The 15-row sample above (step 0) shows what a typical record looks like and where the data is messy. **What a typical row looks like** - Each row is one surveyed user, a student or a professional, identified by `User_ID`. It records demographics (age, gender, country, education level, profession), the AI tool and purpose, usage hours per day, a free-text list of tasks, and monthly AI cost. - It also holds outcome ratings (productivity, accuracy, satisfaction), time saved per week, work or study hours per day, and whether the user would recommend the tool. - The sample mixes students (e.g. U0207, a 16-year-old in the Philippines using Claude for content creation) and professionals (e.g. U0151, a 39-year-old senior software developer in India using Microsoft Copilot for coding). It covers many countries, tools and purposes: coding, studying, writing, research, customer support and content creation. **Data-quality quirks visible in the sample** - **Inconsistent formats:** - Age appears as "16 yrs" instead of 16. - Cost appears as "$20 " instead of 20. - Time saved appears as "6.5 hrs". - Usage hours appears as "1.6 hrs". - Gender appears as "F" as well as "Female". - Education appears as "undergrad" as well as "Undergraduate". - The tool appears as "copilot" as well as "Microsoft Copilot". - These are text columns, so they need cleaning before any numeric analysis. - **Missing values:** some rows lack the profession, tool, country, task list, or ratings. In U0138 the profession, tasks and accuracy are blank. - **Implausible combinations:** U0085 is a 17-year-old listed as a PhD student, and U0087 is a 17-year-old listed as undergrad. A German-language profession and task list also appears (U0230), so free text is multilingual. - **Duplicates and label sprawl (step 1):** the table has 300 rows but only 288 distinct `User_ID`s, so some users repeat. Gender has 12 distinct spellings, education level 17, AI tool 17 and recommendation answers 4. Most of these should collapse to a few clean categories. - **Gaps in the numeric scores (step 1):** productivity is present in 284 rows, accuracy in 285 and satisfaction in 293. I left `Monthly_Income` out of the sample because it is person-identifying. I can summarise it in aggregate if that would help.