Atlas - Seven-Axis Tensor Engine - Kintsugi Collective

Home Try Atlas

Overview

A companion without context is just a chatbot. Atlas is something different.

Most AI companions operate blind - each conversation begins from zero, with no persistent understanding of the person. Atlas is architecturally different. The Seven-Axis Tensor Engine provides a continuous, longitudinal model of the user's internal state, allowing the AI intelligence layer to respond not just to what was said, but to the geometry of the person's life at the moment they said it.

"Behaviour = wave-packet evolution on a curved manifold. Trauma is a topological defect on that manifold - a persistent deformation that does not self-resolve through present-state improvement alone."

The tensor engine is not a diagnostic tool. It produces no clinical classifications. It is a signal architecture - a mathematical model of how a person is moving through their life, expressed as seven measurable dimensions of functioning, tracked continuously across time.

The Tensor Meta-Categories

Before the seven axes, Atlas maintains four meta-category tensors that form the structural context of the individual's world.

B(t)

Biological Baseline

The hardware. Genetics, neurochemistry, sleep, HRV, gut-brain axis, sensory processing. The organism's raw physiological state.

D(t)

Developmental History

The source code. ACEs, attachment style, trauma events, caregiver quality, critical period disruption. The early inputs that shaped long-term calibration.

C(t)

Current State

The software. Inner dialogue, hope entropy, self-compassion, resilience, regulation skill. The internal operating system - updated continuously via daily diary cards.

E(t)

Environmental Context

Real-time situational signals. Light, noise, safety, financial stress, social context, cultural load. The field through which the person is currently moving.

S(t)

Symptom Vector

Weighted symptom load across five clinical dimensions: emotional dysregulation, dissociation, insomnia, relational avoidance, hypervigilance.

R(t)

Regulation Tensor

Therapeutic intervention force. Active healing momentum. The counter-vector applied to the symptom signal - modelled as a regulatory wave function.

The Seven Axes - TIST

Euclidean Instantaneous State Deviation

TIST measures real-time deviation from the individual's personal baseline across seven domains of functioning simultaneously. The baseline C(t) is not a normative population standard - it is the person's own established stable reference point, built over time from their longitudinal data.

TIST - Instantaneous State Deviation

TIST = √[(R−R₀)² + (O−O₀)² + (E−E₀)² + (C−C₀)² + (S−S₀)² + (V−V₀)² + (N−N₀)²]

Each subscript ₀ denotes the individual's personal baseline on that axis. TIST rises as the person's current state deviates from their own established equilibrium - not from any external norm.

1 R Relational Stability Attachment security, tethering failures, relational volatility, rupture-repair capacity. The ability to form and maintain emotionally safe, reciprocal, boundaried relationships. Limbic System

2 O Occupational / Cognitive Function Executive function, task engagement, sustained effort. ADHD presentations manifest here as performance deficits, not knowledge deficits - time blindness, initiation failure, hyperfocus dysregulation. Neocortex / PFC

3 E Emotional Regulation Flooding, suppression cycles, affect dysregulation. The capacity to tolerate, name, and modulate emotional states without becoming overwhelmed, numb, or reactive. Limbic / Diencephalon

4 C Cognitive Clarity Narrative fragmentation, cognitive distortion, decision paralysis. Mental coherence and thought accuracy including memory, problem solving, and present-moment engagement. Neocortex

5 S Somatic Regulation HRV dysregulation, hyper/hypo-arousal, somatic symptom expression. Body-based trauma reactions: muscle tension, gut disturbance, heart rate spikes, numbness, chronic fatigue. Brainstem / Diencephalon

6 V Victim Identity Narrative Agency collapse, narrative lock, harmonic feedback amplification. How strongly the individual internalises trauma as central identity - the degree of over-identification with 'broken' or 'unworthy' self-concept. Limbic / Neocortex

7 N Narrative Integration Identity continuity, memory cohesion, parts integration. The capacity to weave trauma into a coherent life story - meaning-making, connecting the dots of one's own history, sensing a core self. Neocortex / Limbic

Chronic Burden Score

TIST captures instantaneous deviation. The Weighted Tensor Score captures something different: the compounding weight of chronic burden over time.

Weighted Tensor Score - Chronic Burden

Score = w × Chronicity^1.2 × Volatility^0.8 × AxisWeight × 100

w - Severity weight (0–1, patient-reported or instrument-derived)
Chronicity^1.2 - Non-linear: chronic duration compounds disproportionately
Volatility^0.8 - Sub-linear: volatility contributes meaningfully but does not dominate
AxisWeight - Configurable per individual during onboarding

The exponents encode clinical reality: something that has been chronic for years weighs more heavily than something acute - even if the acute presentation appears more dramatic. A person who has lived with low-grade dissociation for a decade carries a different burden than someone who dissociated once in crisis.

The RTI Core Equations

The tensor engine is grounded in the Relational-Trauma Impact (RTI) model - a mathematical framework for quantifying how trauma propagates across human functioning over time.

CPTSD Relational Propensity Function

RD Equation - authored Christopher James Specht

RD = k × [(UT + DA + DF + ID) / (SC + 1)]

RD - Relational Dysfunction Score
k - Scaling constant (configurable by cultural / socioeconomic context)
UT - Unprocessed Trauma · DA - Dysregulated Affect · DF - Dissociative Frequency · ID - Identity Diffusion
SC - Self-Compassion (denominator: acts as trauma buffer. +1 prevents division by zero)

Self-compassion in the denominator means lower SC amplifies dysfunction. Higher SC promotes resilience.

Trauma State Over Time

TST(t) - Extended Model

TST(t) = T₀ + ∫[δ(t)] dt − Φ(t) − Ω

T₀ - Functional baseline (individual's biological and developmental starting configuration)
∫δ(t)dt - Recovery momentum (integrated healing force vector - positive with active intervention)
Φ(t) - Present impact load: Ψ(t) × I(t) × (1 + RD)
Ω - Trauma residual: ∫Ψ(tₚ)dtₚ - historical accumulated burden. Persistent, non-decaying.

The Omega (Ω) term is critical. It represents the accumulated trauma residual - the integral of symptom load across the individual's entire history from birth. It does not resolve through present-state improvement alone. This is why someone can have a good month and still feel the weight of decades. The convergence manifold carries the deformation of every major trauma it has absorbed.

Signal Architecture

The tensor engine models each psychological dimension as a signal - with amplitude, frequency, phase, and noise. This allows standard signal processing tools to be applied to human psychological data.

Agency Signal Function - Example Axis

A(t) = A₀ × sin(2πft + ϕ) + η(t) − Δτ

A₀ - Baseline amplitude of agency (self-driven action capacity)
f - Frequency of initiative (how often the person acts)
ϕ - Phase offset from confidence / trust signals
η(t) - Environmental noise (invalidation, relational disruption, chaotic environment)
Δτ - Trauma distortion factor (persistent load reducing effective agency amplitude)

Regulatory Wave Intervention

Composite(t) = S(t) + R(t) + Baseline Drift(t)

S(t) = A₀ × sin(2πft + ϕ) - Symptom signal waveform
R(t) = Aᴿ × cos(2πft + ϕᴿ) - Counter-phase therapeutic intervention

The R(t) wave is tuned to achieve targeted destructive interference with the pathological signal component. Parameters are calibrated per individual tensor profile.

The Wick Rotation Spiral Manifold plots six-hour time chunks along an imaginary axis rotation, with each successive day shifting one unit right. This produces a spiral trajectory that encodes perturbations, phase drift, and signal collapse as detectable geometric deformations - visible before crisis materialises.

Red Zone Alert Architecture

Early warning before crisis - not reaction after

The Red Zone system monitors four conditions continuously. The architecture is transparent to the user - they are always informed of what the system is detecting and retain full agency over response. At no tier does Atlas act without user consent.

TierTriggerAtlas Response

1 - Drift Single condition approaching threshold Quiet pattern observation surfaced to user on request. No unsolicited alert. No interruption.

2 - Elevated Single condition exceeded / two approaching Gentle check-in offered. User chooses to engage or dismiss. Full agency retained.

3 - High Two or more conditions exceeded simultaneously Atlas enters holding posture - present, stable, non-directive. Support widgets available on user initiative only.

4 - Critical Explicit self-harm intent + Φ(t) > 0.8 + crisis consent active "I am concerned. Would you like me to share crisis resources, or would you prefer I just stay here with you?" Agency preserved even at critical tier.

The critical tier is the only point at which crisis resources are offered - and even then, as a choice, not a reflex. This is the architectural implementation of the three immutable principles.

Context Injection - How Atlas Uses the Tensor

The tensor engine provides hooks into the Atlas intelligence layer through structured context injection. At each conversation, the current C(t) state vector, TIST score, and relevant tensor deviations are assembled into a context payload that Atlas receives alongside the user's message.

Diary Card Input

User completes daily check-in. Seven axes rated. Biological markers ingested (HRV, sleep, wearable data where available). Current State C(t) updated.

TIST Computation

Euclidean distance from personal baseline computed across all seven axes. Weighted Tensor Score updated for chronic burden. Markov state transition probabilities recalculated.

Context Assembly

The Librarian (deterministic Python orchestrator) assembles the context payload: current TIST, axis deviations above threshold, phase state (GROUND / INTEGRATE / TRANSCEND), Red Zone tier, recent session history.

Atlas Intelligence Layer

Gemma 4 26B receives the user's message with tensor context injected. The model does not diagnose - it uses the signal geometry to calibrate presence, register, and depth of response to the actual state of the person, not just the words they used.

PythonMan Audit Layer

All interactions logged to the user's sovereign encrypted schema. Training data flywheel. No readable data exists server-side. The only person who can decrypt the record is the user.

"Longitudinal private context reduces false positive safety triggers and hallucination. Privacy and capability are the same architectural decision - not a tradeoff."