Current release
This page is the short, honest description of what exists in the application repository on 11 October 2026. It describes application commit e9733d7 and Python SDK version 0.4.4. The final README update in that commit does not alter simulation behaviour; the preceding 80ed129 commit contains the verified account-deletion migration. Deployment evidence should still record the exact submitted commit and deployment ID.
What a researcher can do
- Create an account, confirm an email address, reset a password, optionally enable authenticator-based two-factor authentication and permanently delete the account from Security after an explicit confirmation.
- Create a research project from a supplied arena template, add project members and give each member only the modules they should edit.
- Build a custom arena, configure doors, lights, sound, odour, treatment, timing and the random seed, then save a versioned experiment.
- Run the experiment in the browser, control doors during a run, change camera views and record interventions.
- Review a live or saved run, seek to a step, inspect readable results and export JSON or CSV.
- Run the same simulation without graphics from Python or Google Colab using
rodent-sdk. - Train a researcher-defined control model through a Gymnasium-style environment, record episodes, rebuild selected episodes and compare results.
- Save versioned project models, export an ONNX copy and use a compatible saved or uploaded ONNX model to drive the rat live in the website.
Current technical release
| Area | Current implementation |
|---|---|
| Research portal | HTML, CSS and JavaScript hosted on Vercel, with the Godot web export embedded for 3D execution and replay. |
| Simulation | Godot 4.7.1 fixed-step kernel, six supplied presets, custom arena support, deterministic seeds and headless execution. |
| Stimuli | Step-scheduled light, sound and odour. Floor and overhead light use geometry, material transmission, reflection or shadow rules as appropriate; sound is distance and material attenuated; odour releases, diffuses, decays and follows door and material conductance over time. |
| Research models | rodent-sdk 0.4.4, Gymnasium-style training, reproducible episode logs, versioned model storage and live browser execution through ONNX Runtime Web. |
| Data | Supabase Auth, Postgres, Row Level Security, project-scoped memberships, experiments, replays, model metadata and private model files. |
| External integration | Open-Meteo provides optional weather-derived setup suggestions that the researcher reviews before applying. |
| Quality | Gitea Actions runs Ruff and gdlint, checks JavaScript syntax and presets, enforces at least 85% measured Python line coverage, runs Godot and SDK tests, and records an API load-smoke report. |
What this release does not claim
RODENT supplies an experiment platform and model interface. The built-in moving rat is a software baseline, not a validated biological model. A researcher-trained model is also not automatically scientifically valid merely because it trains successfully or produces repeatable results.
The platform reports configured fields, calculated exposure, observations, actions and recorded outcomes. Scientific interpretation still requires an appropriate model, calibration, validation, experimental design and analysis outside the platform.
Where to start
- New researcher: Accounts and projects
- Browser experiment: Experiments
- Python or Colab: Python and Google Colab
- Train or connect a model: Python SDK
- Understand the system: Architecture
- Assess project evidence: Rubric and evidence map
AI Attribution: This status page was prepared with OpenAI Codex assistance from the merged application source. The team remains responsible for verifying the deployed build and scientific claims.