Data Model

MindPrism stores meaning not as a flat set of tokens and not as a continuous cloud of numbers, but as a structured algebra of relations governed by strict thermodynamic laws. This is a fundamentally different way of representing the world: the system does not simply remember what was said or seen, but assembles what is happening into hierarchical semantic constructions where each element has a role, connection, confidence level, and lifecycle. In such a model, data ceases to be "input" and becomes a living cognitive environment in which conclusions, verifications, and new associations are born — but only if they pass the architectural filters of structural integrity.

At the core of MindPrism lies not a monolithic memory but a strict division between a stable species core and a mutable individual circuit. The core defines the alphabet of representations: what types of entities exist, the immutable rules of their composition, and the fundamental system roles. The individual circuit stores the current session, local connections, temporary states, and adaptation dynamics. This means that the system does not rewrite itself entirely with each new experience. It unfolds the world anew, but does so on a solid foundation, maintaining stability and interpretability.

Ternary Representation and the Absolute Signal Barrier

The basic unit of MindPrism is the ternary vector. It does not reduce to the familiar binary logic of "yes/no" but allows three states: presence (+1), absence (0), and negation (-1). For a cognitive architecture, this is critical because the real world rarely fits into a simple on/off scheme. Sometimes an object exists. Sometimes it doesn't. And sometimes what matters is not just absence, but specifically active displacement, inhibition, conflict, or prohibition.

However, zero in MindPrism is not merely a "missing value" — it is an absolute algebraic barrier. Data in MindPrism is stored as structures with explicit semantics, where zero acts as a hard break in the synaptic graph, programmatically blocking the propagation of information. This strict separation of signal presence and sign ensures that meaning does not blur during processing and semantic layers never mix.

Algebraic Composition and Strict Binding Rules

In MindPrism, every meaningful concept can be represented not only as an object but also as a name, role, or relation. This allows the system to distinguish "what it is," "what function it performs," and "in what context it acts." Such architecture is especially important for complex reasoning chains where the same object can be a cause, a consequence, an obstacle, or a landmark.

Unlike naive compositional models, MindPrism enforces strict algebraic rules on binding (BIND). Direct binding of two sparse, unstructured representations is categorically forbidden, as it generates an irreversible noisy hash that destroys the semantics of the connection. Binding is only architecturally valid for specific structural pairs (e.g., a dense, fully present role vector binding with a sparse data vector).

Furthermore, extraction (UNBIND) from a superposition is never direct. To extract a role or object from a complex overlapping state, the system must first project the data into a linear field, apply the algebraic extraction, and then strictly sanitize the result — clearing any structural "garbage" bits left in the absence masks. This guarantees that retrieved structures are clean and do not pollute the working memory.

Tree Composition and Recursive Chunking

MindPrism assembles complex representations not into chaotic superpositions but into hierarchical trees. Simpler elements form stable nodes, nodes unite into larger structures, and these structures become the basis for new meanings. However, trees cannot grow indefinitely. Due to the algebraic limits of signal stability during retrieval, the depth of a tree is strictly bounded by an architectural constant.

To overcome this depth limit without losing compositional power, MindPrism uses Recursive Chunking. When a branch reaches the maximum allowed depth, the system forcibly compacts the entire detailed sub-tree into a single, dense "macro-step" (a new name), resetting the depth counter. This process is recursive: detailed sparse sub-trees are folded into dense macro-steps of the first level, which themselves can become sub-trees for second-level macro-steps. This allows the system to store arbitrarily long temporal and logical sequences as a hierarchy of abstractions without collapsing the algebraic stability of the computation.

Compaction, Resource Regulation, and Recycling

One of the key properties of the MindPrism data model is the ability to compact complex fragments into dense semantic units (PROMOTE). When the system encounters repeating structures or highly connected sub-trees, it can translate them into a dense name that represents the entire fragment as a single atom.

However, compaction is not a mere convenience; it is a thermodynamically regulated operation. It requires internal systemic energy (drive) to execute. Under conditions of high internal conflict or stress, compaction is blocked to prevent the crystallization of contradictory structures — except in cases of absolute survival necessity, where the system forces compaction regardless of stress.

Conversely, if a compacted structure loses relevance or exhausts its resource, it can be unfolded back (DEMOTE) into a sparser representation through a process of recycling. The dense name is freed, and the original structure returns to a freer form. Memory thus maintains internal order not by hoarding, but through regulated structural breathing.

Memory Traces, Phantoms, and Emergent Sleep

MindPrism distinguishes short-lived and long-term structures. Working memory stores the current state of the scene, active hypotheses, and what the system needs right now. Data in working memory possesses a thermodynamic resource that decays if not confirmed by predictive matching. Long-term memory works on an additive principle: new events do not erase old ones but are built on top of them, preserving historical trajectory.

When the resource of a node is exhausted, it does not disappear instantly. It transitions to a phantom state — structurally present in the archive but no longer participating in active computation. This maintains historical continuity without overloading the active working field.

Crucially, memory cleaning is not a scheduled task. When the pool of available structural nodes is physically exhausted, the system enters an emergent sleep state. This forced low-energy attractor halts active cognitive processing, allowing background mechanisms to safely reorganize, compact, and clean the memory graph, ensuring the system survives its own resource limits without crashing.

Resonant Retrieval and Structural Sanitization

Search and retrieval in MindPrism occur not through crude scanning but through resonance — iterative projection onto orthogonal bases. If the match is sufficiently strong, the node is extracted. However, the system enforces strict limits on the depth and duration of resonant search to prevent the process from falling into a "black hole" of noise where signal is lost.

If a search is interrupted due to urgency or low confidence, the partially extracted data must pass through a structural sanitization gate before re-entering the computational field. This gate algebraically zeroes out any unstable or unsubstantiated components, ensuring that noise or incomplete residues are not mistaken for confident knowledge.

Modality Isolation and Subsumption

MindPrism does not mix different types of information into one mass. Visual, linguistic, spatial, and audio data exist as specialized modalities, but they do not form a "federation" exchanging raw data on equal terms. Because different processing tracts operate at radically different dimensionalities and speeds, direct algebraic mixing (superposing a fast sensory vector with a deep linguistic vector) is architecturally impossible.

Instead, modalities interact through subsumption and translation interfaces. Lower-order modalities do not send raw packets to higher-order ones; they project scalar modulators and structural role indexes. Higher-order circuits read these indexed roles, not the raw sensory vectors. This strict isolation prevents cross-modality noise and ensures that the system builds a coherent world model through hierarchical coordination, not chaotic data blending.

Data Lifecycle and Thermodynamic Conflict

Data in MindPrism is not static. They are born, activated, compacted, lose resource, go into phantom state, and are then either recycled or disappear. But this lifecycle is subject to the law of thermodynamic conflict (cognitive dissonance).

If mutually opposing concepts actively annihilate each other in the system's linear field, they do not silently cancel out to zero. If the energy expended on this mutual destruction exceeds a critical threshold, the system enters a state of unresolved conflict. This state physically freezes plasticity and blocks motor output, forcing the system to resolve the paradox through structural restructuring or associative drift, rather than ignoring the contradiction.

This makes the architecture not only efficient but honest. If an element is no longer supported by prediction, it should not pretend to be alive. If structures conflict, they must be resolved, not averaged. If meaning has matured, it must be compacted. This is what a mature data model consists of: not in accumulating everything indiscriminately, but in the ability to maintain structural discipline under the pressure of a living, contradictory world.