SYSTEMS / AURORA

—

TR-2025-28

Aurora v1.0

Adaptive Reasoning Model. A reasoning-first language model built on the Adaptive Cognition Layer with emotional memory encoding and predictive hallucination detection.

ACTIVE
|Version 1.0|Reference: TR-2025-28

94%

Hallucination Detection

78%

Emotional Memory

96.7%

Local Routing

0.73

Confidence Correlation

OVERVIEW

Aurora is Thynaptic's primary cognitive model implementation. It operates within the 10-component Adaptive Cognition Layer pipeline, receiving structured context from memory systems, reasoning routers, and safety validators at every inference step.

Unlike standalone language models, Aurora maintains self-awareness through the ACL's confidence scoring system. It knows what it remembers, what it's uncertain about, and when to defer to external verification.

MODEL ROUTING

Default Local— Fast, thinking mode enabled
qwen3:1.7b
Fallback— Reliable backup
granite3.2:2b
Cloud— 94.83% MMLU accuracy
gpt-oss:20b
Stickiness— Maintains consistency
3 turns
Thinking Trigger— Non-casual queries
>80 chars

CAPABILITIES

Emotional Memory

Tracks valence, tone, intensity for each memory entry

Hallucination Detection

Recall-based verification against workspace knowledge

Hybrid Routing

Local-first with cloud fallback for capability

Confidence Scoring

Self-aware metrics exposed for transparency

RELATED

TR-2025-28: Model Card →ACL System →Aurora Research →