Skip to content

quick_start.md

Latest commit

 

History

History
96 lines (71 loc) · 2.75 KB

quick_start.md

File metadata and controls

96 lines (71 loc) · 2.75 KB

Quick Start

Prerequisites

  • SWI-Prolog (for Layer 2 validation)
  • Gemini API access (for Layer 1 generation, optional)
  • Claude API access (for Layer 3 synthesis, optional)

Running Layer 2 Audit Only

Test the validation system on existing models:

cd prolog
swipl

?- [validation_suite].
?- run_scenario('./testsets/columbia_2026_elections.pl', columbia_2026).

Output:

[CONSTRAINT INVENTORY: REALITY AUDIT]
  Constraint | Claimed | Actual | Action
  ----------------------------------------------------------------------
  c1_term_limit | mountain | mountain | navigate
  c2_runoff | rope | noose | replace
  c3_coalition | rope | tangled_rope | reform

[STRUCTURAL SIGNATURE ANALYSIS]
  c1_term_limit: natural_law (confidence: high)
    → No alternatives ever existed, universal inaccessibility
  
  c2_runoff: constructed_constraint (confidence: high)
    → Active enforcement detected (suppression=0.55, resistance=0.65)
    → Alternatives existed (single-round, ranked-choice)

[PERSPECTIVAL GAP ANALYSIS]
  c2_runoff:
    - Individual (powerless): mountain
    - Institutional (manager): rope
    → GAP: Same structure experienced as unchangeable by powerless,
            as coordination tool by powerful

Generating New Models (Layer 1)

Manual Process:

  1. Copy prompts/constraint_story_generation_prompt.md
  2. Provide domain description to Gemini: 
    Analyze the constraint topology of [domain description].
Generate a complete Prolog constraint story file.
  3. Save output as testsets/[domain_name].pl
  4. Run Layer 2 validation (see above)

What Layer 1 generates:

  • Base measurements (extractiveness 0.0-1.0, suppression 0.0-1.0)
  • 3+ indexed perspectives (different WHO/WHEN/WHERE/HOW MUCH)
  • Test suite proving indexical variance
  • Beneficiary/victim asymmetry markers
  • Model commentary explaining reasoning

Validation: Layer 2 auto-repairs syntax errors, checks schema compliance

Synthesizing Analysis (Layer 3)

Manual Process:

  1. Run Layer 2 audit, capture executive summary
  2. Perform web search for empirical evidence (Perplexity)
  3. Provide Claude with UKE_W protocol: 
    Execute UKE_W v1.0 protocol.

Substrate: [Layer 2 audit summary]
Evidence: [Web search results]
Voice: [System Architect | Critical Essayist]
Temperature: [hot | warm | cool]

Generate constraint story essay.
  4. Claude outputs essay with metadata blocks

What Layer 3 produces:

  • Evidence-backed narrative (all claims trace to substrate)
  • Collapsed Omegas (uncertainties → defensible assertions)
  • Falsifiable predictions (explicit success/failure criteria)
  • Stakes anchor (why this matters beyond being true)
  • Quality gate validation (substrate fidelity, counterfactuals, etc.)