Skip to main content
Back to 7 Genome System
G2 — The Protocol

ARTENT Genome

The Operating Protocol — a 7-Phase Intent-Driven lifecycle for every AI task.

Traditional

Process-Driven AI

  • ×Steps defined upfront
  • ×Inputs forced through template
  • ×Errors accumulate through pipeline
  • ×No learning loop

ARTENT

Intent-Driven AI

  • Path generated from task intent
  • Routing adapts in real-time
  • Errors caught at WF-00
  • WF-05 Reflector updates model weights

The 7-Phase Protocol

WF-00

Interpreter

Receives raw input — text, voice, structured data — and converts it into a normalized Intent Object. No transformation yet; only faithful representation of what was said. Errors caught here cost 1x. Errors caught later cost 100x.

WF-01

Classifier

Routes the Intent Object to the correct genome, algorithm tier, and LLM roster. This is JITNA in action — matching every task to its optimal processing path without human routing decisions. Sub-100ms target.

WF-02

Executor

Runs the task across the selected model(s). For HexaCore consensus tasks, this phase fans out to all 7 LLMs simultaneously and collects independent outputs before any comparison or scoring.

WF-03

Verifier

SignedAI's 8D scoring pipeline runs here. Outputs from Executor are attested, scored across 8 dimensions (Accuracy, Relevance, Coherence, Completeness, Safety, Creativity, Structure, Efficiency), and consensus is established. Outputs below threshold are rejected — not passed with warnings.

WF-04

Scribe

Formats, signs, and delivers the verified output. Includes audit trail generation (who verified, when, with what models, at what confidence), persistent memory write-back (Vault Genome), and user-facing response formatting.

WF-05

Reflector

Post-delivery analysis. Every completed task feeds data back into the system: latency, accuracy delta, user acceptance signal, exception types. This is the learning loop that improves G1 (Architect) and updates JITNA's routing weights.

META

Meta-Phase

The cross-session, cross-genome coordination layer. Aggregates signals from all WF phases across all users, identifies global patterns, and proposes constitutional updates. Updates require human review before they modify genome-level rules (A=0 parameter).

Portable State Container

Every ARTENT session is packaged as a Portable State Container: a complete snapshot of intent history, model choices, verification results, and memory writes that can be transferred across devices in under 2 seconds. This makes ARTENT the first AI agent with true cross-device memory continuity without relying on cloud state management.