How Interviewers Know Whether You Actually Built Your Notebook

Interview Insights · Data Analytics

How Interviewers Know Whether You Actually Built Your Notebook

SECTION 01Questions only the person who cleaned the data can answer

My favourite ownership question is not about code at all. It is: 'What was the most annoying thing about this dataset?' Someone who spent hours cleaning it answers instantly and with feeling. Maybe dates came in two formats, or some orders had negative quantities that turned out to be returns, or the same city appeared as 'Bengaluru', 'Bangalore' and 'BLR'.

Someone who downloaded a finished notebook usually answers in general terms: 'there were some missing values'. Every dataset has missing values. What I want is the specific, slightly irritating detail that you could only know by living in the data. That memory is very hard to fake and very easy to keep if you write it down while you work.

import pandas as pd

orders = pd.read_csv('data/orders.csv')

# Quirk 1: dates switched from DD-MM-YYYY to YYYY-MM-DD mid-year
orders['order_date'] = pd.to_datetime(orders['order_date'],
                                      format='mixed', dayfirst=True)

# Quirk 2: one city, three spellings
city_map = {'Bangalore': 'Bengaluru', 'BLR': 'Bengaluru'}
orders['city'] = orders['city'].replace(city_map)

# Quirk 3: staff test orders
is_test = orders['email'].str.endswith('@internal.example', na=False)
orders = orders[~is_test]

# Quirk 4: negative quantity rows are returns, not errors
returns = orders[orders['quantity'] < 0]
sales = orders[orders['quantity'] > 0]

DATA QUIRKS LOG (keep this while you work)

Date      | Quirk found                      | What I did
----------+----------------------------------+--------------------
Day 2     | Two date formats in order_date   | format='mixed', dayfirst
Day 2     | Negative quantities              | Split into returns
Day 3     | City spelled 3 ways              | Mapping dictionary
Day 4     | Staff test orders in data        | Filtered by email domain

Read this log the night before the interview.
Each line is a ready ownership answer.

One technical note: format='mixed' in pd.to_datetime needs pandas 2.0 or later. If an interviewer asks why you did not simply let pandas guess, the honest answer is that mixed day-first and year-first strings are exactly where guessing goes wrong, so you told it what to expect.

SECTION 02The small live edits that reveal fluency

The second check is a live change. During a screen share, I ask for something small: 'Can you show the median instead of the mean here?' or 'What if we only look at Delhi?' or 'Group this by quarter instead of month.' None of these are hard. They take a builder a few seconds, because the builder knows where the relevant line is and what it does.

A borrower has two problems. First, finding the right cell takes time, because they do not know the notebook's structure. Second, they are not sure what else depends on the variable they are changing, so they hesitate. I am not testing whether you can write perfect pandas from memory. I am watching how quickly you find your way around your own work.

summary = (sales
           .assign(month=sales['order_date'].dt.to_period('M'))
           .groupby(['city', 'month'])
           .agg(avg_order=('amount', 'mean'),
                revenue=('amount', 'sum'))
           .reset_index())
summary.head()

# Edit 1: median instead of mean
# Edit 2: only Delhi
# Edit 3: quarter instead of month
# Edit 4: add an order count

delhi = sales[sales['city'] == 'Delhi']

summary = (delhi
           .assign(quarter=delhi['order_date'].dt.to_period('Q'))
           .groupby('quarter')
           .agg(median_order=('amount', 'median'),
                revenue=('amount', 'sum'),
                orders=('order_id', 'nunique'))
           .reset_index())
summary

Named aggregations like avg_order=('amount', 'mean') make live edits easier, because the output column name tells you exactly which line to change. Writing your notebook this way is a small investment that pays off under pressure.

While you make an edit, say what you are doing: 'The metric is calculated in this agg call, so I will change mean to median here.' Talking while typing proves you know the structure even before the cell runs.

SECTION 03What your repository history says before you speak

The third check happens before the interview. If your notebook is on GitHub, I may glance at the commit history. A notebook that appears in a single commit called 'upload' with a polished final state is not proof of copying. Many people only push at the end. But it gives me nothing to support your story either.

A notebook that grew over several commits, with messages like 'add cohort table', 'fix date parsing for mixed formats' and 'move exploration to appendix', tells a story of real work. Even better if the messages match the quirks you mention in the interview. You cannot rewrite history for old projects honestly, but for every new project you can commit as you go, and for old ones you can explain how you worked.

# Commit as the analysis grows, with specific messages
git add analysis.ipynb
git commit -m "Parse mixed date formats in order_date"

git add analysis.ipynb
git commit -m "Split negative quantities into returns table"

git add analysis.ipynb data/README.md
git commit -m "Add Delhi quarterly summary and data notes"

# See how the notebook changed over time
git log --oneline -- analysis.ipynb

# Notebook diffs are noisy because outputs are stored
# inside the .ipynb file. Two common helpers:

# 1. nbdime shows notebook-aware diffs
pip install nbdime
nbdiff analysis.ipynb

# 2. Clear outputs before committing a work-in-progress
jupyter nbconvert --clear-output --inplace analysis.ipynb
git add analysis.ipynb
git commit -m "WIP: cohort grouping"

If part of your notebook came from a course, a tutorial or a teammate, say so plainly. 'The plotting helper came from a course exercise; the cleaning and cohort logic are mine' is a strong answer. It is specific, honest and puts the focus on what you did.

SECTION 04An evening plan to prepare for ownership checks

If the notebook is genuinely yours, these steps make that obvious in the interview. If parts are adapted, they help you separate and explain your own contribution clearly.

STEP 1: REBUILD YOUR QUIRKS LIST (30 minutes)

Open the raw data again, not the cleaned version.
For each column, ask:
  - Any odd formats or units?
  - Any impossible values (negative, future dates)?
  - Any spelling variants?
  - Any rows that should not be there?

Write at least four quirks with what you did
about each. These are your ownership answers.

STEP 2: LIVE EDIT DRILL (30 minutes)

Time yourself on each, starting from the open notebook:

[ ] Change one aggregation (mean -> median)
[ ] Add a filter for one category or city
[ ] Change the time grain (month -> quarter)
[ ] Add one new column to a summary
[ ] Sort the result by a different column

Speak each step aloud while typing.
Target: find the right cell in under 15 seconds.

STEP 3: CONTRIBUTION MAP (15 minutes)

Section             | Mine | Adapted | From where
--------------------+------+---------+-----------
Loading and parsing |  [ ] |   [ ]   | _________
Cleaning            |  [ ] |   [ ]   | _________
Main analysis       |  [ ] |   [ ]   | _________
Charts              |  [ ] |   [ ]   | _________
Conclusions         |  [ ] |   [ ]   | _________

Practise one honest sentence that summarises this map.

SECTION 05Habits that make ownership obvious

Build these habits into every new project, and ownership checks become easy.

• Log quirks as you find them — Keep a short text file of data surprises and your fixes. It becomes the most convincing interview material you own.
• Commit at milestones — Push after each meaningful step with a specific message, so your repository tells the story of how the work grew.
• Use named aggregations — Output names such as median_order make every metric easy to locate when an interviewer asks for a change.
• Credit borrowed parts — Note in a comment where a helper function came from. Honesty about small borrowed pieces makes the rest more believable.
• Narrate while you type — In live edits, say where the change goes and why before you run it. Silence while searching looks like unfamiliarity.

SECTION 06Interview Q&A — Notebook ownership

SECTION 07Test yourself — Notebook ownership quiz

SECTION 08Frequently asked questions

SECTION 09Related reads