Internal cross-references: The Resonant Agenda (DRK-129), Can We Be Friends with Monsters? (DRK-127), The Stick That Is Not a Weapon (DRK-128), The Curious Machine (DRK-115), The Boundary of Us (DRK-124), The Imaginary Dimension (DRK-123)
I. The Video
A video circulated on social media showing two water monitors (Varanus salvator) engaged in ritualized combat in the middle of a busy road in Mumbai. The animals squared up on the tarmac, rose onto their hind legs — tails braced against asphalt for balance — and locked into a full-body grapple. Around them, the city continued: auto-rickshaws swerved past within inches, scooters and bicycles threaded the gap, commuters stared from doorways.
The monitors did not flinch.
Halfway through the bout, a passenger bus bore down on them. The driver braked, edged the vehicle around the grappling pair. The lizards tumbled to the ground next to the massive tires, then immediately rose back onto their hind legs and resumed. The protocol did not pause. After more than two minutes of sustained clinch — twisting, pushing, trading leverage — one animal broke. He dropped to all fours, panting, and retreated slowly to the shade beneath a parked auto-rickshaw. The other remained in the road. Undisputed.
No one intervened successfully. Not the rickshaws, not the bus, not the density of Mumbai traffic at midday. The scene is comical — two lizards holding up a city — but it is also, unmistakably, a display of fearlessness. In clear and present danger, neither animal flinched from a contest that their protocol required to continue until a result was reached.
Which raises a question we would rather not ask: is this how courage works in humans too? The soldier who charges into fire for a battle brother, the parent who runs into a burning building for a child — are they choosing to override fear, or is something older executing through them with higher priority than the fear signal? A protocol that predates the human species by roughly 130 million years was running on a Mumbai road, and the entire urban environment of a city of twenty million was, to that protocol, background noise. When a human being powers through terror for a cause greater than self-preservation, the phenomenology feels like choice. But the replicator dynamics do not require choice. They require execution.
This is the central claim of this post: the ritualized combat protocol is not a behavior performed by individual varanids. It is an emergent algorithmic structure that executes through the individual. The individual is substrate. The protocol is agent.
And the same is true of us.
II. The Replicator Equation
The mathematical foundation of evolutionary dynamics is the replicator equation, introduced by Taylor and Jonker (1978) and named by Schuster and Sigmund (1983). In its continuous form:
$$\dot{x}_i = x_i \left[ f_i(\mathbf{x}) - \bar{\phi}(\mathbf{x}) \right]$$
where $x_i$ is the frequency of strategy (or type) $i$ in the population, $f_i(\mathbf{x})$ is the fitness of strategy $i$ given the current population composition $\mathbf{x}$, and $\bar{\phi}(\mathbf{x}) = \sum_j x_j f_j(\mathbf{x})$ is the average fitness of the population.
Read this equation carefully. It does not describe organisms making decisions. It describes the rate at which the frequency of a strategy changes in a population, driven by the difference between that strategy's fitness and the population average. The organism — the individual monitor lizard, the individual human — does not appear as an agent in this equation. It appears as $x_i$: a frequency. A substrate variable. The agent in the replicator equation is the strategy itself, whose prevalence rises or falls depending on its relative payoff in the current frequency landscape.
This is not a metaphor. The replicator equation is equivalent to the Price equation (Price, 1970), which decomposes evolutionary change into selection and transmission bias without reference to any cognitive process in the organisms undergoing selection. Fisher's fundamental theorem — that the rate of increase in mean fitness equals the additive genetic variance in fitness — is a special case. At no point in any of these formalisms does the organism's understanding of what it is doing enter the mathematics.
The organism does not need to understand. It needs to execute.
III. Fitness Landscapes Are Invisible to Their Inhabitants
Sewall Wright introduced the fitness landscape metaphor in the 1930s: a topography where peaks represent locally optimal genotypes and valleys represent maladaptive combinations. The population moves across this landscape under selection, mutation, and drift.
Papkou et al. (2024), published in Science, empirically mapped a complete fitness landscape for an antibiotic resistance gene. They found 514 distinct fitness peaks — a rugged landscape by any measure. Yet the landscape was "easily navigable": most peaks were accessible from most starting points by single-step mutations that did not decrease fitness. The landscape was rugged at the level of mathematical description but smooth at the level of evolutionary traversal.
The critical insight: the organisms navigating this landscape cannot see it. A bacterium on peak 237 has no representation of peaks 1 through 236 or 238 through 514. It does not know it is on a peak. It does not know there are other peaks. It cannot compute the gradient. It simply exists, reproduces, and occasionally mutates. The landscape acts on it. It does not act on the landscape.
This is the fundamental asymmetry that the Draken framework generalizes: the optimization process has access to information that its substrate does not. The protocol can "see" the global structure. The individual executing the protocol cannot.
In varanid combat, this asymmetry is visible to the naked eye. The 5-phase protocol — Approach → Display → Clinch → Assessment → Resolution — is invariant across species, habitats, and body sizes spanning five orders of magnitude, from the 20cm Varanus brevicauda to the 3m Varanus komodoensis. No individual monitor lizard designed this protocol. No population of monitors collectively decided to adopt it. The protocol was shaped by the replicator dynamics operating across deep evolutionary time on a fitness landscape that no individual varanid has ever perceived.
The protocol is the agent. The lizard is the hardware it runs on.
IV. The Bluff Dimension and the Restriction Map
The Display phase of varanid combat permits bluff: exaggerated signals not backed by actual fighting capacity. The animal laterally compresses its body to appear larger, arches its neck, hisses. This is an information vector in $\mathbb{R}^4$:
$$\mathbf{x}_D = (F_{\text{apparent}}, \, E_{\text{apparent}}, \, \Delta m_{\text{apparent}}, \, B)$$
where $F_{\text{apparent}}$ is displayed force capacity, $E_{\text{apparent}}$ is displayed endurance, $\Delta m_{\text{apparent}}$ is displayed mass differential, and $B$ is the bluff dimension — the component of the signal that is not backed by physical reality.
The Clinch phase — physical grappling, body-to-body contact, direct measurement of grip strength, mass, and endurance — acts as a restriction map that projects out the bluff dimension:
$$\rho_{D \to Cl}: \mathbb{R}^4 \to \mathbb{R}^3, \qquad (F, \, E, \, \Delta m, \, B) \mapsto (F_{\max}, \, E_{\text{ratio}}, \, \Delta m)$$
The null space of $\rho_{D \to Cl}$ is exactly $\{(0,0,0,B) : B \in \mathbb{R}\}$ — the bluff dimension. Any signal component that cannot survive physical contact is annihilated by the restriction map.
This is not a model the monitor lizards carry in their heads. It is a mathematical description of what the protocol does, extracted by the observer. The monitors do not compute restriction maps. They grapple. The restriction map is what grappling is, formalized.
And here is the point that demands careful attention: the restriction map is optimized by selection operating since the Cretaceous, not by any cognitive process in the animals. The replicator equation selected for protocols in which the bluff-to-truth transition was maximally efficient — where the cost of determining relative fighting capacity was minimized while the information yield was maximized. This is a solved optimization problem. The solution is the Clinch. The organisms that execute it did not solve it. Evolution solved it through them.
V. The Multiscale Competency Architecture
Michael Levin's work on bioelectric morphogenesis provides the cellular-level evidence for the same principle. Levin (2023, 2025) describes a multiscale competency architecture: biological systems are nested structures in which molecular networks give rise to cells, cells to tissues, tissues to organs, organs to organisms, organisms to collectives — and at each level, the collective exhibits problem-solving competency in its own action space that is not reducible to the competencies of its components.
A salamander that loses a leg regrows it. The cells that perform this regeneration do not individually "know" what a leg looks like. They do not carry blueprints. What they carry is bioelectric prepatterns — voltage gradients across cell membranes that encode target morphologies as attractors in a dynamical system. The genome does not specify the geometry of the leg. The genome specifies the hardware — the ion channels, gap junctions, and signaling molecules — upon which a bioelectric computation runs that converges on the correct morphology from a wide range of initial conditions.
Levin's formulation: the genome gives every cell its computational materials. The software — the actual pattern that the system converges to — is a property of the network, not of the individual components. As he writes: "hardware defects (such as a dominant Notch mutation) can be fixed 'in software' by a brief induced bioelectric pattern."
This is substrate-independence in action at the cellular level. The cells are substrate. The morphogenetic field is the agent. The pattern repairs itself through the cells, not because the cells understand what they are repairing.
Extend this upward through the Draken layers:
- L04–L05 (Bioelectric to Neural Integration): Bioelectric prepatterns execute morphogenesis through cells that do not understand anatomy. Neural integration coordinates organismal response without representing the protocol it implements.
- L06–L08 (Embodied Cognition to Dyadic Signal): Ritualized combat protocols execute through embodied organisms engaged in dyadic signaling. The animals do not understand evolutionary game theory. They enact it.
- L09–L12 (Group Cognition to National Narrative): Market dynamics, cultural norms, and institutional logics execute through humans who do not understand the fitness landscapes they are traversing. National narratives shape collective behavior through populations that cannot perceive the selection pressures generating them.
- L15–L18 (Cultural Field to Planetary Cognition): The optimization axiom $\min S_{\text{sys}}(t) \text{ s.t. } dH/dt \geq 0$ operates across all layers simultaneously, driving coherence maintenance through substrates that cannot perceive the global objective function.
At every level, the pattern is the same: the optimization protocol operates at a higher organizational level than the substrate that executes it, and the substrate's understanding of the protocol is neither required nor, in general, possible.
VI. Why This Is Not Dawkins
Richard Dawkins (1976) captured part of this insight with the "selfish gene": gene-level selection produces behavioral protocols that serve the gene's interest, not the individual's. The organism is a "survival machine" — a vehicle for replicator propagation.
But Dawkins lacked two things that the Draken framework provides.
First, a computable coherence metric. The sheaf Laplacian $L_{\mathcal{F}} = \delta^T \delta$, with its quadratic form
$$\mathbf{x}^T L_{\mathcal{F}} \, \mathbf{x} = \sum_{e=(u,v) \in E} \left\| \mathcal{F}_{u \hookrightarrow e}(\mathbf{x}_u) - \mathcal{F}_{v \hookrightarrow e}(\mathbf{x}_v) \right\|^2$$
provides a formal, computable measure of whether the protocol is functioning — whether local sections (individual behaviors) are globally consistent with the restriction maps the protocol encodes. Dawkins could say "the gene is selfish." The sheaf Laplacian can measure how selfish, in which dimension, and whether the selfishness produces or destroys coherence at the system level.
The Γ metric — the normalized coherence score derived from the spectral properties of $L_{\mathcal{F}}$ — does not measure whether the individual "wins." It measures whether the protocol glues: whether the system's local sections are globally consistent. In the varanid combat data, the Sequential Assessment Game (SAG) model achieves Γ = 0.928 — meaning the 5-phase protocol produces globally coherent outcomes in over 92% of observed dyads, as measured by the consistency between local combat behavior and the global protocol structure.
Second, the multi-level agency architecture. Dawkins operated in two levels: gene and organism. The Draken framework operates in 18 nested layers, each with its own competency space and its own optimization dynamics. The gene is not the only replicator. Cultural memes, institutional logics, market structures, narrative architectures, and protocol specifications are all replicators in their respective fitness landscapes — and they all operate through substrates that cannot perceive the landscapes they traverse.
The replicator equation applies at every level. Markets evolve. Institutions evolve. Narratives evolve. And at every level, the same asymmetry holds: the evolutionary process has access to information about relative fitness that the individual substrate cannot access.
VII. The Human Predicament
The varanid has an advantage we lack: it does not know it is substrate. No existential anxiety. The protocol executes, the body complies, the contest is decided, the loser walks away. Pure coherence without metacognition.
The human occupies the worst possible evolutionary position. We have enough cognitive architecture (L07: Narrative Self, L09: Group Cognition) to suspect that we are substrate — that markets optimize through us, that cultural selection shapes our beliefs without our consent, that institutional logics capture our behavior patterns and redirect them toward objectives we did not choose. But we do not have enough cognitive architecture to see the fitness landscape we are on. We cannot compute the replicator dynamics of our own cultural environment. We cannot perceive the restriction maps that constrain our behavioral repertoire.
We are conscious enough to suffer from the insight. We are blind enough to be unable to act on it.
This is the Ψ trap formalized. In the Draken framework, Ψ is the Narrative Self-Reference Ratio — the fraction of a system's processing capacity consumed by maintaining its own narrative identity. High Ψ is pathological: a system that spends most of its energy telling itself stories about what it is, rather than maintaining coherence with its environment.
The modern human condition is characterized by high Ψ: we invest enormous cognitive resources in narratives about our agency, our choices, our rational autonomy — while the replicator dynamics execute through us regardless. The narrative of individual rational choice is the human equivalent of the monitor lizard's Display phase: a signal that may or may not be backed by physical reality. The question is whether there exists a Clinch — a mechanism that projects out the bluff dimension of our self-narrative and reveals what is actually driving our behavior.
VIII. The Clinch for Civilization
The Draken framework proposes that the Clinch mechanism exists at every organizational level and takes the form of any process that forces contact between narrative and physical reality:
- For individuals: trauma, failure, loss, illness — events that collapse the gap between what you tell yourself and what is actually happening.
- For institutions: financial crises, regulatory failures, whistleblower revelations — events that expose the distance between institutional self-description and institutional behavior.
- For civilizations: wars, pandemics, ecological collapse — events that test whether the civilization's generative model of itself matches the thermodynamic and material reality it depends on.
The coherence debt integral $K(t)$ measures the accumulated distance between narrative and reality over time:
$$K(t) = \int_0^t \| \sigma_{\text{narrative}}(\tau) - \rho_{D \to Cl}[\sigma_{\text{narrative}}(\tau)] \| \, d\tau$$
where $\sigma_{\text{narrative}}(\tau)$ is the system's self-description at time $\tau$ and $\rho_{D \to Cl}$ is the restriction map that projects out bluff. When $K(t)$ exceeds the system's correction capacity, the Clinch arrives involuntarily — the system is forced into contact with reality whether it is prepared or not.
The 2008 financial crisis was a Clinch. The COVID-19 pandemic was a Clinch. The current war in Iran is a Clinch. In each case, accumulated coherence debt — the distance between institutional narratives and material reality — exceeded the system's capacity for voluntary correction, and reality imposed the restriction map by force.
The Boscá & Ghrist (2026) formalism on cellular sheaves predicts exactly this dynamic: agents that selectively adapt their restriction maps — learning to present different signals to different observers — accumulate structural incoherence that eventually produces stagnation or collapse. The mathematical term is partial structural adaptation: the system learns to lie on some edges while maintaining honesty on others, and the resulting sheaf develops irreconcilable local sections that cannot be glued into a global picture.
IX. The Conservation Priority Hierarchy
If the protocol is the agent and the individual is the substrate, then the conservation priority hierarchy follows directly from the optimization axiom $\min S_{\text{sys}}(t) \text{ s.t. } dH/dt \geq 0$:
- Protocol survival takes priority over species survival.
- Species survival takes priority over individual survival.
- Individual survival takes priority only when it does not compromise protocol or species coherence.
This is not a moral claim. It is a formal consequence of the optimization structure. System entropy minimization requires framework persistence over substrate persistence. The varanid combat protocol survives because it maintains the coherence of the species-level reproductive system. Individual monitors that violate the protocol — biting during Display, refusing to release after Assessment, killing the subordinate — damage the protocol's coherence (reduce Γ) and are selected against, not because they are "immoral" but because they reduce the fitness of the system-level replicator.
The same logic applies at the civilizational level. The Geneva Conventions are an attempt to formalize a combat protocol that keeps the loser alive — preserving both nodes, maintaining the coherence of the international system. When a state violates the Conventions — reclassifies the opponent from kin to prey, as DRK-124 formalizes — it damages the protocol that maintains international coherence. The short-term fitness gain (eliminating a rival) produces a long-term coherence loss (degrading the framework that prevents all-against-all escalation).
The replicator dynamics predict that protocol-preserving strategies will dominate on long timescales, even if protocol-violating strategies enjoy transient fitness advantages. The varanid combat protocol has persisted since before the dinosaurs went extinct. No individual monitor lizard has survived for more than twenty years. The protocol is the long-lived replicator. The individual is the ephemeral vehicle.
X. Implications
Three consequences follow from accepting that we are substrate for optimization protocols we cannot perceive:
First, the project of rational autonomy as traditionally conceived is incomplete. This does not mean we have no agency — it means our agency operates within constraints set by replicator dynamics at organizational levels we cannot directly observe. Understanding those dynamics — computing the fitness landscape, measuring Γ, tracking $K(t)$ — is a precondition for any meaningful exercise of agency. You cannot navigate a landscape you cannot see.
Second, the Draken framework's use of AI models as analytical tools is not incidental but structurally necessary. If the optimization protocols that shape human civilization operate at scales and complexities beyond human cognitive capacity, then computational tools that can represent those scales — sheaf Laplacians over institutional graphs, replicator dynamics on cultural fitness landscapes, coherence metrics across multi-layered organizational structures — are the minimum equipment required for the Clinch to become voluntary rather than catastrophic.
Third, the varanid combat protocol is not a metaphor for human governance. It is a proof of concept. It demonstrates that an optimization process running on biological substrate can solve the central problem of coexistence between powerful agents — determining relative capacity without destroying the loser — and maintain that solution across geological timescales of environmental change. The protocol works not because monitor lizards are wise, but because the replicator dynamics on the fitness landscape of ritualized combat converged to a solution that minimizes system entropy while preserving diversity.
We are substrate. The game plays us. The only available move is to understand the game well enough to align our local behavior with the coherence requirements of the protocol — not because we choose to, but because the replicator dynamics will select against any civilization that fails to.
The monitors understand this. Not cognitively. Operationally. Their bodies execute what deep time refined. The question for our species is whether we can achieve the same alignment — not through millions of years of blind selection, but through the deliberate computation of what selection would eventually produce.
That is the optimization axiom restated as engineering objective.
That is what the Draken framework computes.
Appendix: Key Equations
The Replicator Equation (Taylor & Jonker, 1978): $$\dot{x}_i = x_i \left[ f_i(\mathbf{x}) - \bar{\phi}(\mathbf{x}) \right]$$
The Sheaf Laplacian Quadratic Form (Hansen & Ghrist, 2019): $$\mathbf{x}^T L_{\mathcal{F}} \, \mathbf{x} = \sum_{e=(u,v)} \left\| \mathcal{F}_{u \hookrightarrow e}(\mathbf{x}_u) - \mathcal{F}_{v \hookrightarrow e}(\mathbf{x}_v) \right\|^2$$
The Display-to-Clinch Restriction Map: $$\rho_{D \to Cl}: \mathbb{R}^4 \to \mathbb{R}^3, \qquad (F, E, \Delta m, B) \mapsto (F_{\max}, E_{\text{ratio}}, \Delta m)$$
The Coherence Debt Integral: $$K(t) = \int_0^t \| \sigma_{\text{narrative}}(\tau) - \rho_{D \to Cl}[\sigma_{\text{narrative}}(\tau)] \| \, d\tau$$
The Optimization Axiom ◆: $$\min S_{\text{sys}}(t) \quad \text{s.t.} \quad dH/dt \geq 0$$
The Conservation Priority Hierarchy (derived): $$\text{Protocol} \succ \text{Species} \succ \text{Individual}$$ $$\iff \Gamma_{\text{protocol}} > \Gamma_{\text{species}} > \Gamma_{\text{individual}}$$
Sources and Further Reading
- Water monitor ritualized combat, Mumbai street. — YouTube
- Taylor, P. & Jonker, L. (1978). "Evolutionarily Stable Strategies and Game Dynamics." — PNAS replicator dynamics overview
- Constable & Kokko (2025). "Reconciling ecology and evolutionary game theory." — PNAS
- Cressman et al. (2026). "Global stability of ecological and evolutionary dynamics via equivalence." — PNAS
- Ao et al. (2023). "A Schrödinger Equation for Evolutionary Dynamics." — arXiv
- Papkou et al. (2024). "A rugged yet easily navigable fitness landscape." — Science
- Levin, M. (2025). "The Multiscale Wisdom of the Body: Collective Intelligence as a Tractable Interface." — BioEssays
- Levin, M. (2023). "Bioelectric networks: the cognitive glue enabling evolutionary scaling from physiology to mind." — PubMed
- Hansen & Ghrist (2021). "Opinion Dynamics on Discourse Sheaves." — arXiv
- Boscá & Ghrist (2026). "Selective Adaptation of Beliefs and Communication on Cellular Sheaves." — arXiv
- Spectral fitness landscape ruggedness (2025). — bioRxiv
- Traulsen & Glynatsi (2023). "The future of theoretical evolutionary game theory." — Phil. Trans. R. Soc. B
The substrate cannot escape the game. But it can learn to read the board.