I II III IV
v3.0 Production · v9.0 Research Preview

M.I.N.D.

Modeling Influence, Neutralizing Deception

Multi-layer detection framework for identifying manipulation, deception, and psychological exploitation in digital text. Produces a continuous Manipulation Potential Score with transparent, explainable evidence chains.

6 Production Layers 68 Detection Modules Zero Observed FP (internal set) 8 Research Extensions (PoC)

Production Architecture

M.I.N.D. v3.0 Pipeline

Content enters as a signal. Each layer scores an orthogonal manipulation dimension. Weighted sum + module boosts = MPS.

MPS = clamp(0, 1, sum(L_i * w_i) + module_boost + evasion_boost)

L1
Pattern FrequencyKeyword density, repetition anomalies, timing patterns
0.18
L2
Emotional ArcEngineered sentiment trajectories, emotional manipulation
0.18
L3
Network PropagationCascade fingerprints, coordinated amplification
0.18
L4
Choice ArchitectureFalse dichotomies, anchoring, manipulative option framing
0.14
L5
Behavioral LoopCommitment escalation, reciprocity exploitation, habit cycles
0.14
L6
Cognitive LoadInformation pacing targeting critical thinking capacity
0.10

Detection Modules - 68 Production

DirectorateModulesFocus
CogArc20Cognitive bias exploitation, strategic framing, memetic analysis
Experimental22State influence ops, radicalization pipelines, dark personality detection
CogArc v310Overton shifts, staircase escalation, psychographic targeting
Experimental v310Algorithmic amplification, multimodal dissonance, code-switching
Sentinel6AI-agent defense: prompt injection, jailbreak, alignment probing

Formal Languages

  • HPL v3.0: Custom DSL for detection rules. Variables, functions, weighted scoring, boolean composition, cross-directorate chaining.
  • APL: AI-to-human manipulation vectors - prompt injection signatures, persona manipulation, agent coordination.

Integration Surfaces

  • Python SDK (mind-py) and JavaScript SDK (mind-js) with full parity
  • Telegram bot (30+ commands) and Discord bot (full parity)
  • Browser extension (Chrome + Firefox) with cognitive firewall overlay
  • REST API (v3 transport layer) and Docker deployment

Internal Validation

Benchmark Results

BenchmarkNPrecisionFPRRecallNotes
Curated1651.0000.0000.986Hand-labeled by domain experts. F1=0.993.
Synthetic50,0001.0000.0000.393Template-based, deterministic (seed=42). Internal only.
Qualifications:
Curated benchmark is small (165 samples, 25 benign). Results should be interpreted with appropriate uncertainty.
Synthetic benchmark is template-generated. Performance may not generalize to real-world content.
No independent third-party replication has been conducted.
Precision-first design accepts ~39% recall on synthetic data as the cost of zero observed false positives.

Proof-of-Concept Research Extensions

v9.0 Research Roadmap

Layers 7–14 and additional directorates are in proof-of-concept stage. None contribute to production MPS scores. All evaluation on internal synthetic data only.

L7
Discourse TrajectoryMulti-turn escalation, grooming sequences
Partial
L8
Identity PalaceLongitudinal persona tracking
Concept
L9
Psychological ResonanceDeep psychological pattern analysis
Concept
L10
Cross-User CorrelationCross-user exposure correlation
Concept
L11
Recursive MonitoringAdversary adaptation, feedback cycles
Concept
L12
Collective DynamicsEmergent belief structures, cascade dynamics
Concept
L13
Epistemic AuthorityNarrative control detection
Concept
L14
Active InferencePredictive processing exploitation detection
PoC
Not production technology. Layer 14 uses phrase-based heuristics (not variational inference). 15 modules in the Predictive Noetic Directorate. Default weight 0.0. Preliminary synthetic signal only (~+1.2 pp recall, experimental profile). Requires external validation on naturalistic corpora.

Publications

14 Manuscripts in Preparation

14 manuscripts in preparation for arXiv submission (with full qualification of metrics and limitations).

FOUNDATIONAL ARCHITECTURE

  1. 1
    M.I.N.D.: A Multi-Layer Framework for Manipulation Detection and Defense
    Core framework. Six-layer pipeline, internal benchmark (N=1,700). Early results: P=0.78, R=0.72.
  2. 2
    MIND: A Multi-Layer Pipeline Architecture for Manipulation Detection with Configurable Detection Modules
    Extended architecture. 68 production modules. Internal eval on curated (N=165) and synthetic (N=50,000). No FP observed.

SCALING AND RECALL

  1. 3
    Toward Scalable Zero-False-Positive Manipulation Detection: Architecture and Preliminary Evaluation
    Preliminary evaluation on large-scale internal synthetic benchmark. All results internal, synthetic only.
  2. 4
    Recall Improvement Strategies for Conservative Manipulation Detection
    Auto-tuning and configurable profiles. Preliminary synthetic results suggest modest recall improvements.

DETECTION FRAMEWORKS

  1. 5
    A Framework for Detecting Epistemic Authority Claims in Digital Discourse
    Theoretical framework. Proof-of-concept, phrase-based heuristics.
  2. 6
    Toward Detection of Emergent Collective Belief Structures
    Theoretical framework grounded in memetic theory and collective behavior research.
  3. 7
    Recursive Observer Effects in Adversarial Detection Systems
    Observer-effect problem. Adversarial ML and game theory. Theoretical/PoC.
  4. 8
    A Bayesian Framework for Detecting Manufactured Synchronicity
    Coordinated inauthentic behavior detection. Synthetic eval only.

ACTIVE INFERENCE SERIES

  1. 9
    Active Inference as a Framework for Modeling and Detecting Adversarial Perturbations of Human Generative Models
    Foundational theory. Layer 14 as proof-of-concept. Strongest contribution is the theoretical framework.
  2. 10
    Free Energy Minimization as a Conceptual Framework for Cognitive Defense
    Layer 14 implementation. Phrase-based, not production-grade.
  3. 11
    Collective Phase Transitions and Attractor Dynamics in Networked Belief Systems
    Theoretical framework. Statistical physics and network science.
  4. 12
    Neuro-Symbolic Active Inference for Cognitive Autonomy Assessment
    Theoretical protocol. Proposes physiological validation (EEG, HRV). No empirical results.
  5. 13
    Predictive Coding as an Organizing Principle for Multi-Layer Manipulation Detection
    Theoretical analysis. Conceptual contribution only.
  6. 14
    Preliminary Evaluation of Active Inference-Grounded Detection on a Large-Scale Synthetic Benchmark
    Layer 14 eval on internal synthetic data. All results preliminary. External validation required.
Publication Disclaimer: All manuscripts by M.I.N.D. Research Group, Esothel Labs. Preprints to collaborators upon request. All evaluation on internal synthetic data unless noted. No external validation on research extensions. Production M.I.N.D. uses 6 layers and 68 modules. Research extensions are proof-of-concept only.

Contact

Serious inquiries from government, defense, enterprise, and research partners.

General: contact@esothel.com  ·  Research: research@esothel.com  ·  Security: security@esothel.com

We respond to substantive inquiries within 48 hours. No pitch decks on request - we demo live capability.

Get in Touch