Skip to main content

Experiments

Overview​

An experiment is a saved, versioned configuration inside a project. It defines the arena, simulation timing, stimuli, treatment and the current demonstration agent. The platform does not yet provide a validated biological rat or medicine-response model.

The experiment configuration determines factors such as the simulation settings, behavioural characteristics, environmental stimuli and treatment conditions.

Researchers use the browser workspace rather than editing JSON by hand. Choose an experiment from Your projects; it opens in a dedicated tab. Its starting arena is shown on the experiment card when a template ID is saved.

Experiment Configuration​

An experiment can contain configuration for:

  • Simulation settings
  • Random seed
  • Arena settings
  • Rat characteristics
  • Behaviour preset
  • Environmental stimuli
  • Stimulus schedules
  • Treatment conditions
  • Treatment dose
  • Treatment administration timing

Behaviour Presets​

The demonstration agent supports three behaviour presets. They make prototype runs visibly different; they are not measured rodent phenotypes.

Explorer​

Explorer makes the demo agent more active and willing to explore.

Balanced​

Balanced is the default demonstration behaviour.

Shelter Seeker​

Shelter Seeker biases the demonstration agent toward safer areas.

Environmental Stimuli​

The experiment can contain configurable:

  • Light sources
  • Odour sources
  • Sound sources

Stimuli can have configurable properties such as:

  • Position
  • Range
  • Intensity

Stimuli can also be scheduled to activate at specific simulation steps.

This allows controlled environmental changes to be introduced during an experiment.

Treatment Conditions​

Experiments can use one of three treatment conditions:

  • No treatment
  • Placebo
  • Medicine

Treatment settings can include:

  • Treatment type
  • Dose
  • Administration timing

Treatment events are recorded so that they can later be correlated with behavioural data.

Set up an experiment​

The website follows 1 Setup → 2 Run → 3 Review. In Setup, the left panel has four smaller sections so you do not have to scroll through one huge form:

SectionWhat to use it for
ConditionsRename the experiment; set seed and fixed step duration; choose no treatment, placebo or medicine; set dose and administration step; configure available light, sound and odour sources.
ArenaChoose the starting state of each available door. Apply starting doors resets the configured run.
TemplatesSee the current preset, choose an available preset, and edit region materials or colours. A preset change changes the experimental setup.
EnvironmentEnter location, local date and time to fetch weather-based ambient suggestions. Review them before using them in the draft. Temperature is shown as source data, not simulated heat.

Use Apply setup to load your edited values into the simulator. Applying restarts the run, so do it before collecting data. Use Save draft to persist the experiment. A member may only edit modules granted for that project; a missing control may be a permission restriction rather than a loading error.

The checked integration branch has six validated JSON presets. The website displays what is in its current build, so the deployed portal may show fewer until updated. Ahmed has implemented a 2D custom arena builder on feature/arena-editor, with Region, Wall and Door tools, grid settings, JSON preview and Create experiment. That branch has not yet been merged into the checked workspace; its save path needs end-to-end verification in the combined build.

Reproducibility​

To reproduce a software run, keep the same saved configuration, seed, fixed step duration, plugin versions and interventions. The same settings should yield the same trajectory in the deterministic core.

Changing 1x to 100x changes how quickly steps are displayed, not the fixed step size. Randomize the seed only when you deliberately want a different run. Determinism is a software property, not evidence that the demo agent predicts a real animal.

AI Attribution: This document was drafted and edited with AI assistance, including ChatGPT and OpenAI Codex. The team reviewed it against the implemented project.