How code-philia/hackathon-local-simulation works
This repository packages a local ARC-Bench submission simulator, not the competition website. A user supplies an agent ZIP and requirements, optionally public tests, then local_submit.py assembles a workspace and launches the runner in Docker. The runner executes the agent, deploys its generated application, and optionally evaluates it with Playwright; results and logs remain in the workspace. The README documents these stages and optional Meter usage, while sampled source confirms workspace handling, result reading, and runner customization. Production runner internals are not included in the tree, so their internal wiring is not expanded.
Main entry points: local_submit.py Submission CLI · tests Public tests · local_runner.py Local runner adapter · requirements Requirements
Components 17
Submission inputs
- Agent ZIP
- Requirements
public-exercise/bookstack/requirements - Public tests
public-exercise/bookstack/tests - Model environment
env.example
Local orchestration
- Submission CLI
local_submit.py - Run workspace
Container execution
- Docker container
- Local runner adapter
local_runner.py - Production runner
- Agent execution
- Generated application
- Application deployment
Evaluation and results
- Playwright evaluation
- ARC Bench Meter
- Run artifacts
- Result reader
local_submit.py
Other components
- Participant
Connections 21
- Participant → Submission CLI: submits inputs
- Agent ZIP → Submission CLI: provides agent
- Requirements → Submission CLI: provides requirements
- Public tests → Submission CLI: optionally provides tests
- Model environment → Submission CLI: configures models
- Submission CLI → Run workspace: assembles
- Submission CLI → Docker container: starts container
- Run workspace → Docker container: mounts inputs
- Docker container → Local runner adapter: invokes
- Local runner adapter → Production runner: loads and adapts
- Production runner → Agent execution: runs agent
- Agent execution → Generated application: generates
- Production runner → Application deployment: starts application
- Application deployment → Generated application: serves
- Production runner → Playwright evaluation: optionally evaluates
- Public tests → Playwright evaluation: supplies test cases
- Agent execution → ARC Bench Meter: records usage
- Playwright evaluation → Run artifacts: writes report
- Production runner → Run artifacts: writes logs and status
- Run workspace → Result reader: provides run files
- Result reader → Participant: reports outcome