0. Installation | 1. Annotate Graphs | 2. Network Visualization | 3. Topological Analysis | 3.1. Global topology | 3.2. Node-level metrics | 3.3. Is any of this surprising? | 3.4. Robustness Analysis | 3.5. Hubs | 3.6. Bottlenecks | 3.7. Calculate Roles | 3.8. Modules | 4. Extracting a subnetwork | 5. Information flow | 5.1. Network Diffusion | 5.2. Reverse Network Diffusion | 6. Credits and Contributions

The problem | The default is to ignore weights, loudly | What changes when weights are used | Diffusion propagates along strengths | Seeding from a continuous signal | What is not a weight | Reusing a diffusion kernel | Summary

The question that separates them | How the packages line up | FakeDataR, which came first | The rule of thumb

Step 1: make the blinded copy | Step 2: read the summary before you share anything | Step 3: give the assistant the file and the summary | Step 4: run what comes back on the real file | Four options that come up in this workflow | Doing it without writing R | Before you send it

It is not a formal privacy guarantee | The structure stays visible, on purpose | The blinded data is for writing code, not for analysis | Some code will not survive the trip | Detection can be wrong, and the summary is where you see it | Exact numeric coincidences are reported, not forbidden | keep_real is the one place the promise is switched off | What the package never does | Not in this version