Skip to main content
Back to 7 Genome System
G7 — IMPROVEMENT

RCT-7 Genome

Mental OS and Self-Evolution Engine — a system that continuously learns and improves itself.

Continuous Cycle: G7 → G1 → G2 → ... → G7

RCT-7 is the final genome in the cycle but not the endpoint. The signals G7 generates feed back into G1 (Architect's Genome), allowing the system's 'WHY' to evolve with real-world experience. This cycle is what makes the RCT Ecosystem not merely a tool suite, but a living system.

4 Improvement Variants

RCT-S

Systemic

4-Step system-level improvement

  1. 01OBSERVE — Monitor system-wide performance metrics across all 7 genomes
  2. 02ANALYZE — Identify root cause of metric divergence (not symptoms)
  3. 03REDESIGN — Propose architectural changes at genome level (requires Codex review)
  4. 04VALIDATE — A/B test redesign against baseline; constitutional review before deploy

Use for: Core architecture changes, new genome-level rules, major capability expansions

RCT-I

Interpretation

7-Step intent-level improvement

  1. 01RECEIVE — Collect intent classification failures from WF-00 Interpreter logs
  2. 02DECODE — Map failure patterns to ambiguity categories
  3. 03HYPOTHESIZE — Generate candidate disambiguation rules
  4. 04TEST — Run candidates against historical false-classification corpus
  5. 05COMPARE — Score candidates against current classifier performance
  6. 06INTEGRATE — Merge winning rules into JITNA I-primitive schema
  7. 07MONITOR — Track live classification accuracy for 30-day stability window

Use for: JITNA protocol improvements, intent schema updates, classification accuracy

RCT-V

Visual

Visual output and UX improvement loop

  1. 01CAPTURE — Record user interaction patterns with visual outputs
  2. 02CLUSTER — Group interaction patterns by output type and user segment
  3. 03REDESIGN — Update visual output templates based on cluster analysis
  4. 04DEPLOY — Progressive rollout with satisfaction measurement

Use for: UI layout improvements, visual output quality, dashboard enhancements

RCT-F

Framework Expansion

4-Dimension new capability framework

  1. 01IDENTIFY — Define the new capability's scope and constitutional impact (Codex 09 review)
  2. 02DESIGN — Build with mandatory Janus Destroy-face (Codex 10 compliance)
  3. 03PILOT — Controlled rollout to 1% of sessions with full telemetry
  4. 04SCALE — Progressive expansion with constitutional gate at each tier (1% → 10% → 100%)

Use for: New product features, new genome additions, major autonomy level changes

7-State IntentLoop

IDLE
RECEIVE
PARSE
ROUTE
EXECUTE
VERIFY
ADAPT
↻ G1

Every agent session runs this 7-state IntentLoop. ADAPT is the critical seventh state — it does not merely deliver output, but writes learnings back into the system before the next loop begins. This ensures each interaction leaves the system marginally smarter than before.