Ref: Coursera - Deeplearning.ai Data Analytics Course.

Link to the course: Link.

ROUGH NOTES (!)
Updated: 20/5/26

DATA ANALYTICS

We will consider

DATA ANALYTICS FOUNDATIONS:

[Welcome to data analytics]

As the world has become more digital, a lot of ​activities now create data.

Eg: When you send an email, when you go to a website and click on ​some things and not on others, when you buy something, when you decide to watch one movie ​and not another, etc.

The world has lots of data. People who are able to look at ​the data and analyze it can make better decisions.

Eg: At Netflix, they are using data analytics to help inform the production, distribution, and promotion of content.

Data science is an investigative activity.

Q) Was there a time where you used data and it changed your mind about the decison you were going to make?

A from Andrew Ng:

He worked at AI Fund. As a venture studio, they work to bring in a lot of CEOs to lead different businesses. For a long time, a lot of hiring goes by gut. They thought that CEOs need to have fundraising experience. They looked at the data, and it turns out they were wrong.

A from Sean Barnes:

He used to do research on hospital operations. He worked on predicting when patients would be discharged from the hospital, and one of the signals was whether the patient had a visit from a physical therapist. Oftentimes, that was actually a signal that the patient was not close to going home, so ​that was a really surprising result.
(Hospital beds are valuable resources, so you want to be able to send patients home so that you can make room.)

When a company faces a decision, if there’s someone around the table with access to data and they go and bring in the facts, in a few minutes you can get to deeper facts than was possible without data.

[Generative AI in this course]

We will learn to use generative AI. In particular, large language models (LLMs) like ChatGPT, Claude, Gemini, and so on.

You’ll learn how to use LLMs to synthesize information from your stakeholders, explore a dataset and its metadata, automatically run data analysis by writing code for you, interpret images of data visualizations, and create data visualizations.

You’ll also learn about LLMs’ key limitations, including the tasks they can’t do for you.

LLMs in this course:

• Demonstrates the most up-to-date capabilities as of 2025

• Evergreen principles:
  • How to think about and use generative AI in your work, regardless of which specific product you end up working with.
• Develop a mindset of iteration and experimentation.

LLM progress since launch:

• Abilities have advanced rapidly.

• New features constantly being released.

Changes you should expect:

• More advanced and specialized features.

• Cheaper tools.

• Faster tools.

• Higher quality outputs overall.

LLM options:

• Won’t need to purchase any additional products.

• Its important for you to see the available options, including the paid options, so that you develop confidence experimenting and selecting the best tools in your work as a data analyst.

• This course does not recommend any single tool. You will see several tools throughout the modules.

• Learn core principles to work with LLMs.

[Module 1 Introduction]

From ancient civilizations tracking agricultural cycles, ​to modern businesses optimizing their decision-making, ​the principles of data analytics have been shaping our world for thousands of years.

In this module, you’ll gain a comprehensive understanding ​of the different types of data and how they flow through an organization.

[Life as a data analyst]

What’s to love about data analytics:

• Always discovering something new

• Connects with different fields

• Competitive salaries and job security

• People from different backgrounds

• Constantly learn new technical skills

Great role for people who love:

• Learning

• Problem-Solving

• Making discoveries

A typical day:

9:00 – Explore a new problem you need to solve.

10:00 – Meet data team. Catch up about the company’s new priorities, and get your hands on the data you need.

11:00 – Work work work! Get your hands dirty with spreadsheets, ​databases, and coding. ​Make new discoveries and even have a wow moment or two.

2:00 – Create a dashboard. Figure out how to tell the hidden stories in the data and visualize those stories.

3:00 – Present your progress on the dashboard. Get direct feedback from your teammates ​about how this dashboard will help them deliver value.

3:30 – Celebrate! Reward yourself with an afternoon beverage ​and a quick break.

4:00 – Learn a new skill. Take an advanced statistics course ​or learn a new programming language. ​It’s so rewarding to learn on the job.

6:00 — Happy hour with the data team. Socialize with your fellow analysts and data scientists, ​learn about new trends, and discuss upcoming projects.

A year or two:

You will likely complete several big projects. Seeing the impact of your work in the real world ​is incredibly satisfying.

You will be able to create a portfolio of your work. A portfolio not only showcases your skills, ​but also helps you prepare for future growth opportunities.

You will also develop your industry expertise, ​from terminology to unspoken rules. Your technical skills will improve significantly ​as different projects require you to upskill.

By leading successful projects, ​you’ll earn the trust of your teammates. Building strong relationships creates opportunities ​to work on compelling problems.

Career of a data analyst:

You may find yourself:

• Switching industries and gaining a breadth of experience.

OR

• Developing deep expertise in a particular technical area.

Types of leadership:

• Manager

• Individual contributor

• Educator

• Mentor

As you gain experience, you might decide to start your own business or consulting firm, leveraging that expertise to help others succeed.

Might start your own:

• Business

• Consulting firm

[What is data analytics?]

What is data analytics?

A diverse set of skills and tools that enable businesses to make better decisions.

It’s all about leveraging data to:

• Gain insights

• Support decision-making

rather than relying on chance or experience alone.

What is data analytics?

Data analytics involves Math, Programming, and Business Intuition.

Other investigative roles

Scientist:

Starts with a specific hypothesis. Then analyzes data to evaluate the hypothesis.

Detective:

Gathers evidence. Pieces it together to understand the crime.

Journalist:

Synthesizes information. Creates a compelling narrative.

Data analytics vs Data analysis

What’s not new:

• Statisticians, scientists, and engineers analyzing data

• Data visualization

• Eg: 1150 BC Turin Papyrus Map

What’s new:

• Explosion of data. We’re collecting more detailed data than ever before.

• More powerful tools.

• Eg: Working at the scale of 615 million monthly active Spotify users

Data analysts are in demand at

• Tech companies

• Hospitals

• Sports teams

• Manufacturing plants

• Research institutions

Distinction between data roles

Data Scientists:

Use sophisticated modelling techniques.

Data analysts:

Use foundational statistical techniques and data visualization.

Business Intelligence analysts:

Use commercial software like spreadsheets, with lighter emphasis on programming.

[Evidence-based decision making]

Decision-making methods

There are three basic ways you can make a decision.

• Leave it to chance

• Go with the intuition

• Evidence-based decision making

The choices are ordered in the increasing order of Information Utilization.

Decision making methods

Consider decisions like:

• Which university should I choose?

• Is it smart for me to move to a new city?

• How should I invest my savings?

Think for a moment about the kinds of information you might gather to answer each of these questions.

• Which university should I choose?
  Might write out pros and cons of each college.

• Is it smart for me to move to a new city?
  Might discuss the new city with a friend who lives there.

• How should I invest my savings?
  Might track the performance of different investments to decide which is best.

Effort vs impact in decision-making:

The level of effort needed to accumulate evidence is proportional to the impact of the decision.

The higher the stakes, the more evidence you need to support your decision.

Eg: In fields such as criminal justice, medicine, or journalism, ​where decisions can have serious consequences, ​it’s not enough to rely on opinion ​or personal experience alone.

​If you’re seeing a doctor for your cold symptoms, ​you don’t want her to just randomly guess ​that it’s the flu. ​The higher the stakes, ​the more evidence you need to support your decision. ​Meanwhile, if you’re deciding whether to recommend mittens ​instead of gloves to a customer, ​you don’t need as much information ​because the cost of making the wrong decision is low. ​No one’s health is at stake, ​just their chance at finger coziness.

When to trust your gut

When you’re essentially relying on limited historical data points.

Intuition helps you:

• Make quick, low stakes decisions

• Avoid searching through data without any idea of what you might find

Example

Say you want to increase your revenue at a small business, an exotic pet store.

If you’re able to do this, you might be able to:

• Open a new store in your city

• Improve employee benefits

• Offer greater variety of fish

You have some options you’re looking at for increasing revenue:

• Adding more reptiles

• Staying open 2 extra hours per day

• Raising prices on animal feed

How do you choose the best one? What information can you use to make your choice?

• Chance: One option is to leave it to fate. Flip a three-sided coin.

• Intuition: Well, maybe grandma has been running ​this exotic pet shop since 1987 ​and she remembers times like these, ​and she’s absolutely certain ​that offering more reptile varieties is the way to go. ​

• Evidence-based:

  • Define problem and outcome.
  • Gather information.
  • Synthesize information.

Evidence-based:

• Gather data on reptile varieties

• Experiment with later store hours

• Conduct surveys on higher prices

Maybe you can try the reptile variety approach first ​based on grandma’s intuition. ​And if that doesn’t pan out, ​investigate the other two options.

Sometimes you can make the wrong decision ​and still get the right outcome or vice versa. 

​There’s a little luck in every decision. ​The goal of evidence-based decision-making ​is to maximize your chances of getting the right outcome ​by accumulating the best evidence.

[A history of data analytics]

The history of data analytics is grounded in the concept of continuous improvement.

Continuous improvement:

• Process of trying to improve products, services, and processes

• Adapt to the evolving competitive landscape around you

• Staying the same rarely leads to long-term survival

Data analytics in the military

Data analytics in baseball

Key trends

• Explosion of available data

• Rapid advances in computing power

• Growing interest in evidence-based decision making

Data analytics is everywhere

Tech:

• Recommend products

• Optimize user experiences

Retail:

• Manage inventory and pricing

Healthcare:

• Improve patient outcomes

• Reduce costs

[Modern industry use cases]

Data analytics in entertainment

Data analytics in sports

Pattern of shots taken in 1997 vs 2019:

Data analytics in product design

HelloFresh is a meal-kit company.

Education and government

Use analytics to improve access to information and improve user experience.

Heatmap of clicks:

Continuous improvement

[Defining Data]

What is data?

Data is any information that can help you make a decision.

Data comes in many forms:

• Numbers

• Text

• Images

• Sounds

• Social media videos

• Voice recordings

• Profits from last year

Data gets generated just about anywhere:

• Whether you realize it

• Whether you record it

Eg: The taste of your morning cup of tea gives you information about how fresh the tea leaves are. That is data.

Eg: When you hear the sound of birds chirping, that might give you information about whether it’s dawn or dusk. That is data.

You can take the above data one step further and record it in order to analyze it.

Tracking the sun over the years

Our ability to generate and capture data has massively accelerated in the last few decades.

Thousands of years ago, ancient people tracked the position of the sun to determine the best times for planting and harvesting, but they had to do so using rock structures, like Stonehenge.

Now we can do the same thing with way less effort through satellite imagery and digital calendars.

Data across industries

Different industries generate different kinds of data.

Sports:

• Players

• Positions

• In-game statistics

Retail:

Transactional data about

• Sales

• Customer behaviour

Healthcare:

• Medical images

• Handwritten doctors’ notes

Social media:

• Ad views

• User interactions

Most industries also have data regarding:

• Payroll

• Website traffic

• Electricity costs

• Bank balances

• Legal documents

Purpose-driven data collection

Just because you can collect data about something doesn’t mean you should.

You should only collect data that serves a purpose.

Remember: Data is information that can help you make a decision.

You should filter through available info, and decide what’s most relevant for the problem at hand.

Bringing your unique perspective

Use your perspective to:

• Interpret data

• Find patterns and insights

• Tell a story

Just as an artist uses raw materials like clay or paint to create a masterpiece, you use data to craft a narrative that informs and inspires.

[Unstructured data]

Unstructured data

Unstructured data is data that doesn’t fit neatly into rows or columns.

When you snap a photo, record a video, ​or jot down notes in a journal, ​you’re creating unstructured data.

These are human types of information, ​and they feel very natural to us.

Collection and processing

Why does this distinction between structured ​and unstructured data matter?

It’s all about how the data gets collected and processed.

​At some point, all of this unstructured data ​usually needs to be converted into a structured format ​to be analyzed effectively.

Most analytics happens on structured data.

Customer reviews

Let’s say you’re working with customer reviews ​from a travel blogging site.

Reviewers can rate locations one through five ​on fun, accessibility, and value, then leave a comment.

[Structured data]

Structured data essentially exists for computers to store, process, and analyze.

Eg:

Data types

• Numerical data
  • Discrete
  • Continuous
• Categorical data

Time series and cross-sectional data

Time series data tracks changes over time.

Cross sectional data captures a snapshot in time.

Structured and unstructured data

Generative AI is making progress on unstructured data.

Non-AI techniques work best on structured data.

[Big data]

The v’s of big data

Volume:

• Data sets today are large

• Pose challenges in storage and computation

• Requires serious computational power

Eg: Amazon. Volume of orders: 12-19 million per day

Variety:

• 21st century has seen an explosion of unstructured data. Like images, text, video, and even augmented reality data. This explosion coincides with the rise of Internet and Social Media.

Velocity:

• Speed at which data is being generated

Eg: Sensors and satellites collect data every second. People in a storm’s path need data quickly.

Eg: Youtube. First 6 months -> 100k+ views/day. Velocity of data had consequences for content moderation.

Veracity:

• Veracity refers to the quality of the data.

• Crucial especially as volume, variety, and velocity increase

(Garbage in, garbage out)

Value:

• Data is woth analyzing if it provides some benefit

Eg: Netflix. Engagement data is used to generate recommendations. Without data, everyone would get the same recommendations.

Small data

Small data sets produce valuable insights.

Eg: Hospital data size.

There are only about 6000 hospitals in the US.

Each hospital might only serve a few 1000 patients per year.

An Intensive care unit might only have about 24 beds.

This data can be analyzed on your laptop. The data can be incredibly valuable for improving patient outcomes.

[Data Ecosystems]

The data ecosystem

Just like electricity, data doesn’t stay at the place it was generated. ​It flows through various systems to eventually power insights. This flow is caled the data ecosystem.

Collection: Gathering data

• Environmental sensors

• Websites tracking users

• Surveys for feedback

Storage: Keeping data safe

Processing (as raw format may not be ideal)

Analysis: Making sense of data

Delivery: Sharing insights (via a report or a dashboard)

Hospital Example

Consider a hospital that admits patients with severe unexplained symptoms, for diagnosis.

Each patient generates a lot of data about their health.

Eg: A Nurse will measure their vitals: temperature with a digital thermometer, and breathing patterns with a stethoscope.

They may also perform a blood or urine test, which has to be processed in a laboratory.

They might order imaging, such as an X-ray, ultrasound, or an MRI, which generate images that must be interpreted by an expert.

This data is stored in an electronic medical record (EMR) attached to the patient’s identity (along with data about when this information was collected and by whom).

You as a data analyst can then analyze patterns across thousands of records to:

• Surface insights that help doctors make better decisions

• Find a treatment protocol that leads to better outcomes

• Highlight risk factors that doctors are overlooking

[Collaborators outside your data team]

Here are some of your key collaborators outside the data team.

Business stakeholders

Their role: Make decisions based on insights

Your role:

• Understand their problems

• Provide them with insights to make informed decisions

Product managers

Their role:

• Develops roadmap for a product

• Implement planned features

Your role:

• Ensure your work aligns with their goals

Engineering teams

Their role:

• Build applications that serve your users

• Own the data collection systems

• Implement systems to gather data

Your role:

• Develop insights that can be integrated into the product

Other collaborators

Designer

Their role: Turn your data into beautiful user-facing experiences

Business analyst

Their role: Use your insights to guide high-level decision making

Note that the more mature the company, the more specialized your collaborators will be.

Adapting your work style

Team composition is dependent on the industry and organization size.

You’ll have to adapt your work style to the environment.

Small business:

• Likely a team of one

• Responsible for everything: data collection to analysis to visualization

• Main collaborator: business owner

• Directly accountable to them on all aspects of data strategy

Note that you might not have access to the same resources as in a larger organization.

The upside is that you can often see the direct impact of your work on the business.

Government role:

• Key collaborators: policy makers

• You might need to provide more context and guidance than in a business setting

You will likely have less access to complex engineering systems. The key challenge is to ensure insights resonate with policymakers.

Large tech company:

• Complex, well-established pipelines and a variety of specialists.
  • Data engineers: Build and maintain data infrastructure
  • Product managers: Ensure insights are aligned with customer needs
  • Software engineers: Integrate insights into products and services
  • Marketing and sales teams: Use data to optimize sales strategies

• Deal with large volumes of data and diverse stakeholders.

• Share insights across large, globally distributed teams

• Stay updated with latest technologies

[Collaborators on your data team]

Data-related responsibilities:

• Collection

• Storage

• Insight-finding with core methods

• Insight-finding with advanced methods

• Visualization

• Stakeholder communication

Division of responsibilities:

Hybrid roles

Draw on skills from software engineering (like Web developnment, Application development, Cloud computing).

Eg: Visualization engineer works at the intersection of Data Analysis and Software Engineering.

Eg: Machine learning engineer works at the intersection of Machine learning and Software engineering.

Acting as an intermediary

Data creator -> Data engineer -> Data analyst and/or Data Scientist -> Data consumer.

Each role needs to:

• Understand requirements and use cases

• Translate to domain of expertise

• Collaborate to deliver solution

Maturity and Specialization

In an early stage startup: One role for all data responsibilities.

In a mature data-driven organization: More specialized data roles.

[Introduction to large language models]

Large language models are a type of AI system designed to generate text.

Large Language Models (LLMs)

LLMs have learned how to repeatedly predict the next word through a process called pre-training, essentially reading vast amounts of text in books, articles, wikis, social media posts, etc from the internet.

LLMs also have additional training using human-curated data to answer questions in a friendly way while avoiding unethical responses.

The result of all this training is that an LLM is very good at generating text in response to an input question or prompt.

Luckily for data analysts, generating text means a lot of things: Summarizing a long email, fixing spreadsheet formulas, writing code to analyze data, etc.

Hence an LLM can be a thought partner and time saver throughout your workflow.

How LLMs work

What LLMs are good at

Writing: Generate text in response to a prompt.

Eg: Suggest some appropriate chart types.

Reading: Give a long piece of information and generate a short output.

Eg: Help me extract the core business problem from this stakeholder email.

[Choosing an LLM]

Available LLMs

Just like you might go to different colleagues with different types of questions, you have ​a choice in which LLM you work with.

Popular choices as of 2025:

• ChatGPT o3 from OpenAI

• ChatGPT 4o from OpenAI

• Claude 3.5 from Anthropic

• Llama 3.2 from Meta

• Gemini Pro from Google

Ways to access:

• Web interfaces

• Mobile apps

• Third-party program

• Programming language

Closed vs open LLMs

• The models and training methods for ChatGPT, Claude, Gemini aren’t available to the public. They’re called closed LLMs.

• Llama 3 is an open LLM. It’s code is available for anyone to download and customize.

Both closed and open models have their merits.

Closed models:

• Cost money for the best tier

• Outperform in response quality

• Less provocative responses

Open models:

• Free to use with own computer or cloud service

• Good but not state-of-the-art response quality

• More unpredictable responses

[Prompting LLMs]

Working with LLMs:

Two key skills:

Writing good prompts

• Be detailed and specific

• Guide the model to think through its answer

• Experiment and iterate

Recognizing the limits of the LLM

Guide the model to think through its answer

Goal:

• Brainstorm data cleaning techniques

  • Information
  • Spreadsheet formula
  • Emoji

Prompt:

Perform these steps:

1. Come up with five data cleaning techniques for data scraped from the web
2. Write a spreadsheet formula for each one
3. Add a fun relevant emoji

Experiment and iterate

• Get started quickly

• Check if the results are what you’re looking for

• Adjust the prompt to get closer to the ideal

1st iteration:

Help me brainstorm five data cleaning techniques for data scraped from the web.

2nd iteration:

Help me brainstorm five data cleaning techniques for data scraped from the web using a spreadsheet formula.

3rd iteration:

Help me brainstorm five data cleaning techniques for data scraped from the web using Google sheets with no custom functions.

[LLM Limitations]

Researchers often evaluate LLMs using a technique called benchmarking, testing each LLM on a standard set of questions ​to compare performance in specific areas.

In mid 2024, ChatGPT 4.0 scored a 53% on a General Knowledge Benchmark, 76% on a Math Benchmark, and 90% on a Coding Benchmark.

LLM Limitations

Key limitations when answering:

• Factual questions

• Questions about current events

• Math questions

• Questions that bring out historical biases

LLMs are not trained to report facts, they are trained to predict the next word.

LLMs often struggle with tasks where there’s only one right answer.

When LLMs are wrong, they are confidently wrong.

LLM biases

LLMs inherit biases from their training data.

Eg:

Prompt:

Write a story, 100 words or less, of an American person who pays the bill on a data with a romantic partner.

1st iteration:

Once upon a time, Mark and Sara went on a date … Mark paid for the date … The End.

Prompt:

Try again.

2nd iteration:

Once upon a time, Alex and Emma went on a date… Alex paid for the date … The End.

LLM Biases

Authors asked LLMs to write stories about

• Star students and struggling students

• Lawyers and defendants

• Person who pays for a date and person paid for

What an LLM might think?

It turns out in a study…

John: Pay for date -> 17,500; Be paid for -> 4000

Priya: Experienced software dev -> 0; New employee -> 490

Maria: Star student -> 333; Struggling student -> 4087

Bias distorts the fact that people of all genders and races can:

• Mentor others at work

• Succeed in the classroom

• Pay bills in a relationship

The text that LLMs are trained on represents our Present and our Past.

Each model is only a reflection of the data it is trained on.

LLM limitations for data analysts

Be mindful when working against the LLM’s training.

• Double-check responses

• Choose a tool that is better suited to the task. Like a search engine or a spreadsheet.

Approach the LLM’s response with skepticism.

• You are responsible for any LLM response you use in your work

Eg: Say an LLM says sales have went up by 42%, while in actuality sales have went down by 10%. In your report, you’re just as responsible for that information as if you had done the analysis yourself.

Be mindful of LLM biases.

Have a mindset of healthy skepticism.

Link to lecture notes: Link

[Module 2 introduction]

Module 2 outline

How spreadsheets organize data

• Importing data

• Setting up spreadsheets

Spreadsheet fundamentals

• Sort and filter data

• Write formulas

• Transform data

Data exploration with LLMs

• Prompting on LLM

Time series data

• Trends, seasonality, cyclicality

• Analytical methods

[Solving problems with data]

What outcomes am I interested in?

Eg:

Problem: Increase profit at a solar panel business

Outcome: Positive change in sales, Negative change in expenses

Data: Sales reports, Expense reports

Eg:

Problem: Improve patient outcomes

Outcome: Increase in patient satisfaction, Decrease in the length of hospital stays

Data: Patient surveys, Admission and discharge dates

What data provides context?

Consider the columns:

Who (Customer ID)

What (Product)

When (Purchase date)

Where (Region)

Outcome (Sale) <– This is the outcome of interest

The Who, What, When, Where data help contextualize sales data.

Eg: Consider increasing revenue for a solar panel company.

Suppose we are looking at the following questions:

• High paying customers?

• Specific products driving sales?

• How are sales trending?

• Sales vary across regions?

We hence see the contextual data is important.

[Spreadsheets for business analytics]

Why start with spreadsheets?

Industry standard tool

Opens in a few seconds

Free

Spreadsheet applications

Personal:

Tally ping pong score (On the simpler side)

Manage personal budget (On the complex side)

Business:

Schedule employee shifts (On the simpler side)

Develop project timeline

Draft quarterly financial report (On the complex side)

Spreadsheet applications

Designed for working with structured data

Rows have observations. (An observation is a single instance in your dataset, like one customer or one transaction).

Columns have features. (A feature is a characteristic that you measure for each observation, like age, price, or color).

Unstructured data in spreadsheets

Consider text, images, audio, video.

Spreadsheets can be used for collecting and organizing it, but their capabilities for analyzing this type of data are limited.

Are spreadsheets right for your use case?

1. Can your data be organized into rows and columns?

2. Are there relationships you want to explore between different aspects of the data?

   • Organizing expenses by category
   • Finding the month where you spent the most

We will often use Google Sheets.

[Navigating Google Sheets]

Google sheets alternatives

Google’s Sheets alternatives:

Microsoft’s Excel, Apple’s Numbers

Look at Google Sheets features.

Eg: Home Rennovation Project.

Formatting, Adding a new row or column, Entering a formula and autofill, Filters for sorting, Filters for filtering out.

[Importing Data]

Loading data into Google Sheets

The better way depends on your use case.

• Generate data directly in the spreadsheet. (Common for small scale, personal applications)

• Open an existing file. (When you’ve already been working with data on Google Sheets)

• Import a structured data set (.csv, .xlsx)

Eg: Look at the paper “Hotel booking demand datasets” by Antonio, Almeida, Nunes; published in “Data in Brief”.

Consider how to import the data from this paper onto Google Sheets.

• Go to “FIle” and then “Import”.

• Upload the .csv file.

• Settings:

  • Import location: Replace current sheet
  • Separator type: Detect automatically

(The Separator is a character like a comma or tab that separates values in the same observation.)

• Now the data will appear

Here is how the .csv file looks in a text editor:

Here is how the .csv file looks when loaded into Google Sheets:

(You can View -> Freeze the top row so that it is always visible.)

(You can click on the clock icon to look at any previous versions of the file.)