First 30 Minutes: Guided Exercise

Stop reading, start doing. This guided exercise uses your own data to demonstrate Claude Code's value in under 30 minutes.

Before You Start

What You Need

• [ ] Claude Code installed and working (Setup Guide)
• [ ] VS Code open
• [ ] A dataset you know well (any format: CSV, Excel, R data, etc.)
• [ ] 30 uninterrupted minutes

Choose Your Dataset

Pick something:

• Not sensitive (no PII, no IRB-protected data)
• Familiar to you (you know what the values should look like)
• Messy enough to be realistic (perfect data won't show Claude's value)
• Moderate size (100-10,000 rows is ideal for this exercise)

Good choices:

• Old project data you know well
• Public dataset from your field
• Preliminary data from a current project (if shareable)
• A teaching dataset you use with students

Minute 0-5: Setup

Create a Working Directory

mkdir claude-code-test
cd claude-code-test

Copy Your Data

Put your dataset file in this directory. We'll call it my_data.csv in examples, but use your actual filename.

Start Claude Code

claude

Set Context

Tell Claude what you're working with:

1"I'm testing Claude Code with my own research data.
2The file is my_data.csv.
3This is [briefly describe: survey data, field observations, experimental results, etc.]
4from my research on [topic].
5 
6Read the file and give me an overview."

Minute 5-10: Data Quality Assessment

Ask for a Quality Report

Type this prompt:

1Give me a data quality report for my_data.csv:
21. Dimensions (rows, columns)
32. Column names and data types
43. Missing values by column
54. Obvious outliers or anomalies
65. Any apparent data entry issues
7 
8Be specific about what you find.

What to Look For

Claude should return something like:

1DATA QUALITY REPORT: my_data.csv
2 
3DIMENSIONS: 847 rows × 12 columns
4 
5COLUMNS:
6- site_id (string): 12 unique values
7- date (string): mixed formats detected...
8- temperature (float): range 12.3-89.4°C ← possible outlier
9...
10 
11MISSING VALUES:
12- species_count: 45 missing (5.3%)
13- salinity: 8 missing (0.9%)
14...
15 
16POTENTIAL ISSUES:
171. Row 234: temperature = 89.4°C (unit error?)
182. Dates in two formats: MM/DD/YYYY and YYYY-MM-DD
19...

Checkpoint ✓

At this point, you should:

• See a summary that matches your expectations
• Notice some issues you already knew about
• Possibly discover issues you didn't know about

If Claude found something you didn't know: Good! That's the point.

If Claude made an error about your data: Correct it. This teaches you how to guide it.

Minute 10-15: Quick Fix

Pick One Issue to Fix

From the quality report, choose something specific:

Fix the date format inconsistency you identified.
Convert all dates to YYYY-MM-DD format.
Show me what you're changing before you do it.

Review Before Approving

Claude should show you:

• Which rows are affected
• What the changes will be
• Ask for confirmation before modifying

Important: Always review changes before accepting them.

Create the Cleaned File

Apply that fix and save as my_data_cleaned.csv.
Also create a log file (cleaning_log.txt) documenting what was changed.

Checkpoint ✓

You now have:

• A cleaned data file
• A log of changes (for reproducibility)
• Experience reviewing Claude's proposed changes

Minute 15-20: Simple Analysis

Run an Appropriate Test

Ask for an analysis that makes sense for your data:

For continuous data:

Calculate descriptive statistics for [your main variable]
grouped by [your grouping variable].
Include: mean, SD, n, SE, 95% CI.
Format as a table.

For categorical data:

Create a frequency table for [your categorical variable]
cross-tabulated by [another variable].
Include row and column percentages.
Run a chi-square test if appropriate.

For temporal data:

Show me the trend in [your variable] over time.
Calculate summary by [month/year/season].
Flag any notable patterns.

Verify the Results

Here's the crucial step:

I want to verify this result.
Show me how to calculate [one specific statistic]
manually from the raw data.
Walk me through the calculation.

This isn't about distrust—it's about understanding.

Checkpoint ✓

You've:

• Run a real analysis on your own data
• Learned how to verify Claude's calculations
• Seen how fast this can be

Minute 20-25: Quick Visualization

Request a Figure

1Create a figure showing [the relationship you just analyzed].
2Make it publication-quality:
3- Clear axis labels with units
4- Appropriate font sizes
5- Colorblind-friendly colors
6- High resolution (300 DPI)
7 
8Save as figure1.png

Iterate on the Design

Ask for one change:

Make the title more descriptive and add error bars.

Or:

Change the color scheme to match Nature style.

Or:

Add a trend line with 95% CI shading.

Checkpoint ✓

You have:

• A real figure from your data
• Experience iterating on visualizations
• A sense of how quickly you can refine output

Minute 25-30: Documentation

Generate a README

1Create a README.md for this analysis:
21. What data this is (keep it general—no sensitive details)
32. What cleaning steps were applied
43. What analysis was performed
54. What the figure shows
65. How to reproduce this analysis
7 
8Make it suitable for including in a repository.

Create a CLAUDE.md for Future Work

1Based on this session, create a CLAUDE.md file for this project type.
2Include:
3- Common commands I'd need
4- Data format expectations
5- Analysis conventions for my field
6- Quality checks to always run
7 
8This is for [R/Python] analysis of [your data type].

Checkpoint ✓

You've:

• Documented your work
• Created a template for future projects
• Completed a full mini-workflow

What You've Done in 30 Minutes

1. Data assessment — Quality report identifying issues
2. Data cleaning — Fixed a real problem with audit trail
3. Analysis — Ran statistics appropriate for your data
4. Visualization — Created a publication-ready figure
5. Documentation — Generated reproducible records

This is the core loop. Every project is variations on this.

Now Try These Extensions

Next 30 Minutes (Optional)

If you found data issues:

Create a comprehensive cleaning script that fixes
all the issues you identified in the quality report.
Run it and show me the before/after comparison.

If your analysis worked well:

Now run [a more complex analysis appropriate to your data].
Explain the assumptions and check if my data meets them.

If you want to explore visualization:

Create three different ways to visualize this relationship.
Explain the pros and cons of each for a journal submission.

Common Questions After First Use

"It made a mistake about my data"

This happens. Fix it:

Actually, [explain the domain context].
Given that, revise your analysis/recommendation.

Claude doesn't know your field's conventions—teach it.

"How do I know the statistics are right?"

Always verify critical results:

Show me the formula you used for [statistic].
Now calculate it step-by-step so I can verify.

For publication, run final analyses in validated software.

"This seems too easy"

It is easy for routine tasks. That's the point.

The value is:

• Fast iteration on exploration
• Less time on boilerplate
• More time on thinking about your science

"What about my more complex workflows?"

See Research Case Studies for complete examples of:

• Publication pipelines
• Reviewer responses
• Multi-author coordination
• Legacy code rescue

Red Flags to Watch For

If during this exercise you noticed:

• Claude made statistical errors → Always verify before publishing
• It didn't understand your domain conventions → Add context, use CLAUDE.md
• The code didn't run → Specify your environment (R version, packages)
• It assumed things about your data → Be more explicit in prompts

These aren't dealbreakers—they're the learning curve.

Your First CLAUDE.md

Based on this exercise, here's a starter template:

1# [Your Project Name]
2 
3## About This Data
4- Type: [survey/experimental/observational/etc.]
5- Format: [CSV/Excel/R data]
6- Size: [approximate rows]
7- Key variables: [list main ones]
8 
9## Common Tasks
10- `python clean_data.py` - Apply standard cleaning
11- `Rscript run_analysis.R` - Run main analysis
12 
13## Data Conventions
14- Missing values coded as: [NA/-999/blank/etc.]
15- Date format: [YYYY-MM-DD]
16- [Other conventions specific to your field]
17 
18## Quality Checks to Always Run
191. Check for duplicate IDs
202. Verify date ranges are plausible
213. Check [variable] is within [expected range]
22 
23## Notes
24- [Any quirks about this dataset]
25- [Field-specific terminology Claude might not know]

Next Steps

Now that you've experienced the basics:

• Research Case Studies — Complete workflow examples
• Collaboration Workflows — Working with teams
• Limitations — When not to use Claude Code
• CLAUDE.md Generator — Create project-specific configs

You did it. You went from "what is this tool?" to "I used my own data" in 30 minutes.

That's the pitch: real work, faster, with appropriate caution.