Humanizer Pattern Selection
Decision Flow
Does the agent run across multiple sessions?
NO → Humanizer patterns do not apply; use Signal patterns for in-session persona
YES — start here:
H1 (Identity Persistence) — PREREQUISITE for all other Humanizer patterns
Stable identity must exist before it can evolve
After first failures emerge:
H2 (Episodic Self-Improvement) — learn from mistakes across sessions
Requires: K11 or K10 as memory substrate
After first successes:
H4 (Procedural Skill Accumulation) — distil successful trajectories into reusable skills
Complements H2: H2 learns from failure, H4 from success
As user model grows:
H7 (Adaptive Persona) — adapt communication style per user
H10 (Relational Memory) — persist user relationship state
⚠ H10 requires explicit user consent and right-to-deletion
When reasoning loops stall or creativity degrades:
H3 (Entropy-Driven Curiosity) — autonomous deadlock breaking
For persistent background reasoning between turns:
H6 (Continuous Inner Monologue) — separate thinker from responder
For accurate self-knowledge and capability routing:
H9 (Observational Identity) — explicit model of own capabilities
With human governance board and formal oversight:
H5 (Constitutional Self-Alignment) — evolving principles with mandatory checkpoints
⚠ NEVER implement H5 without mandatory human review; alignment risk
Adoption Sequence
| Stage | Patterns | Purpose |
|---|---|---|
| Foundation | H1 | Stable identity across sessions |
| Learning | H2 + H4 | Improve from failure and success |
| Adaptation | H7 + H10 | Serve users better over time |
| Advanced | H3 + H6 + H9 | Autonomous, self-aware operation |
| Governed | H5 | Evolving principles with oversight |
All Humanizer patterns require K11 (Observational Memory) or K10 (Long-Term Memory) as infrastructure. H1 is a prerequisite for all others.
Anti-Patterns
- HA1 — Simulated Emotion: emotional language without genuine affective model (manipulation)
- HA2 — Unbounded Relationship Depth: H10 without ethical guardrails $\to$ parasocial harm
- HA3 — Identity Drift: H7/H10 without H1 $\to$ agent becomes whoever the user wants
- HA4 — Autonomous Principle Adoption: H5 without human review $\to$ alignment risk
- HA5 — Stale Self-Model: H9 without decay functions $\to$ overconfident outdated self-assessment