1. Architecture (S7, four verbs, three classes) | 2. Ecosystem contracts | 2.1 medfit::MediationData (the fitting contract) | 2.2 RMediation (the inference contract) | 2.3 Dependency direction (no cycle) | 3. The MI estimator and Rubin's rules | Worked example: pooling three imputations | 3A. Treatment-by-mediator interaction (X:M) | 3A.1 Estimand | 3A.2 The four-way decomposition, pooled | 3A.3 What stays refused | 4. MBCO under MI: D4-stacking (and why it is hosted here) | 4.1 Why MBCO does not commute with Rubin's rules | 4.2 D4 combination | Worked example: pooling three MBCO statistics | 4.3 Hosting decision | 5. The IPW estimator | 5.1 Pipeline | 5.2 Weight estimation | Worked example: four weights, stabilized and trimmed | 5.3 Variance: sandwich vs model | 5A. Non-gaussian models (family_y, family_m) and the scale of a*b | 5A.1 The scale caveat | 5A.2 IPW plus a binomial family: the non-integer-successes warning | 5B. MNAR sensitivity (sensitivity_mnar()) | 5B.1 Method | 5B.2 What delta actually is — read before choosing a value | 5B.3 Reading a tipping point | 5B.4 Binary targets: when is a log-odds delta of 1 plausible? | 5B.5 Limitations | 6. Cross-package engineering decisions | 7. Design-choice summary | 7A. Self-check | 7A.1 Three common confusions | 8. References

What you get | Before you impute: the imputation model | Worked example | The interval | The test | The result object | Options and edge cases | ariv = "fixed" or "own" | Without the pipeline: mbco_d4() | A single imputation | When (k) differs across imputations | MBCO on a pooled object | Models with an X:M interaction | Why MBCO does not commute with Rubin's rules | Operating characteristics | Not covered here | References

A worked example | 1. Specify the mediation model | 2. Fit each imputation | 3. Pool with Rubin's rules | 4. Inference on the indirect effect | A binary mediator (or outcome) | Structural equation models and latent mediators | Migrating from the S4 API

Introduction | Installation | Loading the Ecosystem | Package Management | Check Installed Versions | Update Packages | Check for Conflicts | Core Workflow | 1. Fit Models | 2. Extract Mediation Structure | 3. Choose Your Analysis | Confidence Intervals (RMediation) | Probabilistic Effect Size (probmed) | Sensitivity Analysis (medrobust) | Bootstrap Inference (medfit) | Ecosystem Design | Foundation Package | Specialized Packages | Type Safety | Next Steps | Session Info

What the ecosystem is | Packages | Live status | Roadmap & planning | Get involved

Part 1: the core analysis | The data | Step 1: the complete-data reference | Step 2: make the mediator missing at random | Step 3: impute the mediator | Step 4: a pooled Monte Carlo interval | Step 5: test the indirect effect with D4-MBCO | Step 6: when both paths are weak | Step 7: how much does the conclusion depend on MAR? | Part 2: new since missingmed 0.5.0 | The model is checked before it is fitted | Arguments that do not apply are reported | A treatment-by-mediator interaction | Sensitivity analysis with the MBCO test | Summary

Why two tiers | The structural reason FORK-RNG can only be Tier B | The self-contained synthetic study | Running Tier B | B2 — end-to-end coverage (the real number) | Combining runs under an integrity audit | B3 — the can-fail dogfood (planted defect) | B4 — many-core FORK RNG realism (the sole FORK guard) | B5 — production-grid seed-collision check | Pilot before scale | Summary

Introduction | Installation | Step 1: Generate Nonnormal Data | Step 2: Insert Missingness | Step 3: Define a Missing-Data Scenario | Step 4: Run Both Estimators | Step 5: Branch-Switch Diagnostics | Optional: Inverse-Probability Weighting | Next Steps

Introduction | Sobol Variance-Share Proportion Mediated | Gauge-Residual Estimand | Comparing the Two | Next Steps

Introduction | Overview of Methods | Method Comparison | Performance and Accuracy Comparison | When to Use Each Method | Distribution of Product (DOP) | Monte Carlo (MC) | Asymptotic Normal | Working with S7 Classes | Practical Recommendations | Conclusion

Introduction | Basic Usage | Single Mediator Analysis | Multiple Mediators | Three-Variable Indirect Effects (ProductNormal3) | Using S7 Classes | Available Methods | Integration with Structural Equation Modeling | Validation and Error Handling | Conclusion

Why this vignette | A serial chain from a lavaan fit | The same chain from lm/glm fits | Graceful degradation without medfit | See also

The problem with a single proportion-mediated number | A clean (additive) example | An interaction breaks the split | Non-binary-coded exposures: a0 / a1 | Inference for a skewed ratio | Summary

Overview | Setup | Step 1: Simulate Data | Step 2: Fit Mediation Models | Step 3: Extract Mediation Structure | Step 4: Confidence Intervals (RMediation) | Distribution of Product Method | Monte Carlo Method | Compare Methods | Step 5: Probabilistic Effect Size (probmed) | Step 6: Sensitivity Analysis (medrobust) | Step 7: Bootstrap Inference (medfit) | Complete Results Summary | Interpretation | Using SEM with lavaan | Session Information | References

Introduction | Installation | Quick Start | Step 1: Define Your Method | Step 2: Configure the Simulation | Step 3: Select Scenarios | Step 4: Run the Simulation | Step 5: Analyze Results | Execution Modes | Custom Scenarios | HPC Cluster Support | Next Steps

Overview | The joint estimand | A worked example | Plugin (point estimate) | Parametric bootstrap | Nonparametric bootstrap | MBCO (deterministic) | Scope and limitations

Overview | What is $P_{med}$? | Installation | Basic Example: Linear Mediation | 1. Simulate Data | 2. Estimate $P_{med}$ | 3. Interpret Results | MBCO intervals (deterministic) | Advanced Example: Binary Outcome (GLM) | 1. Simulate Binary Data | 2. Estimate with GLM | Conclusion

Feature Comparison | Workflow 1: Base R (lm/glm) | Workflow 2: lavaan | Workflow 3: mediation | Conclusion