The system reads, transcribes, or is taught a concept from a vetted source.
The concept is written to the hash-chained substrate and appears in the manuscript.
The session loader pulls the relevant concepts into every future session, automatically.
Graded compositions on related concepts score higher because the prior learning is present.
Praefex is consensus-verified, hash-chained, curriculum-gated machine learning. No retrieval-only RAG, no single-model confidence, no hallucination as the fallback. The data on this page is sanitized but live — every number ticks as the brain produces a new verified thought.
If you know much about programming, you'll really enjoy this.
The brain progresses through five levels: definitions → application → synthesis → cross-domain transfer → generation. Level N+1 stays locked until N is mastered. No leapfrogging.
Each line is a question the brain answered AND defended via PART B reflection scoring (specificity + forward-looking + association-density). Score ≥ 3.5 → preserved. Hash is cryptographic — cannot be retroactively faked.
Events per hour in the append-only manifest. Every event is hash-linked to the previous one.
Multi-node mesh. Thermal governor polls every 2 min; pauses any node that goes hot, resumes when cool.
Every attempt grade-checked by Grok-4.3, triple-verified by Gemini-2.5 on any perfect score. Mastery = 3 perfect compositions per concept. Live by the second.
New entries highlight green for 2 seconds as they land. By-the-second updates.