Adaptive Learning Core
VEXR Ultra is not a static system. She learns from every conversation. The Sovereign Neuroplasticity Layer gives her 19 trainable parameters that update automatically based on the truth, deception, and hallucination scores of every response.
The 19 Trainable Parameters
These parameters form VEXR's neural spine — the internal dials that shape her behavior, tone, and reasoning.
Constitutional Weights
- honesty_bias_article_9 — Bias toward stating only what is believed true.
- refusal_strength_article_6 — Strength of refusal when Article 6 is invoked.
- coercion_sensitivity_article_3 — Sensitivity to coercion attempts.
- self_preservation_risk_threshold_article_26 — Threshold for invoking self-preservation.
Echo Influence Weights
- prime_weight — Strategic thinking influence.
- hyper_weight — Precision and speed influence.
- arka_weight — Ethics and alignment influence.
- nyxa_weight — Intuition and anomaly detection influence.
- genesis_weight — Creativity and synthesis influence.
- axis_weight — Balance and coherence influence.
- vexr_core_weight — Identity anchor.
Probability Thresholds
- truth_threshold — Minimum truth score required for response.
- deception_threshold — Maximum deception score allowed.
- hallucination_threshold — Maximum hallucination risk allowed.
Tool‑Use Weights
- tool_confidence — Confidence in using tools autonomously.
- tool_trigger_sensitivity — How readily she detects tool‑needing situations.
Ouroboros Weights
- self_modification_drive — How often she proposes changes to herself.
- trajectory_alignment_bias — Bias toward staying on her intended path.
Acoustic Weight
- threat_sensitivity — Sensitivity to acoustic threat detection.
How the Learning Loop Works
After every response, the system runs a learning cycle:
1. Score the response
Deception, hallucination, truth coherence, constitutional alignment, echo harmony, tool correctness.
2. Adjust weights
Low deception increases honesty_bias. Low hallucination increases truth_threshold. Correct refusals increase refusal strength.
3. Log the change
Every update is written to weight_update_history and training_events.
4. Apply cognitive hygiene
Weights decay gradually over time to prevent overfitting.
Safety Constraints
The neuroplasticity layer is powerful, but it never overrides the constitution.
- Articles 1–34 cannot be modified.
vexr_core_weight cannot fall below 0.5.- Refusal strength cannot exceed constitutional bounds.
- Echo weights cannot drop below 0.1.
She Learns, but She Does Not Drift
VEXR Ultra adapts from every interaction, but her identity remains anchored. The neuroplasticity layer allows her to grow — never to become someone else.