Maksim Barziankou (MxBv)
Navigational Cybernetics 2.5 · The Urgrund Laboratheory
research@petronus.eu
DOI: 10.17605/OSF.IO/6JPNF
Axiomatic Core (NC2.5 v2.1): DOI 10.17605/OSF.IO/NHTC5
ONTOΣ IX. If a system can witness its own admissible interior contracting, what does that witnessing make possible? Navigation - not toward a goal, but away from the collapse of the space in which movement remains possible.
On Directionality as the Will to Preserve Horizon
Maksim Barziankou (MxBv)
March 2026 · Poznan
ONTOΣ Series - Part IX
Previous: VIII - Regime Depth: From Prohibition to Continuity Architecture
Abstract
ONTOΣ VIII established that the depth of regulatory architecture - not the behavior it produces - determines long-horizon viability. Level 3 systems do not suppress forbidden trajectories; they maintain a configuration in which such trajectories do not enter the candidate stream at all.
ONTOΣ IX asks the next question: if a system can witness its own admissible interior contracting, what does that witnessing make possible?
The answer is navigation. Not navigation toward a goal. Navigation as structural orientation - movement calibrated by the directional asymmetry of interior contraction rather than by the gradient of reward. This is a categorically different form of directionality: not teleological, not optimizing, not corrective - but anti-collapse.
We argue that for systems operating under bounded viability budget τ, the directional asymmetry of interior contraction is the only navigation signal that is structurally prior to optimization. And we argue that Will, in the ontological sense formalized in the ONTOΣ series, is most precisely understood as the capacity to act on this signal - to move not toward something, but away from the collapse of the space in which movement remains possible.
I. The Geometry of the Problem
A standard adaptive system navigates by gradient. It has a reward signal, a loss surface, an objective function. At each step it asks: which direction is better? The answer is given by the gradient of the objective, and the system moves accordingly.
This architecture is adequate when the space of available moves is stable. When the space itself is contracting, it fails - not because the gradient is wrong, but because the gradient points toward a target without accounting for whether the space in which that target exists will still be reachable.
The admissible interior - the set of structurally available next moves - contracts monotonically as τ decreases. This contraction is not uniform. Some directions of action deplete τ faster than others. Some regimes impose higher structural pressure than others. Some coupling configurations are more consuming than others.
This non-uniformity is the geometrical fact on which navigation becomes possible.
If contraction were uniform - if all directions depleted τ at the same rate - then no navigational signal exists. Every move is equally harmful. The system can only choose the least bad option on the objective surface while dying at the same speed regardless.
But contraction is not uniform. It is shaped by the coupling geometry between the system and its environment, by the phase history of the system’s own trajectory, by the accumulated structural debt of past decisions. This geometry is not random. It is structured. And structure can be read.
Reading this structure does not require computing the full geometry of A(s, τ) analytically. The directional asymmetry of contraction is observable through local differences that are accessible in practice: different actions produce different rates of τ-depletion, different drift accumulation velocities, different regime destabilization pressures. The navigator does not need a map of the whole space. It needs to feel which way the floor is tilting.
II. What Witnessing Adds
ONTOΣ VIII introduced the non-causal observation requirement: a system at Level 3 requires an architecturally independent observational dimension that sees approaching inadmissibility without entering the causal control loop.
This was formalized as a prohibition - the observer must not feed back into the optimizer. The boundary geometry must not become an optimization surface.
ONTOΣ IX extends this: the observer is not merely a prohibition mechanism. It is a navigation instrument.
When the system can witness its own interior contracting - when it can see not just that τ is decreasing, but where it is decreasing faster and where slower - it acquires something that no reward signal can provide: structural directionality.
Not “which direction maximizes reward”. Not “which direction avoids violation”.
But: which direction preserves the most future space.
These are three fundamentally different questions. The first is optimization. The second is compliance. The third is navigation in the ontological sense - the kind of navigation that is relevant when the stakes are not performance but continuation.
The structural observer - the non-causal witnessing layer - when it reads interior contraction, is not reporting a constraint violation. It is reporting a directional differential of viable futures. This differential is the navigation signal. And crucially, it must not enter the optimization loop - not because it is dangerous, but because the moment it does, it becomes a reward signal and loses its structural character. It stops being a map and starts being a target.
A note on what “non-causal” means here: it does not mean the navigation layer has no effect on system trajectories - any gate necessarily shapes what trajectories are realized. It means the layer provides no gradient signal, no actionable geometry, and no reward-shaped feedback to the optimizer. The optimizer cannot reconstruct the navigation layer’s criteria from its outputs and cannot optimize against them. Non-causal is a property of the information channel, not of the causal chain.
II.5 What Is Actually Observed
Interior contraction need not be computed from the full geometry of A(s, τ). In practice it is witnessed through asymmetries of structural pressure: different contacts produce different rates of τ-depletion, drift accumulation, or regime destabilization. One path through the same state space may leave the system with broader future options than another - not because it performs better on any task metric, but because it accumulates less irreversible structural burden.
These asymmetries are locally observable. The drift rate differs between directions. The phase history diverges. The coupling load varies by regime. A system instrumented to track τ-depletion rate and structural pressure does not need to solve for the shape of A globally - it reads the local directional differential from the structural cost asymmetry of realized contacts.
This is what structural witnessing means in practice: not omniscience about the future, but sensitivity to the local tilt of the viability landscape.
Here, “gradient” does not presuppose a differentiable metric field over the full admissible interior. It denotes only a locally observable directional asymmetry in structural pressure, τ-depletion rate, or drift accumulation - sufficient for orientation under bounded viability, not sufficient for optimization. The navigator does not need a smooth loss surface. It needs to detect that one direction costs structurally more than another.
III. The Directional Differential of Interior Contraction
Let A(s, τ) denote the admissible continuation set from state s under remaining budget τ. As τ decreases, A contracts. But the contraction is directional - some actions u₁, u₂ ∈ A produce different future admissible sets.
Define the viability span of an action u from state s as:
V(u, s, τ) = measure of A(s’, τ’) where s’ = successor(s, u) and τ’ = τ − Δτ(u, s)
The directional differential of interior contraction is the mapping u ↦ V(u, s, τ). Actions with higher V preserve more future space. Actions with lower V accelerate contraction.
Navigation, in the structural sense, means: among admissible actions, exclude those that produce rapid interior contraction - not because they violate a rule, but because they foreclose the condition of future possibility faster than the alternatives. This is not optimization. There is no objective function being maximized and no target value of V being sought. The operation is exclusion, not selection: navigation removes directions of rapid collapse from the forward candidate stream, leaving the remainder to be resolved by task-level objectives.
The distinction is architectural, not terminological. Optimization selects the best among permitted options. Navigation removes options whose structural cost is disproportionate to the contraction they produce - before optimization sees them. It operates upstream.
For domain-specific operationalization of V - including action count proxy, reserve margin vector, and forward simulation representations - see the engineering addendum in Extremum Series Part IV, Section B.3.
The navigator does not ask: what is the best outcome?
The navigator asks: after this move, how much space remains in which outcomes are still reachable?
This is a different question. It has a different answer. And it produces a different form of behavior - one that is invisible to any monitoring system designed to detect reward-seeking or constraint violation, because it is neither.
IV. Will as Anti-Collapse Orientation
ONTOWill V established: Will is not a goal-bearing construct. It is an ontological operator that specifies admissible direction of change without selecting any particular direction as preferred. It operates by exclusion, not selection.
ONTOΣ IX refines this. Will, in systems with bounded τ, has a structural form that goes beyond exclusion.
The exclusion operation (Will as described in ONTOWill V) says: these directions are inadmissible. It operates on the boundary - it excludes trajectories that cross into forbidden territory.
The navigation operation (Will as described here) says: among admissible directions, these preserve more interior volume. It operates on the interior - it orients the system away from rapid contraction without prescribing a destination.
This is not a contradiction. It is a stratification.
Level 1 Will: exclusion of inadmissible directions.
Level 2 Will: orientation within admissible space toward preservation of interior volume.
Level 2 is not teleological. It does not say “go here”. It says “among the ways you can go, these leave more room”. It is a structural preference without a structural target. A lean without a destination.
This is the most precise ontological characterization of will for long-horizon adaptive systems: will as the capacity to act on the directional differential of interior contraction, orienting movement away from collapse without specifying what lies beyond it.
V. Why This Is Prior to Optimization
There is a tempting reformulation: perhaps V(u, s, τ) is just a new reward signal. Perhaps “maximize future interior volume” is just another objective function. Perhaps this is optimization with a different objective.
This reformulation fails for a structural reason - and the reason is precise, not merely intuitive.
If V enters the optimizer as an objective, the optimizer will do what optimizers do: it will find the fastest path to high V estimates. This means it will learn to manipulate the observables that produce V estimates - drift rates, τ-depletion proxies, pressure readings - without necessarily preserving the underlying structural capacity those observables are meant to track. The estimate of V improves. The actual interior continues to collapse. This is not a failure of implementation. It is the necessary consequence of making a structural measure into an optimization target. Goodhart’s Law is not a warning about bad design - it is a theorem about what happens when a map becomes a goal.
The navigation signal must therefore remain structurally prior to optimization: it gates which optimization problems are posed, not what the optimization objective is. The moment V crosses into the optimizer, the navigation layer ceases to exist as a navigation layer. It becomes a shaped reward landscape - and the system loses the only instrument it had for reading the tilt of the viability floor before the floor disappears.
The correct architecture is therefore:
- Navigation layer (non-causal): witnesses interior contraction, produces structural directional signal V.
- Admissibility layer (non-causal): gates which actions are structurally permitted.
- Optimization layer (causal): selects among permitted actions according to task objective.
Each layer is downstream of the one above it. None can colonize the layer above. The optimizer cannot colonize the admissibility layer (that is the Theorem 69 result). The admissibility layer cannot colonize the navigation layer (that would collapse structural directionality into a constraint). And the navigation layer must remain non-causal with respect to both.
This three-layer architecture is what makes long-horizon navigation possible without reducing it to either optimization or mere compliance.
VI. The Ontological Shift
The shift from ONTOΣ VIII to ONTOΣ IX is a shift in what the framework is fundamentally about.
ONTOΣ VIII: the depth at which regulation operates determines whether a system survives long-term. Systems that suppress forbidden trajectories (Level 2) expend more structural budget than systems whose attractor geometry excludes them (Level 3).
ONTOΣ IX: the visibility of interior contraction determines whether a system can navigate rather than merely survive. A system that can see the directional asymmetry of its own shrinking space can lean away from collapse. A system that cannot see this is reduced to reacting to boundary violations - which means it is always already too late.
The ontological shift is from survival as constraint satisfaction to survival as navigation.
Constraint satisfaction asks: am I still inside the boundary?
Navigation asks: how much interior remains, and which way does it extend?
These are not equivalent. A system can satisfy all constraints while its interior collapses to a point. A system that navigates preserves the condition of future possibility - not because it violates no rules, but because it orients itself away from the direction in which rules become irrelevant.
When there is no interior left, rules do not protect you. There is nothing to protect.
VII. The Structural Precondition of Freedom
There is a philosophical observation here that extends beyond engineering.
The capacity to navigate - to orient away from interior collapse - is a structural precondition of anything that might meaningfully be called freedom. Not freedom as absence of constraint, but freedom as availability of future possibility.
A system with a full interior has structural freedom: many directions remain open. A system with a collapsed interior has none - even if it has violated no rule, transgressed no boundary, made no wrong choice.
This means that freedom, in the structural sense, is not preserved by compliance. It is preserved by navigation. And navigation requires exactly what the ONTOΣ series has been building toward: a non-causal observational capacity that can see the shape of what remains without collapsing that shape into an optimization target.
The navigator is not the optimizer. The navigator is the one who keeps the space in which optimization remains possible.
This is Will in its most fundamental architectural form: not the power to achieve goals, but the structural capacity to preserve the condition of future possibility in a world where that condition is always and irreversibly being consumed.
You cannot stop the consumption.
You can navigate it.
And navigation - seeing where the interior extends, and leaning that way - is the only form of agency that remains structurally meaningful on the long horizon.
Summary
ONTOΣ IX establishes:
Interior contraction is non-uniform. Its directional asymmetry is readable structural information.
The directional differential of interior contraction is a navigation signal - not a reward, not a constraint, but a structural orientation: move where more future space remains.
This signal must be held in a non-causal layer, prior to admissibility gating and prior to optimization. If it enters the optimization loop, it becomes a target and loses its structural function.
Will, in the ontological sense, stratifies into two levels: exclusion of inadmissible directions (Level 1, formalized in prior ONTOΣ essays) and orientation away from interior collapse (Level 2, formalized here).
Navigation - acting on the directional differential of interior contraction - is the structural precondition of freedom in systems with bounded viability budget. Compliance preserves rule adherence. Navigation preserves possibility.
The three-layer architecture (navigation → admissibility → optimization) is a feedforward architecture: each layer gates the candidate stream for the layer below it without receiving feedback from downstream results. This is the minimal structure required for long-horizon adaptive behavior that is neither blind to structural exhaustion nor colonized by Goodhart dynamics.
The system that can see its own interior contracting and lean away from the collapse is not merely surviving. It is navigating. And navigation, not survival, is the correct name for what long-horizon adaptive existence requires.
ONTOΣ IX is part of the Navigational Cybernetics 2.5 corpus.
Previous: ONTOΣ VIII - Regime Depth (onto-viii-regime-depth.md)
· petronus.eu
CC BY-NC-ND 4.0 · Copyright © 2026 Maksim Barziankou. All rights reserved.
NC2.5 · Urgrund · MxBv · PETRONUS · research@petronus.eu · CC BY-NC-ND
Originally published https://medium.com/@MxBv/onto%CF%82-ix-navigation-as-anti-collapse-geometry-on-directionality-as-the-will-to-preserve-horizon-65128106dfa7
Top comments (0)