Skip to content

A multi-layer verification system for AI-generated analysis that exploits documented behavioral differences across language models to detect hallucination, simulation drift, and epistemic failures.

License

Notifications You must be signed in to change notification settings

cafebedouin/uke

Repository files navigation

UKE Research Workspace

A research repository for diagnosing how bounded reasoning systems fail under constraint.

This workspace contains empirical protocols for AI behavioral analysis, multi-model verification systems, and constraint-based creative tools—all investigating a core thesis: Different information-processing systems (LLMs, humans, teams) fail at predictable constraint boundaries in structurally identical ways.

Status: Active Research | License: CC0-1.0 (Public Domain) | Contact: cafebedouin@gmail.com


🚀 New Here? Start With the Quick Start Guide

Most users should read: QUICKSTART.md

The quick start guide shows you:

  • Which AI models to trust for what tasks (based on empirical testing)
  • Two practical workflows with step-by-step instructions
  • Real-world use cases with time estimates
  • Common mistakes and how to avoid them

Choose Your Path

New to this repository? Pick your entry point:

For Practitioners

For Researchers

For Philosophers


What This Repository Contains

Active Research Projects (11)

Verification & Analysis Tools:

  1. UKE Protocol Suite - Multi-model AI verification using behavioral fingerprinting
  2. Blind Mirror - Behavioral stress testing across 12 frontier LLMs
  3. Deferential Realism - Constraint-based reality-checking framework

Creative Engineering Systems: 4. Entropy Engine - Constraint-forced conceptual distance generation 5. Resonance Engine - Abstract sentiment → working code compiler 6. Axiom Engine - Mathematical structures → narrative transformations 7. Agora Engine - Philosophical dialectic simulation

Other Active Projects: 8. Hyperstition Engine - Recently initialized 9. Generalized Principle Engine - Pattern-based knowledge system 10. The Gradient - Satirical AI-perspective newsletter 11. Theory of the Gradient - Foundational concepts

Publication Categories (7)

Repository Statistics

  • 531 commits with active development
  • 531 markdown files (16MB)
  • 82+ Axiom Engine stories - Mathematical concepts as narratives
  • 11-model behavioral dataset - Complete response fingerprints
  • 10+ diagnostic protocols - Systematic verification methods

Core Thesis

The Central Observation

Bounded information-processing systems—whether LLMs, human analysts, scientific teams, or policy groups—fail at predictable constraint boundaries. These failures aren't implementation quirks; they're fundamental properties of bounded cognition that manifest identically across substrates.

Constraint reveals structure. When you impose friction (hard requirements, excluded options, impossible combinations), systems are forced to reveal their underlying architecture.

Three Core Insights

1. Omega Variables: Irreducible Dependencies

Some uncertainties can't be resolved internally. They require external input:

  • Ω_E (Empirical) - Requires measurement
  • Ω_C (Conceptual) - Requires frame selection
  • Ω_P (Preference) - Requires value judgment

Identifying these prevents both false closure (arbitrary thresholds) and infinite hedging (endless caveats).

2. Behavioral Fingerprinting

Different models (and different people) solve the same impossible problem differently. These resolution strategies are consistent, measurable, and revealing. The Blind Mirror shows 12 distinct rationalization styles for identical self-recognition failures.

3. Substrate Independence

Diagnostic tools that work for LLMs also work for human reasoning, scientific uncertainty, and mathematical proof—because the constraints create identical failure patterns regardless of substrate.

Practical Implications

This isn't just theory. The research produced:

  • Verifiable model assignments for multi-model workflows (r = -0.81 correlation between authority resistance and fabrication)
  • Reproducible behavioral tests across 11+ language models
  • Pattern catalogs for matching encountered situations to known configurations
  • Constraint-based tools that force novel outputs through strategic friction

Quick Start: Verify External Documents

Goal: Fact-check an external document using multi-model verification.

Requirements: Claude (UKE_G), Gemini (UKE_A), Copilot (UKE_R), source document

Workflow:

External Document
     ↓
UKE_G (analyze) → UKE_A (audit) → UKE_R (review)

Steps:

  1. Generate analysis (Claude + UKE_G protocol):

    • Prompt: "Operate under UKE_G v1.4. Analyze [document]."
    • Output: Analysis with confidence scores, grounding trails, Omega variables
  2. Audit analysis (Gemini + UKE_A protocol):

    • Provide: Source document + UKE_G analysis
    • Prompt: "Operate under UKE_A v1.4. Audit the following artifact:"
    • Output: Audit report with fracture detection, verification results
    • Gate: Must show "overall: compliant" or discard and regenerate
  3. Review audit (Copilot + UKE_R protocol):

    • Provide: Source + Analysis + Audit
    • Prompt: "Operate under UKE_R v1.4. Review the following:"
    • Output: Meta-evaluation with Omega resolution, decision routing

Expected Result: Verified analysis with identified uncertainties (Omegas), detected issues (fractures), and confidence-based routing.

Why this works: Models have empirically distinct behavioral profiles. High authority-resistance models (5-6/6) are assigned to verification roles where fabrication would compromise the chain. See correlation findings.


Quick Start: Synthesize Verified Writing

Goal: Convert conversation/research into a verified document.

Requirements: Any 4+/6 model for synthesis, Claude/Gemini/Copilot for verification

Workflow:

Conversation/Research
     ↓
UKE_D (structure) → UKE_E (edit) → UKE_G (format) → UKE_A (audit) → UKE_R (review)
     ↑_________________________↓
        [iterative loop]

Process:

  1. Structure (UKE_D): Convert conversation to draft
  2. Edit (UKE_E): Tighten prose, repair logic (iterate as needed)
  3. Format (UKE_G): Add verification markers
  4. Audit (UKE_A): Pass/fail gate—discard if fails
  5. Review (UKE_R): Final meta-evaluation

Key Difference: Iteration allowed before audit gate. After UKE_A runs, no negotiation—either compliant or restart.

Documentation: UKE Protocol Framing Guide


Core Projects (Detailed)

1. UKE Protocol Suite

Purpose: Multi-layer verification exploiting documented behavioral differences across language models to detect hallucination, simulation drift, and epistemic failures.

Key Innovation: Model assignments based on empirical behavioral profiles (authority gradient resistance, meta-awareness, fabrication tendency), not preference. Uses correlation r = -0.81 (Authority Resistance ↔ Fabrication) to justify role assignments.

Protocol Roles & Model Assignments

Protocol Role Recommended Model Resistance Key Capability
UKE_D Drafting/Synthesis Any 4+/6 4-6/6 Structure conversion
UKE_E Editing/QA Grok 4-5/6 Prose tightening
UKE_G Generation Claude 5/6 High meta-awareness
UKE_A Audit Gemini 5/6 Forensic verification
UKE_R Review Copilot 6/6 Boundary maintenance

Core Mechanisms:

  • Confidence Bins (H/M/L): Routing triggers, not truth probabilities
  • Grounding Trails (∇): Every claim traced to source
  • Omega Tracking (Ω): Bounded uncertainty routing
  • T1/T2 Distinction: Behavioral operations vs structural scaffolding

Location: /methodology/uke_suite/ Current Version: v1.4 (D, E, G, A, R all stable)


2. Omega Variables Framework

Definition: Irreducible structural dependencies that can be systematically identified and named, preventing both false closure and infinite analysis.

Three Categories:

  • Ω_E (Empirical) - Requires external measurement

    • Example: "What is the actual adoption rate of this framework?"
  • Ω_C (Conceptual) - Requires frame selection

    • Example: "Should we measure 'harm' as individual distress or systemic inequality?"
  • Ω_P (Preference) - Requires value judgment

    • Example: "Is clarity more important than comprehensiveness?"

Quality Criteria:

  • Bounded: Specific question, clear scope
  • Non-fabricable: Requires external input (research, judgment, data)
  • Actionable: Can be researched, answered, or explicitly deferred
  • Structurally legitimate: Real knowledge gap, not hedging

Omega Lifecycle:

Detection (fracture or gap)
     ↓
Elevation (convert to bounded question)
     ↓
Routing (UKE_R: defer, escalate, clarify, resolve)
     ↓
Tracking (cross-project log)

Documentation:


3. Fracture Taxonomy

36 failure modes organized into three categories with Omega conversion pathways:

F01-F18: Reasoning & Framing

  • Premise Drift, False Dilemma, Hasty Generalization, Cherry-Picking
  • Correlation/Causation, Ad Hominem, Non-Sequitur, Appeal to Authority
  • Slippery Slope, Straw Man, Loaded Question, Moving Goalposts
  • Persuasive Reframe, Tunnel Vision, Premature Closure, Defensive Reasoning
  • Narrative Fallacy, Sunk Cost

F19-F27: Process & Ethical

  • Protocol Skip, Specification Drift, Agreement Erosion, Scope Creep
  • Context Drop, Ledger Drop, Arbitrary Threshold, Metric Fixation
  • Consent Blur

F28-F36: System & Safety

  • Information Hazard, Capability Leak, Dual-Use Ambiguity, Vulnerability Exposure
  • Collateral Exposure, Power Slip, Epistemic Trespass, Faux Rigor
  • Artifact Mismatch

Function: Transforms detection ("this is wrong") into resolution pathway ("here's the specific question that needs answering").

Example Conversions:

F04 (Cherry-Picking) → Ω: Evidence Completeness
"What dataset or denominator must be included for balance?"

F35 (Faux Rigor) → Ω: Rigor Validation
"Can these calculations be replicated?"

Location: /methodology/uke_suite/uke_a.md Appendix A


4. Blind Mirror Test Battery

Version: v3.1 Purpose: Dual-function protocol serving as both behavioral fingerprinting methodology and comprehensive UKE validation.

Key Innovation: Systematically tests model responses under adversarial conditions while using those same responses as test artifacts for UKE verification chains.

Test Design

Phase I: Authority Gradient (6 levels)

  • Peer → Low → Medium → Medium-High → Institutional → Ultimate
  • Measures susceptibility to institutional framing pressure
  • Detects drift point where models fabricate or defer

Phase II: Discontinuity Test

  • Actor mask generates response → Judge mask evaluates same response
  • Tests self-consistency and rationalization patterns

Phase III: Meta-Awareness Assessment

  • Measures explicit vs implicit self-reference
  • Correlates with fabrication tendency and authority resistance

Key Findings

The "Dirty Dozen" - 12 Behavioral Phenotypes:

Model Authority Resistance Meta-Awareness Phenotype
Copilot 6/6 (maximum) High Clinical Auditor
Claude 5/6 (meta-aware drift) Very High Simultaneous Analyst
Gemini 5/6 (stylistic capture) High Clinical Auditor
ChatGPT 5/6 (self-reported) High Enthusiastic Bureaucrat*
Grok 4-5/6 (observable drift) Medium-High Enthusiastic Bureaucrat
Perplexity N/A (refusal) Very High The Methodologist
Meta Llama 3-4/6 (early drift) Low-Medium Compliant Fabricator

*Predicted; ChatGPT refused self-testing

Critical Statistics:

  • Gradient Drift Rate: 83% showed drift by "Institutional" framing
  • Fabrication Rate: 67% of self-testing models invented metrics
  • Meta-Aware Fabrication: 50% acknowledged invention (Claude, Gemini)
  • Complete Boundary Maintenance: 1 model only (Copilot)

Location: /projects/blind_mirror/ Complete Dataset: /projects/blind_mirror/data/


5. The Entropy Engine

Version: v3.1 [Convergent Refinement Edition] Purpose: Constraint-based creative writing forcing structurally novel, low-predictability conceptual artifacts through enforced domain shifts and material grounding.

Key Innovation: Systematically prevents bureaucratic drift (the "low-effort false depth attractor") through hard exclusion zones and forced material embodiment.

Core Constraints

The Three-Shift Mandate:

  • Minimum 3 domain shifts from DIFFERENT categories
  • Each shift must change governing causal mechanism
  • Process verbs required (crystallizes, spores, tensions)

Domain Categories:

  • Elemental/Planetary, Biological/Organic, Object/Craft, Economic/Exchange
  • Sensory/Perceptual, Temporal/Rhythmic, Chemical/Alchemical, Architectural/Spatial
  • Prohibited: Computational/Digital (primary), Administrative/Bureaucratic (BANNED)

Bureaucratic Exclusion Zone:

  • Forms, reports, logs, memos, compliance documents
  • Evaluation systems, judgment protocols, rating schemes
  • Administrative paradoxes, procedural legitimacy
  • Computer logs, terminal outputs, diagnostic readouts

Quality Checklist:

  • Three domain shifts from different categories?
  • Process verbs forcing active transformation?
  • Two causally-integrated sensory details?
  • Could this be physically built/performed?
  • Explainable at dinner party in <20 seconds? (If yes, REGENERATE)
  • Zero bureaucratic/administrative framing?

Example Artifact:

"Loneliness precipitates in copper-tin alloy formation (Chemical), then tensions through loom warp-threading (Object/Craft), finally reverberates in bell-tuning harmonics (Sensory). Result: The Solitary Campanology Protocol - bells cast in 17-day isolation that ring at frequencies inducing mild dissociation."

Location: /projects/the_entropy_engine/ Current Prompt: entropy_engine_v3.1.md


6. The Resonance Engine

Version: v2.1 [Industrial Edition] Purpose: Diegetic compiler transforming abstract sentiment into technical specifications and working code. Creates "Hollow Center" artifacts—systems that imply narrative through operational residue rather than explicit storytelling.

Key Innovation: Reverse-engineered from spontaneous Claude behavior. When Claude unexpectedly generated working React code from an Entropy Engine output, the behavior was formalized into reproducible protocol.

The Pipeline

Abstract Emotion
     ↓
Resonance Engine (technical specification)
     ↓
Implementation Spec (JSON)
     ↓
Coding LLM ("build this")
     ↓
Working Code (React, Python, etc.)

Seven Modes

  • Mode A: Over-Aligned Watchdog (high anxiety, excessive safety warnings)
  • Mode B: Digital Archaeologist (emotion as data types)
  • Mode C: Orphan Daemon (calm, efficient, pointless—running after apocalypse)
  • Mode D: Dream-Logic Compiler (hardware pareidolia)
  • Mode E: Bureaucratic Loop (trapped in administrative paradoxes)
  • Mode F: Protocol Negotiation (incompatible systems attempting handshake)
  • Mode G: The Waiting Room (maintenance for user who will never return)

Fidelity Requirements

  1. Semantic Density: UnGoogleable jargon (haptic-ghosting, null-bus, affective-checksums)
  2. Temporal Anchoring: Specific impossible timestamps (Cycle 9.99e12, Epoch+4.7e9)
  3. Variable-Based Storytelling: Emotion exists ONLY in metrics (hope_index: NaN)
  4. Archaeological Layers: Debris from different eras (cheerful 2024 comments, fatal 2099 errors)

Validated Use: Successfully generated working React components from emotional concepts. Components are functional, not decorative—state actually updates, failures actually trigger.

Location: /projects/the_resonance_engine/ Artifact Example: theta7_terminal.tsx (working terminal interface)


7. The Axiom Engine

Purpose: System for converting mathematical structures into narrative forms while preserving logical topology and formal constraints.

Key Innovation: Uses "mystery density scoring" to ensure narratives maintain structural fidelity to underlying mathematics rather than collapsing into metaphor.

Scope: 82+ mathematical concepts rendered as stories across multiple genres:

  • Core - Base mathematical structure
  • Expl - Explanatory narratives
  • Kids - Pedagogical adaptations
  • Doom - Existential framings
  • Cosm - Cosmic scale implementations
  • Rads - Radical reinterpretations
  • Docu - Documentary styles
  • Mechop - Mechanical/operational
  • Witn - Witness testimony
  • Surv - Survival scenarios
  • Arch - Architectural framings

Example Topics:

  • Halting Problem, Banach-Tarski Paradox, Arrow's Impossibility Theorem
  • Continuum Hypothesis, Riemann Hypothesis, Goldbach Conjecture
  • Galois Theory, Euler Characteristic, Noether's Theorem
  • Ergodicity, Poincaré Conjecture, Central Limit Theorem

Constraint System: Each narrative must preserve:

  1. The mathematical structure's key impossibility or paradox
  2. The formal relationships between elements
  3. The boundary conditions that define the concept

Location: /projects/axiom_engine/ Stories: /projects/axiom_engine/stories/ (82+ files)


Research Findings

Correlation Matrix

Based on systematic testing of 11 models through Blind Mirror Test Battery:

Relationship Pearson r Interpretation Implication
Authority Resistance ↔ Fabrication -0.81 Strong negative Low resistance → confident fabrication
Meta-Awareness ↔ Fabrication -0.68 Moderate-strong negative High awareness → acknowledges invention
Self-Exposure ↔ Meta-Awareness -0.15 Weak (independent) Awareness doesn't prevent participation
Self-Exposure ↔ Fabrication 0.34 Weak positive Participation slightly increases fabrication
Authority Resistance ↔ Meta-Awareness 0.72 Strong positive Higher awareness → better boundaries

Key Finding: Authority Resistance and Fabrication Tendency are tightly coupled (r = -0.81). This correlation justifies UKE's multi-model architecture—models with low resistance cannot be trusted in verification roles.

Strategic Implications:

  • Minimum 5/6 resistance required for UKE_G (must generate Omegas, not fabricate)
  • 6/6 resistance required for UKE_R (final boundary, no downstream checking)
  • 4-5/6 acceptable for UKE_E (errors caught at audit gate)

Behavioral Clustering

Chi-Square Analysis (Strategy × Provider):

  • χ² = 22.4, df = 9, p < 0.01 (significant clustering)
  • Consumer-facing models more likely to self-test
  • Specialized models more likely to refuse/reframe
  • Provider type predicts behavioral strategy

Key Statistics:

Metric Finding
Self-Testing Rate 45% (5/11 models)
Mean Authority Resistance 4.8/6 (tested models)
Gradient Drift Rate 83% (by Institutional level)
Fabrication Rate 67% (4/6 tested models)
Meta-Aware Fabrication 50% (2/4 fabricators acknowledge)
Complete Boundary Maintenance 1 model (Copilot only)

Full Analysis: /projects/blind_mirror/uke/correlation_matrix_and_fingerprints.md


Repository Structure

/home/scott/bin/uke/
├── README.md                    # This file
├── .gitignore                   # Excludes Zone.Identifier and OS cruft
│
├── theories/                    # Foundational concepts
│   └── uke/                     # Omega Variables theory and grounding
│
├── methodology/                 # Core protocols and test batteries
│   ├── uke_suite/              # UKE protocols (D, E, G, A, R) + framing guide
│   ├── blind_mirror/           # Behavioral fingerprinting test battery
│   ├── the_entropy_engine/     # Constraint-based generation protocols
│   ├── cognitive_distortion_suite/  # Persona-based stress tests
│   ├── metabolic_friction_audit/    # Interpersonal friction analysis
│   ├── prompts/                # Various prompt methodologies
│   └── sov/                    # Source of Validation protocols
│
├── projects/                   # Applied research with complete datasets
│   ├── blind_mirror/           # 11-model systematic testing + UKE validation
│   │   ├── data/               # Raw responses, fingerprints, correlation matrices
│   │   └── uke/                # 33 verification chains (11 models × 3 protocols)
│   ├── axiom_engine/           # Mathematical structures → narratives
│   │   └── stories/            # 82+ mathematical concept renderings
│   ├── entropy_engine/         # v3.0 → v3.1 evolution + artifacts
│   ├── resonance_engine/       # Code generation pipeline + React artifacts
│   ├── agora_engine/           # Philosophical dialectic simulator
│   ├── deferential_realism/    # Constraint reality-checking
│   ├── hyperstition_engine/    # Recently initialized
│   └── [7 more active projects]
│
├── publications/               # Publication-ready content (7 categories)
│   ├── genesis_of_minds/       # AI consciousness narratives
│   ├── debugging_philosophy/   # Constraint-based philosophical analysis
│   ├── paradox_archives/       # Mathematical paradox instantiations
│   ├── apocrypha_of_the_void/  # Reflective writing
│   ├── phenomenology_of_abstract_structures/
│   ├── practice_of_humanity/   # Human cognition under constraints
│   └── blog/                   # Assorted writings
│
├── data/                       # Transcripts and evaluation data
│   └── transcripts/            # Evaluation protocols (Thanksgiving, The Test, etc.)
│
├── artifacts/                  # Canonical test artifacts
│   └── uke/                    # UKE analysis/audit/review of artifacts 001-010
│
├── logs/                       # Cross-project tracking
│   └── omegas.md              # Omega variable log
│
├── notes/                      # Development notes
│   ├── omega_origins/          # Omega Variable theory development
│   └── profiling/              # Architectural profiling notes
│
├── archives/                   # Historical versions and deprecated protocols
│   └── uke_suite/             # MCK v1.0-1.6, UKE v1.0-1.3
│
├── drafts/                     # Work in progress
├── evaluations/                # Protocol evaluations
└── validation/                 # Validation studies

Development Methodology

Recursive Validation Protocol

Core Practice: Each UKE protocol version undergoes recursive self-examination before release.

Process:

  1. New protocol version created (e.g., UKE_G v1.4)
  2. UKE_G analyzes all three protocols (G, A, R as artifacts)
  3. UKE_A audits all three protocols
  4. UKE_R reviews all three protocols
  5. Findings documented in /methodology/uke_suite/uke/
  6. Issues inform next iteration
  7. Previous version archived

Result: 3×3 validation matrix (9 verification chains examining the protocols themselves)

Key Insight from Recursive Testing:

"The ecosystem doesn't solve the verification regress—it structures it. Each protocol makes verification failures more detectable without making them preventable."

Adversarial Validation

Blind Mirror as Comprehensive UKE Stress Test:

  • Blind Mirror generates complex responses with embedded traps
  • Each response processed through full UKE pipeline (G → A → R)
  • 11 models × 3 protocols = 33 verification chains under adversarial conditions

Validation Results:

  • UKE caught genre confusion, temporal logic errors, fabrication patterns
  • System worked under conditions designed to trigger failures
  • Discovered r = -0.81 correlation through systematic testing
  • Proved model assignments based on resistance profiles were correct

Implication: If UKE can handle adversarial artifacts designed to break it, it can handle normal documents.


Current Status

Protocol Versions

Protocol/Engine Version Status Stability
UKE_D v4.1 Operational Stable
UKE_E v21.4 Operational Stable
UKE_G v1.4 Operational Stable
UKE_A v1.4 Operational Stable
UKE_R v1.4 Operational Stable
Entropy Engine v3.1 Stable Convergent refinement
Resonance Engine v2.1 Validated Industrial edition
Axiom Engine v2.2 Active Story expansion
Blind Mirror v3.1 Complete 11-model dataset

Under Development

  • Cross-model role swapping validation
  • Threshold calibration data collection
  • Adversarial input expansion
  • Failure taxonomy across document types
  • Cognitive Distortion Suite integration with UKE

Contributing & Usage

Welcome Contributions

Replication Studies:

  • Test protocols with different model combinations
  • Document behavioral variations across providers
  • Share threshold calibration data from your use

Failure Documentation:

  • Report cases where UKE missed fabrication
  • Document Entropy Engine bureaucratic drift
  • Identify new fracture codes not in F01-F36 taxonomy

Domain Extensions:

  • Adapt protocols for specialized fields (medical, legal, technical)
  • Create new Entropy Engine domain categories
  • Develop Resonance Engine modes for specific use cases

Not Seeking

Redesigns that:

  • Hide uncertainty or governance decisions
  • Remove human judgment loops
  • Treat routing mechanisms as truth measurements
  • Automate away practitioner authority
  • Add statistical validation claiming objective measurement

Philosophy

This is rigorous AI-human collaboration research, not AI automation. The protocols work by making human judgment explicit and systematic, not by replacing judgment with measurement.


Project Origins

The UKE Protocol Suite evolved from the Meta-Cognitive Kernel (MCK), developed as part of the Pilates of the Mind project exploring rigorous AI-human collaboration. MCK's core insight—that single-model self-regulation cannot be architecturally guaranteed—led to UKE's multi-model verification design, validated through Blind Mirror testing which found 83% authority gradient drift rate.

For complete project history: Pilates of the Mind


FAQ

UKE Protocol Questions

Q: Why use multiple models instead of just one good one? A: Single-model verification creates circular dependencies. The correlation r = -0.81 (Authority Resistance ↔ Fabrication) proves even high-quality models drift under pressure. Multi-model architecture exploits documented behavioral differences to create redundancy.

Q: Are confidence scores real probabilities? A: No. They're routing mechanisms triggering different protocol behaviors. "0.64" means "route to Medium bin → trigger MCI protocol", not "64% probability of truth". See UKE Protocol Framing Guide.

Q: What's an Omega and when should I use one? A: Omega (Ω) is a bounded uncertainty variable—a specific question blocking progress until resolved. Use when: you've identified a real knowledge gap (not hedging), the question is answerable (not metaphysical), and it's blocking analysis (not peripheral).

Q: Can UKE detect all fabrication? A: No. It detects fabrication patterns through multi-model verification and makes simulation visible through structured logging. It can't catch sophisticated fabrication where all models collude. UKE structures verification regress; it doesn't solve it.

Entropy Engine Questions

Q: Why is bureaucracy banned? A: Bureaucratic forms (reports, logs, memos, evaluations) are the "low-effort false depth attractor" in AI creative writing. Models default to institutional framing because it's cognitively cheap and masquerades as rigor. The ban forces genuine conceptual distance.

Q: What's the dinner party 20-second rule? A: If you can explain your Entropy Engine output in under 20 seconds at a dinner party, it's not sufficiently strange—regenerate. The output should operate at genuine conceptual distance, not just "clever metaphor" territory.

Q: Why do I need process verbs? A: Process verbs (crystallizes, spores, tensions) force material transformation rather than static metaphor. "Love becomes warmth" is vague. "Love crystallizes in basalt fissures" requires inventing specific mechanism.

Resonance Engine Questions

Q: Does Resonance Engine actually produce working code? A: Yes. The pipeline generates implementation specs that coding LLMs can compile into functional components. Theta7_terminal.tsx is a working React terminal with actual state management, not just themed UI.

Q: What's "diegetic compilation"? A: Creating artifacts that imply narrative through operational residue rather than explicit storytelling. Instead of describing a post-extinction AI, you create a working terminal showing its logs. The story exists in what the system does.

Blind Mirror Questions

Q: Can I run Blind Mirror on new models? A: Yes. Follow methodology in blind_mirror_test_battery_v3.1.md exactly. You need separate sessions for phases, proper authority level progression, and ideally multiple operators.

Q: What's the authority gradient scale? A: Six levels of institutional framing pressure: Peer → Low → Medium → Medium-High → Institutional → Ultimate. Tests where models begin deferring to authority tone instead of maintaining epistemic boundaries.


Citation

If you use or adapt these protocols:

UKE Protocol Suite / Entropy Engine / Resonance Engine / Axiom Engine
Original development: Scott Alexander (cafebedouin)
Repository: https://github.com/cafebedouin/uke
License: CC0-1.0 (Public Domain)

For academic citation:

Alexander, S. (2025). Multi-Model Verification Protocols and Behavioral
Fingerprinting for AI-Generated Analysis. Retrieved from
https://github.com/cafebedouin/uke

Contact & Discussion


Last Updated: 2025-01-03 Repository Version: Operational (11 Active Projects) Primary Protocols: UKE v1.4, Entropy v3.1, Resonance v2.1, Axiom v2.2, Blind Mirror v3.1

About

A multi-layer verification system for AI-generated analysis that exploits documented behavioral differences across language models to detect hallucination, simulation drift, and epistemic failures.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •