Analyze a dataset

Upload a small CSV, ask a statistical question, approve the first code execution, and find a chart and a report among the Artifacts.

• Time: about 15 minutes. The run itself usually takes a few minutes.
• You need: a working task model from the quick start.
• Input: plant-growth.csv, a synthetic sample shipped with the repository: 45 plants in three fertilizer treatments, with the daily light each plant received and its dry mass.
• Skills and connectors: none. The run uses the managed Python environment, which includes pandas, SciPy and Matplotlib.
• You end with: growth-report.md and treatment-boxplot.png in the Artifacts list.

This is the second lesson of the learning path. The quick start is the first.

1. Get the sample file

If you cloned the repository, the file is already at docs/examples/plant-growth.csv. If you run the prepackaged binary, download it:

curl -LO https://raw.githubusercontent.com/openJiuwen-ai/sciencediscovery/main/docs/examples/plant-growth.csv

The values are synthetic and were generated for these tutorials; see the sample data notes.

2. Check the Python environment

Open System configuration → Local Runner → Scientific environments. The Python base environment should show Ready. A new data directory prepares it in the background after the first start; while it shows Installing, wait before you send the task.

3. Upload the file

Create a Project, for example plant-growth, and open its Session. In the Workspace panel on the right, drop plant-growth.csv onto Drop files, or click that area to choose the file. A confirmation appears, and the file is listed under Files and among the Artifacts as a user upload.

4. Send the task

Paste the prompt below into the message box and send it:

Analyze the uploaded file plant-growth.csv. It records the dry mass of 45 plants grown under three fertilizer treatments.

1. Load the file with pandas and report how many plants each treatment has.
2. Summarize dry_mass_g per treatment (n, mean, standard deviation) as a table.
3. Test whether mean dry_mass_g differs between treatments with a one-way ANOVA (scipy.stats.f_oneway), then compare the pairs with Tukey HSD (scipy.stats.tukey_hsd). Report the statistics and p-values.
4. Draw a box plot of dry_mass_g by treatment with the individual plants overlaid, and save it as treatment-boxplot.png.
5. light_hours also varies between plants. Check whether it is balanced across treatments and say whether it could explain the differences.
6. Write growth-report.md with the table, the test results, the figure and a short limitations section.
7. Declare growth-report.md and treatment-boxplot.png as artifacts and name them in your final reply.

Use the managed Python environment, which already has pandas, SciPy and Matplotlib, and run every calculation; do not estimate numbers in the reply.

5. Approve the first code execution

With the default approval mode, Ask for dangerous actions, the first command pauses on a Permission required card for Run local code. The card shows the command the Agent wants to run.

• Allow once approves this command only; the next command asks again.
• Allow same type approves this and later code executions, so the rest of the analysis runs without stopping.
• Deny refuses it, and the Agent has to continue without running code.

Choose Allow same type for this lesson. The code still runs inside the sandbox, with only the workspace and no network. You can revoke the grant later under System configuration → Permissions; the last lesson does that.

6. What you should see

Different models write different code and prose, so expect the shape below rather than particular wording or numbers:

• The timeline shows the Agent listing and reading the file, then several code executions. Expand one to see the command and its output. A failed execution followed by a corrected one is normal.
• growth-report.md and treatment-boxplot.png appear among the Artifacts, often with the analysis script and a results file beside them.
• Opening growth-report.md shows a table with 15 plants per treatment, the ANOVA statistic and p-value, pairwise comparisons between the three treatments, the figure or a reference to it, and a limitations paragraph.
• The report says whether light_hours is balanced across treatments. In this sample it is not perfectly balanced, so a careful report discusses it instead of ignoring it.
• About a minute after the report is registered, the Reviewer may add an automatic review card. Its findings are the starting point of the last lesson.

7. Check your result

• The run finished and no longer appears as running.
• The timeline contains code executions that you approved.
• growth-report.md and treatment-boxplot.png are listed among the Artifacts.
• The numbers in the report come from executed code, and the figure opens in the preview.

If the run cannot import pandas, check that the Python environment is Ready and ask the Agent to use the managed environment. If no Artifact appears, ask it to declare the two files as artifacts.

Next

Keep this Session; later lessons come back to its report. Continue with Search the literature and write a cited summary.