Public demo guide
The demo is at http://34.61.255.147/ (Fragility workbench on the same VM as https://hub.agenticop.io/). Everything is server-hosted — you open samples in the browser; nothing is uploaded from your computer.
Guardrails: the public site is browse-first. Bundled samples and viewers are open. Live Run a scenario requires an operator API key (or install locally: pip install fragility-engine). See Demo guardrails.
Pages at a glance
| Page | URL | What to do here |
|---|---|---|
| Workbench | / | Browse server-hosted samples and open them in viewers |
| Run a scenario | /run.html | Operator-gated live search (API key) — or use local CLI |
| Past runs | /runs.html | See completed server runs and open their results |
| Guided tour | /?tour=1 | Step-by-step walkthrough with written explanations |
| Documentation | /docs/ | Full manual |
| This host | /host.html | Validation snapshot and engine health |
Viewers
Click any sample row on the workbench to open the right viewer. You can also use the Presets dropdown inside each viewer.
| Viewer | Opens | What you see |
|---|---|---|
| Replay | Replay JSON | Step-by-step collapse timeline |
| Pareto | Pareto JSON | Cost vs severity trade-off chart |
| Attribution | Counterfactual / chain JSON | Which steps caused the outcome |
| Composite | Composite JSON | Same attack across multiple domains |
| Coupling sweep | Sweep JSON | Research fork: coupling strength vs instability |
| Coupling compare | Comparison JSON | Research fork: two coupling levels compared |
Local file pickers and drag-and-drop are disabled on this demo. Use the Presets menu or the workbench links. To load your own files, clone the repo and open the viewers from artifacts/ on a local server.
Running a new scenario
Live runs on the public demo need an operator run API key. Without it, use bundled samples or install locally:
```powershell
pip install fragility-engine
fragility search --example capacity-pool
```
If you have a key:
1. Go to Run a scenario (or ?run_key=… once).
2. Choose a domain and search type from the dropdown.
3. Set the seed, horizon, generations, and population (within server limits).
4. Click Run scenario. When the job finishes, links appear to the replay or Pareto viewer.
Results are saved on the server under /runs/<id>/. Nothing on your computer is read or modified.
Server limits: max 150s · 1 concurrent run · 3 runs/hour/IP · anonymous runs off by default.
Six charter domains vs the research fork
Charter domains (aggregate peg, network contagion, resource cascade, service backlog, liquidity ladder, inventory buffer) are the stable reference workbench.
Coupled institution is a research fork: peg panic and overload exchange signals inside one simulation step. It has its own sample section on the workbench and is documented in Coupled fork (research).
Suggested first visit
1. Workbench → click Flagship benchmark replay to see a collapse timeline.
2. Open any Attribution sample to see a counterfactual chain (no live run needed).
3. Optional: work through the Guided tour — read each step, then press Next.
4. For a fresh search: install locally or ask an operator for a run key.
For operators
- Check validation:
/status.jsonand/host.html - Deploy updated site from local machine:
powershell -File scripts/gce_deploy_public_site.ps1 - Optional static mirror:
powershell -File scripts/deploy_github_pages.ps1