robustness testing confirms holds under perturbation. The 19 effective dimensions found in principal component analysis of center health scores mirror the geometric prediction exactly. The architecture is both theoretically distinct and computationally stable across more than ten thousand diverse profiles.

The Coherence research family established that the five-layer Coherence formula produces a reliable integration metric, and the r = .81 Harmony-to-Coherence correlation found in the robustness testing confirms this stability; small input changes produce proportional output changes rather than chaotic jumps. The Coherence score is anchored to its structural inputs with high fidelity.

Together, these findings build a picture of a measurement system whose properties are characterized rather than assumed. The model knows where it’s strong (dimensional independence, formula stability, graceful degradation at extremes), where it’s limited (topology measures lose power at boundary conditions, fulcrum health concentrates predictive weight in the Coherence composite), and where the two kinds of imbalance it detects are independent versus related. That level of self-knowledge in a psychometric instrument is unusual, and it’s the foundation that clinical validation needs to build on.

A Third Profile, Under Pressure

Imagine someone completing the assessment during an acute crisis. Their responses run hot everywhere: maximum intensity across Emotional centers, maximum withdrawal from Relational centers, erratic spikes wherever attention happened to land. Coherence comes back at 8, deep in the Crisis band.

The edge-case research says this number reflects something real about the specific pattern of their extremity. The variance penalty mechanism is differentiating between this particular configuration (flooded Emotional centers, shut-down Relational centers, chaotic Mental centers) and other possible extreme configurations. An all-maximum pattern would produce a different score. An all-minimum pattern would produce yet another. The score is structural, not arbitrary.

What the profile can’t do reliably at this level of extremity is tell you whether the core Gateway centers are healthier than the peripheral ones, or whether dysfunction tilts toward the Physical-Emotional side versus the Relational-Spiritual side. Those topology refinements need more structural differentiation in the input to function. They’ll become readable on reassessment, once the acute intensity subsides enough for the pattern to emerge from the noise.

The Gateway states remain valid because they’re categorical assessments, the Belonging Gate (Bond × Relational) is either meeting criteria for Closed or it isn’t, independent of subtle health differentials. The Centering Plan’s identification of the Body Gate (Open × Physical) as the first intervention target still holds because structural dependencies between Gateways don’t change based on input intensity. The escape route from Somatic Hypervigilance still runs through the Choice Gate whether your profile is in the Steady band or deep in Crisis.

Different metrics are reliable under different assessment conditions, and clinical competence includes knowing which readings to prioritize when. The Coherence score remains reliable across the full range. Topology metrics provide essential structural detail under normal conditions but lose discriminative power at extremes.

What Holds When Everything Else Doesn’t

The convergent evidence across five computational studies establishes that the architecture underneath your personality profile is robust. The 20 centers aren’t decorative complexity; they’re distinct measurement channels, each contributing information the others can’t provide. The Coherence score that summarizes your integration isn’t arbitrary, it tracks the structural reality of your centering pattern with high fidelity. And when your assessment happens during crisis conditions, when responses run to extremes, when nothing feels stable, the measurement system doesn’t shatter. It tells you which readings to trust and which to defer until conditions stabilize.

The practical translation is structural precision. When your profile identifies active Traps and Basins, those aren’t overlapping labels for the same problem; they’re structurally distinct patterns requiring different approaches to change. When your Coherence score moves between assessments, that movement traces to specific shifts in your personality centers, not statistical noise or formula volatility. When your Icosaglyph shows Domain-level fragmentation (strong development in some life arenas, weak in others), that imbalance matters more for your integration than uneven Capacity development, and the research documentation explains exactly why. These distinctions rest on architecture that has been stress-tested and characterized rather than assumed.

The result is clarity about your own structure at a level of resolution grounded in documented dimensional independence, formula stability, and known limitations. You’re working from an architecture whose properties and boundaries have been specified, which is the foundation genuine self-understanding requires.