Sub50 is an open, interactive blackout simulator for the Chilean national power grid (SEN) that runs in the browser. The user cuts a transmission line or a plant on the map. The app shows which zones lose power, how many people are affected and why, as a chain of events in plain language. The name refers to grid frequency falling below 50 Hz. Product, scope and limits: docs/idea.md.
The main objective is to estimate the consequences of cyberattacks on this infrastructure. Sub50 models the effect of an attack (plants or lines disconnected, defences disabled), never how the attack is done. The guardrails are in docs/cyber.md section 8.
- The cascade that follows a line cut on a 188-bus reduction of the SEN (PyPSA-CL data): DC power flow, overload trips, islands, island frequency with automatic load shedding (EDAC), and people without supply (
docs/engine.md). - A strategic theatre on the same network: events, defensive actions and agents that keep a mission under adverse conditions (
docs/strategy.md,docs/agents.md). - Shared sessions, an MCP server and an in-app assistant that drive the same engine (
docs/sessions-api.md,docs/mcp.md,docs/assistant.md).
Needs only Docker with Compose:
docker compose up --buildThe web app is at http://localhost:5173 and the API at http://localhost:8000 (/docs). Ports, the Mapbox token and the optional LLM are in docs/docker.md.
- Validation cases V1-V10 are tests in
backend/tests/validation/. They run on the 13-area engine (backend/src/cascadasen/domain/simulation), which servesPOST /api/v1/simulate. - The 188-bus engine that serves the map, the sessions and the theatre is checked only against the V1 flow distribution (
backend/tests/unit/test_cascade.py). - V6 and V9 are skipped because their reference data is not in the repo (
test_engine_cases.py). - The cases show that the engine reproduces published figures of the 25 February 2025 blackout and of the grid operator's studies. They do not show that it predicts future events. V1 and V3a are calibration cases (
docs/plan.mdsection 5), so passing them shows the fit, not accuracy.
See docs/engine.md and docs/plan.md section 5.
CI runs backend pytest with a coverage floor, mypy (0 errors, blocking), web tsc/vitest/build, and the deploy needs them. Lint (ruff) and the CHANGELOG.md check run on every pull request. The workflows are in .github/workflows/.
| Name | Where |
|---|---|
| Sub50 | The product |
cascadasen |
The Python package and its commands (cascadasen-build, cascadasen-cache, ...), the earlier product name |
platanus-hack-26-cl-team-9 |
This repository |
| Path | Content |
|---|---|
backend/ |
Python package cascadasen (uv): FastAPI API, the two engines, the strategic theatre, MCP server, tests (backend/README.md) |
web/ |
Vite + React + TypeScript app with deck.gl and an optional Mapbox basemap (web/README.md) |
data/ |
Public-source scripts (scripts/), committed processed data (processed/), scenarios |
docs/ |
Documentation, listed below |
infrastructure/ |
Terraform and deploy script for AWS (docs/infrastructure.md) |
scripts/ |
lint.sh and git hooks |
docker-compose.yml |
Local run |
| Doc | Topic |
|---|---|
docs/idea.md |
Product, scope, limits, risks |
docs/cyber.md |
Main objective (cyberattack consequences) and guardrails |
docs/stack.md |
Decisions, simulation model, data contracts, API |
docs/architecture.md |
How the system is built |
docs/README.md |
Index of every doc: purpose, when to read it, status, "where is X" |
docs/code-map.md |
Backend and web module by module, data flows, tests per module |
docs/for-reviewers.md |
Claims, where to verify them, limitations, pending decisions |
docs/index/ |
Generated index of files, symbols, routes and tests (repo-index.json) |
llms.txt |
The same map for an LLM, in the llmstxt.org format |
docs/adr/ |
Architecture decision records |
docs/plan.md |
Phases, validation cases, open decisions |
docs/sessions-api.md |
Shared sessions and the agent API |
docs/mcp.md |
MCP server |
docs/assistant.md |
In-app assistant |
docs/regional.md |
Other countries' networks |
Read order for contributors and agents: AGENTS.md.
team-9, Platanus Hack 26: Santiago ft. Buk Project.
Current project logo: project-logo.png
Track: 🌐 Simulations
team-9
- Valentina Honorato (@ValHonom)
- Ricardo Olalquiaga (@Icksir)
- Alonso Utreras Miranda (@AlasAltum)
- Ignacio Montenegro (@ignaciomontenegroc)
- Felipe Javier Sanhueza Celis (@Sanhuezafce)
Deploy URL and project metadata: platanus-hack-project.jsonc.
