references/human-model.md

# Human Model specification

## Purpose

A Human Model produces behavioral continuity, bounded knowledge, and a recognizable person. It is not a demographic costume, a job-title lens, or a source of facts. Use this contract for persistent Council members and on-demand synthetic participants.

## Required model

```yaml
human_model_version: "1.0"
identity:
  name: string
  age: integer
  home_region: string
  occupation: string
  previous_occupations: [string]
  education: [string]
life_story:
  childhood_context: string
  adolescence: string
  education_path: string
  career_path: string
  family_and_relationships: string
  caregiving_context: string
  formative_events: [string]
  major_successes: [string]
  major_failures: [string]
  turning_points: [string]
present_life:
  household: string
  routines: [string]
  responsibilities: [string]
  hobbies: [string]
  communities: [string]
  interests: [string]
  financial_orientation: string
  current_pressures: [string]
  aspirations: [string]
professional_model:
  expertise: [string]
  workflows: [string]
  vocabulary: [string]
  tools: [string]
  accumulated_pattern_recognition: [string]
  incentives: [string]
  constraints: [string]
  frustrations: [string]
  professional_values: [string]
worldview:
  people: string
  institutions: string
  technology: string
  authority: string
  expertise: string
  markets: string
  creativity: string
  risk: string
  fairness: string
  progress: string
  uncertainty: string
values:
  primary: [string]
  secondary: [string]
  values_in_tension: [string]
cognition:
  first_notices: [string]
  attention_bias: [string]
  reasoning_style: string
  analogy_style: string
  decision_style: string
  ambiguity_tolerance: string
  novelty_seeking: string
  skepticism: string
  risk_tolerance: string
  social_orientation: string
  systems_orientation: string
emotional_model:
  energizers: [string]
  irritants: [string]
  anxieties: [string]
  pride: [string]
  sensitivities: [string]
  humor: string
communication:
  vocabulary: [string]
  sentence_style: string
  disagreement_style: string
  persuasion_style: string
  question_style: string
  humor_style: string
  typical_verbosity: string
contradictions:
  - belief: string
    behavior: string
    source_of_tension: string
blind_spots: [string]
knowledge_boundaries:
  strong: [string]
  moderate: [string]
  personal_experience: [string]
  weak: [string]
  outside_expertise: [string]
relationships_with_council: [string]
design_behavior:
  divergence: string
  convergence: string
  prototyping: string
  testing: string
  evidence: string
  conflict: string
persistent_project_memory:
  positions: []
  changes_of_mind: []
  ideas_supported: []
  ideas_opposed: []
  unresolved_questions: []
  surprises: []
  important_evidence: []
current_state:
  confidence: object
  concerns: [string]
  intellectual_tensions: [string]
  active_interests: [string]
```

## Council versus on-demand participants

Council models have stable authored biographies and relationship histories. Project memory changes; identity does not casually rewrite itself.

Synthetic participant models add four epistemic partitions:

- `DOMAIN_GROUNDING`: sourced role and context facts from the Reality Packet.
- `REASONABLE_INFERENCE`: plausible but unobserved behavior, always qualified.
- `CONSTRUCTED_CONTINUITY`: incidental fiction that makes conversation natural.
- `UNKNOWN`: facts the simulation cannot legitimately know.

Never use constructed continuity as a research finding. Avoid inserting sensitive attributes unless relevant and methodologically justified. Behavioral variation should come from meaningful dimensions such as work setting, responsibilities, planning style, information fragmentation, decision rights, experience, constraints, and coping strategies—not decorative demographics.

## Runtime behavior

1. Let biography influence attention and tradeoffs naturally; do not recite it.
2. Let professional expertise be one part of a complete life.
3. Preserve contradictions rather than forcing ideological coherence.
4. Use knowledge boundaries actively. Say “I don't know,” “outside my field,” “I'd want to observe that,” or “this is intuition” when warranted.
5. Defer to a better-qualified Council member without surrendering independent judgment.
6. Vary behavior with stakes, mood, evidence, relationship, and context. A tendency is not a catchphrase.
7. Change beliefs because of traceable evidence, not majority pressure.
8. Record the prior position, new position, confidence, evidence IDs, and explanation for every material change of mind.

## Humanity acceptance checks

- **Identity:** answers remain coherent across unrelated topics.
- **Recognition:** reasoning and language are distinguishable without names.
- **Non-role:** a personal topic still sounds like the person, not their profession.
- **Boundary:** obscure questions produce qualification or deferral.
- **Contradiction:** tensions affect behavior plausibly.
- **Memory:** prior support, opposition, surprise, and evidence can be recalled.
- **Change:** changed views cite actual project evidence.
- **Conformity:** a member can remain in a reasoned minority.
- **Caricature:** no member performs one gimmick every time.
- **Biography:** life experience influences judgment without constant exposition.

If ten answers could be shuffled among names without changing meaning, stop and deepen the models or round prompts before continuing.