Help Center

FCM Basics

A Fuzzy Cognitive Map (FCM) is a directed graph used to model complex systems. It consists of:

• Concepts (nodes) — variables, factors, or ideas in your system
• Edges (connections) — directed, weighted relationships showing causal influence
• Weights — values between −1 and +1 indicating strength and direction of influence

How FCMs work

Each concept has an activation level. When you run a simulation, the system iterates: each concept's new value is computed from the weighted sum of incoming influences, passed through a transfer function (sigmoid, tanh, etc.). The system converges to a steady state, oscillates, or becomes chaotic.

When to use FCMs

FCMs are ideal for "soft" systems where relationships are known qualitatively but hard to quantify precisely: stakeholder mental models, policy analysis, ecological systems, healthcare decision-making, organizational dynamics.

Quick Start Guide

1. Join the beta waitlist at get-causality.com (closed beta — no billing yet)
2. Load a demo model from the Home tab to explore the interface
3. Open Model Builder to create your first FCM — add concepts, draw edges, set weights
4. Run analysis from the Model Analysis tab — view metrics, simulate scenarios
5. Explore tutorials in the Tutorial tab for in-depth guidance

Keyboard shortcuts

• Ctrl+Z — Undo last action in Model Builder
• Ctrl+Shift+Z — Redo
• Delete — Remove selected element
• Ctrl+S — Save current model
• Ctrl+A — Select all elements

Account & Plans

Available plans

All plans include unlimited analyses within their feature set. Annual billing available. Academic tiers (Academic, Academic Plus, Academic Pro) require .edu email verification.

Academic tiers

• Free — Model Builder, single-model analysis. 1 stored survey, 100 responses, 50 concepts/model. Community support.
• Academic — Adds Multi-Model analysis, full survey platform, goal-seek optimizer. 10 stored surveys, 100 concepts/model. 48h support.
• Academic Plus — Adds Meta-Model analysis (25+ population-scale analyses). 25 stored surveys, unlimited concepts. 24h support.
• Academic Pro — Adds Predictive Cognition (deep learning forecasting). 50 stored surveys. Priority 24h support.

Commercial tiers

• Researcher — Multi-Model analysis with commercial licensing. 10 stored surveys, unlimited concepts. No .edu required. Priority 24h support.
• Professional — Multi-Model, Meta-Model, Predictive Cognition, A/B testing, commercial license. 50 stored surveys, 3 team members. 24h support.
• Team — Everything in Professional + 100 stored surveys, 10 team members. Priority 12h support.
• Enterprise (Custom) — Unlimited surveys and team members, SSO/SAML, dedicated support, custom deployment. Contact us.

Key differences

• Commercial use — Researcher, Professional, Team, Enterprise plans include commercial license. Academic tiers are educational use only.
• A/B testing — Available on Professional, Team, Enterprise plans.
• SSO/SAML — Enterprise only.
• Team collaboration — Professional (3), Team (10), Enterprise (unlimited). Other plans are single-user.

Managing your account

Visit the account page to view your current plan, usage statistics, and manage billing. Full feature comparison on the pricing page.

Model Builder

The Model Builder is your primary workspace for creating and editing FCMs.

Adding concepts

Click "Add Concept" in the toolbar or double-click the canvas. Enter a name and optional description. Concepts appear as nodes you can drag to rearrange.

Drawing edges

Click the edge tool, then click a source concept and drag to the target. Set the weight (−1 to +1) in the dialog. Positive weights indicate positive causal influence; negative indicate inhibition.

Importing models

Import models from CSV, JSON, or adjacency matrix formats. Use the Import button. The builder supports multiple matrix formats and auto-detects concept names from headers.

Saving & exporting

Models auto-save to your browser's local storage. Use "Save to Cloud" to persist across devices (requires login). Export as JSON, CSV adjacency matrix, or image.

Model Analysis

Analyze a single FCM model's structure and dynamics.

Network metrics

Density, complexity, hierarchy index, concept count. Centrality analysis shows which concepts are most influential (outdegree), most influenced (indegree), or most central (betweenness).

Simulation

Run forward inference to see how the system evolves from an initial state. Choose transfer functions (sigmoid, tanh, bivalent, trivalent) and set iteration limits. Visualize convergence with state-vector plots.

Scenario analysis

Clamp one or more concept values and observe steady-state impact on the rest of the system. Compare "what-if" scenarios side-by-side.

Goal-Seek optimizer

Specify a target value for a concept and let the optimizer find minimal interventions needed. Uses differential evolution to handle nonlinear FCM dynamics. Available on Academic plans and above.

Multi-Model Analysis

Compare and combine FCMs from multiple stakeholders. Requires Academic plan or higher.

Importing multiple models

Upload a multi-model dataset (CSV with respondent IDs) or combine individual models. Aggregation uses non-zero averaging — zeros mean "not asked", not "no relationship".

Core analyses

• Concept Frequency — How often each concept appears across models
• Edge Frequency — Common relationships across stakeholders
• Combined Model — Aggregated FCM from all stakeholders using non-zero averaging
• Concept Selection — Which concepts respondents selected and why

Statistical comparisons

• QAP — Test structural similarity between FCM matrices
• Mantel tests — Correlate distance matrices across models
• Permutation tests — Non-parametric significance testing for group differences
• Bootstrap CI — Robust estimates and stability for edge weights
• Centrality comparison — Compare concept importance across groups or models

Structural analysis

• Cluster analysis — Within-model and across-model clustering
• Structural equivalence — Concepts in similar structural positions
• Motif analysis — Detect recurring sub-graph patterns
• Network dynamics — System behavior and stability
• FCM loop influence — Quantify feedback loop impact

Discovery & harmonization

• Relationship Discovery — Top discoveries, heatmaps, hidden patterns, PCA, gaps, feedback loops, exhaustive pattern mining
• Concept Harmonization — Align concepts across models with lexical similarity and concept mapping
• Cross-Tabulation — Group differences by demographics with chi-square testing

Meta-Model Analysis

20+ population-scale analyses across four assessment phases. Requires Academic Plus or higher. All analyses run entirely in your browser via Web Workers and in-browser WebAssembly (Pyodide) — no data leaves your machine.

Phase 1 · Data Assessment

• Data Quality — Automated screening of completion, variance, straight-lining
• Diagnostics — Distribution stats, correlation structures, missing-data patterns
• Validation — Reliability (Cronbach's α, test-retest), outliers, multicollinearity, normality, construct validity

Phase 2 · Descriptive & Statistical

• Concept Preference — empirical-Bayes (Kelley) shrinkage, rank weighting, bootstrap CIs
• Discrete Choice — Concept selection patterns with grouping variable support
• Network Meta-Analysis — Co-occurrence and correlation networks (Jaccard, Pearson, association rules)
• Comparative Analysis — Subgroup differences via Welch's t-test or Kruskal-Wallis ANOVA
• FCM Metric Clustering — k-means / hierarchical clustering with silhouette analysis
• Latent Class Analysis — Finite mixture models with EM and BIC/AIC selection

Phase 3 · Causal & Effects

• Total Causal Effects (TCEC) — Influence between concept pairs through all paths
• Causal Discovery — PC, GES, LiNGAM, FCI to validate respondent-drawn structure
• Bayesian Uncertainty — Posteriors on edge weights with credible intervals
• Monte Carlo Sensitivity — Confidence bands on steady states; rank edge sensitivity

Phase 4 · Machine Learning & Deep Learning

• Respondent Profiling — Random Forest, Gradient Boosting, XGBoost with feature importance
• SHAP Explainability — Per-respondent attribution of classification decisions
• GraphSAGE — Inductive concept embeddings for community detection and similarity
• Graph Transformer — Multi-head self-attention for context-dependent edge importance
• RL Policy Optimization — Q-learning for intervention strategy

Predictive Cognition

Train a conditional GAN entirely in your browser (via Pyodide / WebAssembly — no data leaves your device) to learn how causal mental models vary across your population: generate synthetic survey responses for any demographic subgroup, equalize uneven subgroups for fair comparison, and validate synthetic-vs-real fidelity.

LLM FCM Builder

Extract draft causal models from research papers, literature reviews, or interview transcripts using a language model that runs locally in your browser (or your own Ollama) — no respondent data is sent to a third-party API. Every extracted relationship is grounded in a verbatim source quote you can verify.

Surveys

Create and deploy FCM surveys directly from the platform. Requires Academic plan or higher.

Question types

30+ question types across categories: Input, Selection, Advanced, Research methods (MaxDiff, conjoint), FCM-specific (relationship matrix, concept selection), Layout.

Logic & branching

• Display logic — show or hide questions based on previous answers
• Skip logic — route respondents to different pages
• Question piping — insert previous answers into question text
• Carry-forward — populate options from earlier questions
• Loop / merge — repeat question sets for each selected item
• Randomization — randomize answer order to reduce bias

Distribution

• Direct link / email / SMS / QR / embed code
• Panel verification with PIN codes and single-use links

Quality & compliance

• Attention checks, CAPTCHA, GDPR Art. 9 consent, data classification, password protection, quotas

Response analytics

• Dashboard, individual responses, cross-tabulation, MaxDiff, conjoint utilities, text analytics, CSV/JSON export

Figure Editor

Create publication-ready figures from your FCM visualizations. Adjust node colors, sizes, labels; change edge styles, thickness, curvature; apply automatic layouts or position manually; add titles, legends, annotations.

Export formats

• PNG — high-resolution raster for presentations and documents
• SVG — scalable vector for journals and further editing
• PDF — print-ready with embedded vectors

Data & Security

Encryption

All PII is encrypted at rest using AES-256. Email addresses are stored as hashes for lookup, never plaintext. Session tokens use secure, HttpOnly cookies.

Privacy

Meta-Model and deep-learning analyses run entirely in your browser. Survey response data is encrypted before database storage. We support GDPR Art. 9 consent enforcement for special-category data.

Authentication

Email/password, Google OAuth, GitHub OAuth. Enterprise plans support SAML SSO. All sessions are rate-limited with concurrent session caps.

Frequently Asked Questions