Holistic System Rotation

I. The Process

I do something that I've never seen described as a unified cognitive phenomenon. I've called it "holistic system rotation" for at least five years, probably closer to ten — the term came easily because it described what the process felt like, even if the underlying mechanism was harder to articulate. It doesn't happen in a flash. It happens over hours and days of sustained engagement with a system — legislation, market structure, hardware architecture, organizational design, philosophical framework. The process:

I start wherever curiosity pulls hardest. Some part of the system is interesting — a constraint that seems to bind oddly, an interaction that doesn't add up, a subsystem whose internal logic implies something about a neighboring subsystem. I follow that thread. Along the way, I'm building internal representations — connections between parts, constraints, failure modes, time horizons, actor incentives. Each new connection opens new threads to follow. I think I follow whichever thread offers the most compression progress — Schmidhuber's term for the intrinsic reward signal generated by finding deeper structure that simplifies the representation of the data. The feeling of "this is becoming simpler, I'm seeing deeper structure" is the conscious translation of that reward signal.

The model densifies faster than the linear input of new information, because the mind continuously computes implied relationships. If clause A constrains agency B, and agency B controls pricing incentive C, then a connection between A and C exists — without having to observe it directly. This is transitive closure operating in the background: the cognitive graph fills in indirect links automatically, making the model denser than the data.

This continues for hours, sometimes days. I can't introspect whether the resulting model is held "simultaneously" in any meaningful sense, or whether I've simply built enough connections that traversal between distant parts becomes very fast. The emergent effect is the same: at some point, the model is rich enough that structural patterns fall out of it. Perverse incentives. Mechanism errors. Unintended interactions between subsystems that nobody examined together. Design flaws invisible from any partial view. These findings aren't procedurally derived — they emerge from the accumulated density of the internal model.

Then I move to the next system and the process repeats. The domain doesn't matter. Governance, semiconductor physics, philosophy, financial contracts — the substrate changes, the process doesn't.

Three examples of what this produces:

Governance machinery. Hours of engagement with democratic governance: electoral incentives, bureaucratic mesa-optimization, feedback latency between decisions and consequences, nine capital stocks that the budget doesn't measure, the language layer that hides agency behind passive voice. The structural finding: no closed control loop exists. Nobody measures whether laws produce their intended effects. Nobody tests how laws from different ministries interact. Nobody audits which capital stocks a policy depletes. The governance machine is an open loop — decisions flow out, but consequences never flow back to the decision-maker. This is nearly universal across Western democracies and is the core finding of Mekanismirealismi: mechanisms are the right level of intervention, the level where outcomes are actually determined, and it is the level that is completely missing from how governance works.

Bitcoin mining hardware. Days with ASIC manufacturing variance, electricity spot markets, cooling system thermodynamics, difficulty adjustment algorithms, and financial contract structures. The structural finding: inter-chip performance variance within the same production batch creates an exploitable optimization surface that almost no one in the industry maps, because hardware engineers don't think about financial contracts and traders don't think about semiconductor physics.

The Aliveness framework. Years across thermodynamics, evolutionary biology, game theory, information theory, AI alignment, political philosophy, institutional design. Following connections between physical constraints and institutional outcomes until the model was dense enough. The structural finding: a 250-year-old philosophical gap (Hume's is-ought problem) has an answer derivable from thermodynamic survival constraints. Nobody found it because the question spans six academic departments and no department owns the whole question.

II. Related Concepts

There are several related and distinct concepts.

"Systems thinking" (Meadows, Senge) advocates recognizing interconnections and feedback loops. Useful label. Systems thinking is a philosophy — "think about the whole system." HSR is a cognitive process — actually building a model dense enough that systemic patterns emerge from it. The difference is like "think about all the chess moves" versus being a grandmaster who evaluates fifteen moves deep in three seconds.

"Mental rotation" in cognitive science refers specifically to imagined spatial manipulation of three-dimensional objects. HSR operates on abstract systems — the rotation is through conceptual dimensions (subsystem interactions, temporal horizons, failure modes), in idea-space rather than Euclidean space.

"Perspective-taking" in social cognition means understanding another person's emotional or cognitive state. HSR includes actor perspectives, but the core process is about the rational incentive structure each node in the system faces — what will they actually do, given the constraints they operate under. The municipality's fiscal incentive to sanction welfare recipients onto state-funded programs is a mechanism question, not an empathy question.

"Integrative thinking" (Roger Martin) comes closest. Martin describes holding the whole problem in mind while working on the parts, resolving tensions creatively rather than through tradeoffs. Integrative thinking is primarily a management framework targeting business decisions. HSR is domain-independent and operates on the system's full internal architecture, including dimensions that have nothing to do with decisions.

Charlie Munger's "latticework of mental models" is sequential. Munger takes a problem and views it through the lens of microeconomics, then psychology, then biology. One lens at a time. HSR builds a model dense enough that all lenses are connected — the structural finding emerges from the collision between lenses, which sequential application misses.

III. The Fragments

The academic literature describes every component of this process — separately.

Synoptic judgment (policy theory, Dror/Etzioni contra Lindblom) is the closest terminological match. Mink defined it precisely: the ability to view chronologically, spatially, or conceptually separated events as constituent elements of a single unified whole — where "the whole is prior to its parts and the very condition of their existence and identity." In formal terms (§IV below), exercising synoptic judgment is querying a fully percolated conceptual graph. The systems theory literature treats it as a theoretical ideal — an "impossible dream of comprehensive rationality" that normal human cognition cannot achieve. Institutions therefore default to incrementalism: trial-and-error patches rather than structural redesign. Synoptic judgment names what HSR achieves but declares it impossible rather than studying how some people actually do it.

Eidetic variation (Husserl, phenomenology) maps onto the probing phase. Husserl described a rigorous method: mentally manipulating a concept — exercising freedom to alter one part while enforcing closure on the rest — to discover the invariant structure that persists across all variations. When a feature is altered and the system collapses, that feature is load-bearing. When it persists, it's incidental. HSR applies eidetic variation at the scale of entire socio-technical systems.

Destruction and creation (John Boyd) describes a dialectic engine for building and breaking mental models. Boyd's two-phase cycle: shatter existing mental models into unassociated components (destructive deduction), then discover new connecting threads and synthesize a restructured whole (creative induction). Boyd derived this from Godel, Heisenberg, and the second law of thermodynamics — any closed mental model will eventually fail. The fix is continuous destruction and reconstruction.

Churchman's sweep-in (operations research) describes continuously expanding the boundaries of analysis to include more variables, perspectives, and interconnections. "The systems approach begins when first you see the world through the eyes of another." The sweep-in process gathers comprehensiveness over time.

Coup d'oeil (Clausewitz, military strategy) — the commander's instantaneous grasp of the entire battlefield: terrain, forces, logistics, enemy intentions, temporal dynamics, all at once. Clausewitz described it as a rare gift. Modern military doctrine calls the operational level between tactics and strategy the hardest to master, because it requires holding the whole theater of war in mind while managing parts. Multi-domain operations (air, land, sea, cyber, space) demand the same simultaneous grasp across qualitatively different domains.

Fingerspitzengefuhl — the intuitive mastery that emerges after engaging with a system so thoroughly that its constraints become felt rather than computed. The phenomenological result of extensive HSR, rather than the process itself.

IV. The Topology of Insight

The informal description — "at some point, patterns fall out of the model" — has a formal counterpart in network science. A mental model is a conceptual graph: nodes (entities, constraints, variables) connected by edges (causal relationships, logical implications, operational dependencies). As sustained engagement adds edges — both from empirical observation and from transitive closure computed in the background — the graph densifies. The dynamics of this densification follow well-studied mathematical patterns.

Compression progress as the engine. Schmidhuber's theory: the intrinsic reward driving curiosity is the first derivative of subjective data compressibility over time. The steepness of the learning curve, not the final state, generates the dopaminergic reward. During sustained model-building, this means following whichever thread offers the steepest compression gradient — the next connection that will collapse the most redundancy, eliminate the most contradictions, reveal the deepest generative rule. If a model is built on surface correlations rather than causal mechanisms, it accumulates exceptions — anomalies that expand the model's description length. Causal models don't accumulate exceptions because they encode the mechanisms that produce the variations. Compression progress therefore mechanically forces the model toward reality alignment: superficial patterns get discarded, true generative structures get kept. This connects to Friston's Free Energy Principle — minimizing surprise through hierarchical predictive coding and active inference (purposefully seeking out the boundaries and stress points of the system to resolve uncertainty) — with Schmidhuber providing the algorithmic reward signal and Friston the biological process theory.

Small-world networks and shortcuts. Schilling's Small-World Network Model of Cognitive Insight (2005) formalizes what happens as the graph densifies. Most of the time, the model consists of densely connected local clusters with few connections between them. When a unifying principle or hidden constraint is discovered — a thread connecting two previously isolated clusters — it acts as a shortcut. The mathematical effect: a dramatic drop in average path length across the entire network. This triggers a cascade of connections, as newly proximate clusters reveal further links between them. The "aha moment" is the formation of a shortcut that bridges distant clusters.

Percolation threshold. The ultimate phase transition is described by percolation theory, originally developed in statistical physics. Below a critical graph density (the percolation threshold, p_c), the network consists of isolated clusters — understanding of parts with no traversable paths between them. Above p_c, the clusters merge into a "giant connected component" spanning the entire network. This is a genuine phase transition: a qualitative change in the topology that happens suddenly once enough edges exist.

In cognitive terms: below the threshold, you understand subsystems but can't see how perturbations in one cascade into another. Above it, there is a traversable path from any node to any other. Systemic failure modes, perverse incentives, and second-order effects become visible because the semantic path between cause and distant effect is now connected. The structural insight doesn't emerge gradually — it crystallizes at the threshold.

Three phases:

The phenomenology of simultaneity. When the network is fully percolated, neural traversal time between any two nodes approaches zero. The temporal divisions between representations collapse. You can't perceive an isolated part without automatically perceiving its role in the whole — the activation of one node instantly propagates through the dense graph. This is why the result feels like holding the entire system at once, even though focal attention is serial: the tacit background has become so densely connected that traversal is instantaneous. The phenomenology of simultaneity is the phenomenology of high graph density.

This maps to classic accounts of mathematical discovery. Poincare described four stages: preparation (sustained conscious engagement), incubation (unconscious processing — background transitive closure), illumination (sudden insight — the percolation threshold breached), and verification (testing the emergent model against reality). Hadamard surveyed mathematicians and found the same pattern. The formal model explains why: sustained engagement builds the graph, background computation densifies it, and at the critical threshold, the giant connected component crystallizes into conscious awareness.

V. Why It Has No Name

The irony is structural.

The components of HSR are studied in different academic departments. Synoptic judgment lives in policy theory. Eidetic variation lives in phenomenology. Compression progress lives in algorithmic information theory. Percolation theory lives in statistical physics. Hyper-systemizing lives in autism research. Monotropism lives in neurodiversity studies. Boyd's destruction-creation cycle lives in military strategy.

Each department studies its piece with high local rigor. No department studies the convergence. The process is unnamed because the map is severed — the knowledge production system that would need to name it is fragmented into the same silos that prevent anyone from seeing the whole.

The fact that HSR has no name is itself an instance of the problem that HSR solves. The cognitive process that synthesizes across silos is itself studied in silos. And since nobody owns the synthesis, nobody names it.

VI. The Cognitive Architecture

If the process is real, what enables it? The neuroscience and neurodiversity literature suggests a specific convergence.

Hyper-systemizing (Baron-Cohen): an intense drive to identify the lawful regularities governing input-operation-output mechanics of systems. In high-cognitive-capacity individuals ("twice-exceptional"), this manifests as the construction of massive, dynamically updating mental models that predict system behavior. The Systemizing Quotient (SQ) measures this drive. High SQ provides the motivation to build the model in the first place — the obsessive need to understand how the whole thing works.

Monotropism (Murray et al.): the mind as an interest system where attention is a finite resource pool. Neurotypical minds are "polytropic" — attention distributed broadly across many stimuli. Monotropic minds channel disproportionate resources into fewer interests, creating an attention tunnel of extraordinary depth. Monotropism provides the sustained resource allocation: near-total cognitive bandwidth directed at a single system model for hours or days, preventing the fragmentation that forces sequential thinkers to drop variables when complexity exceeds working memory. Without the capacity to remain deeply engaged in a single conceptual space for extended duration, the transitive closure cascades are interrupted by context-switching, and the graph density decays before percolation can occur.

Global gestalt perception: the neural capacity to perceive "the forest before the trees" — extracting a global invariant structure from local elements. Neuroimaging localizes this to the temporo-parietal junction (TPJ). Patients with TPJ damage can recognize individual objects perfectly but cannot perceive the global pattern they form (simultanagnosia). HSR requires the cognitive scaling of this capacity — perceiving the systemic pattern across an entire socio-technical system.

Latent neural coding: recent neuroscience identifies a dual system supporting cognitive flexibility — an active coding scheme that processes the currently prioritized dimension, and a latent code in the medial prefrontal cortex and hippocampus that maintains unprioritized dimensions without decay. During HSR, the dimensions not currently in focal attention are held in this latent state, instantly swappable into active processing without losing the integrity of the whole model.

The AuDHD hybrid engine

My own cognitive profile seems to be AuDHD — autism and ADHD combined. The clinical literature historically treated this combination as additive impairment — the deficits of both conditions stacked. But recent research supports a different view: AuDHD is a unique cognitive phenotype where the interaction between traits is epistatic, producing emergent properties that neither condition possesses alone.

From a predictive coding perspective, the two conditions operate as competing control policies sharing one neural architecture. The autistic policy dials up precision: tight error bars, analytical depth, low tolerance for unexplained variance, the hyper-systemizing drive to map every input-operation-output regularity. The ADHD policy dials up gain: novelty-seeking, high idea velocity, salience-driven exploration, the restless hunt for stimulation that reaches operating temperature only when something genuinely new appears.

When both policies are focused through a monotropic tunnel on a single complex system, they form a hybrid engine optimized for exactly the process described in this essay:

• ADHD provides the search power. Rapid, non-linear exploration across the topology of the system, jumping erratically to wherever the steepest compression-progress gradient lies. This looks chaotic from outside. Inside the tunnel, it's the most efficient exploration strategy — novelty-seeking as compression-progress hunting.
• Autism provides the model integrity. Every chaotic discovery gets logged into a precise, structurally sound relational graph. The hyper-systemizing drive refuses to let connections remain vague or approximate. The model that accumulates is rigorous because the autistic policy won't tolerate imprecision.
• Monotropism provides the sustained channel. Near-total cognitive bandwidth locked onto one system for hours or days. Without this, the ADHD search engine would scatter across multiple systems, and the autistic model-builder would never accumulate enough density for percolation.

The three together produce something none produces alone. ADHD without autism explores rapidly but builds nothing precise. Autism without ADHD builds precise models but explores too slowly to reach critical density. Both without monotropism fragment across too many targets. The convergence is the engine.

In one research session exploring this, a Gemini instance suggested that the process described here is an emergent feature of this specific architecture. I find this plausible but unverified. The AuDHD literature is young, and research on emergent cognitive phenotypes from trait interactions barely exists. There is also a practical bottleneck: as the internal model densifies, the cognitive load threatens to collapse the model before percolation. In practice this means aggressive externalization — notes, diagrams, whiteboards — to bypass the biological working-memory limit and preserve graph density while the search engine keeps adding edges.

VII. Domain-Independence

The process is substrate-independent because the underlying mathematics is. Percolation thresholds, small-world shortcuts, and average path length reductions are properties of network topology — they operate identically whether the nodes represent legal statutes, molecular structures, or logic gates. The brain doesn't use a specialized module for "financial market physics" and a separate one for "legislative physics." It uses a generalized capacity for graph construction, hierarchical predictive coding, and free energy minimization. fMRI evidence supports this: domain-general signals in the anterior prefrontal cortex predict structural evaluations — confidence, accuracy, relational mapping — across disparate cognitive tasks.

When engaging with a new domain, the model is built from scratch. The transferable skill is the process itself: initialize a sparse graph, lock into monotropic focus, hunt for compression progress, execute transitive closure, reach percolation. The content is domain-specific; the topology is universal. This is why the same process produces structural findings in semiconductor hardware, governance, and philosophy — and why the findings have the same character across domains: structural patterns invisible from any partial view, visible only from a fully connected model.

VIII. What It Produces

The output type is specific: structural findings invisible from any single angle.

Domain experts miss these findings because their expertise is scoped to one cluster of the system. Every ministry understands its own law. Every department understands its own domain. Each analysis is correct within its cluster. The finding that no closed control loop exists — that the governance machine produces outputs without ever measuring whether they work — requires a model dense enough to span electoral incentives, bureaucratic behavior, feedback architecture, and capital stock dynamics in a single connected component.

This is why the findings often feel obvious in retrospect. Once someone points out that nobody measures whether laws produce their intended effects, any policy analyst can see it. The difficulty was never the observation. The difficulty was building a model dense enough that the absence became visible as the binding constraint rather than one problem among many.

The compression paradox is relevant here. Bad compression maps surface attributes. HSR maps relational structure — the full causal architecture of the system. "Governance is an open loop with no mechanism-level feedback" is a relational finding. "The system is broken" is an attribute finding. The difference is the difference between a diagnosis and a complaint.

This essay collection is called The Rotations because every essay in it is the output of one pass of this process applied to a different substrate. The same method on thermodynamics. On institutional design. On epistemics. On AI alignment. On Finnish governance. The domain changes. The process doesn't.

IX. The Recursive Structure

Understanding HSR requires four levels of description. Each lives in a different academic silo. No silo holds more than one.

Cognitive components — what the brain brings. Hyper-systemizing (autism research): the drive to map input-operation-output regularities. Monotropism (neurodiversity): sustained channeling of near-total cognitive resources into a single model. Global gestalt perception (neuroscience): processing whole-field structure before local features. The AuDHD interaction (§VI) that produces something none of the traits produce alone.

Formal models of the process — how it works. Compression progress (algorithmic information theory): the reward signal that drives exploration. Percolation theory (statistical physics): the density threshold at which structural patterns emerge from the model. Transitive closure (graph theory): the mechanism by which the model densifies faster than linear input.

Partial descriptions of what happens — the process observed from outside. Eidetic variation (phenomenology): rotating an object through possible configurations. Boyd's destruction-creation (military strategy): breaking existing models and rebuilding them. Churchman's sweep-in (operations research): progressively widening the system boundary.

Names for the result — what it looks like when it's done. Synoptic judgment (policy theory): seeing the whole system at once. Coup d'oeil (military strategy): the commander's instant grasp of terrain. Both describe the phenomenology of a densely connected model — fast traversal experienced as simultaneity.

Four levels, eight fields, zero integration. The process that synthesizes across silos is itself fragmented across silos.

This essay is itself an HSR pass. The substrate is: the literature about the cognitive process that produces cross-silo synthesis. The finding is: this process has no name because the map is severed in exactly the way the process reveals.

I don't claim HSR is unique to me. Von Neumann, Boyd, Feynman demonstrably did versions of it. Clausewitz described it in battlefield commanders. Alexander described it in architects. The claim is only that the convergence has no name, produces a specific type of output, and that the reason it has no name is itself an instance of the problem it solves.

Related Reading

• The Severed Map — Why no one owns the whole question: the institutional failure that HSR overcomes
• The Compression Paradox — HSR builds relational structure; bad compression maps surface attributes