Week 1 · Foundations of Data Analytics and Visualisation

Run your first task on an AI agent, then look hard at the output and list three things you are not 100% sure are correct. This is the core skill of the whole module: working with agents and staying sceptical.

At a glance

Time	60–75 minutes, in class
Work with	A partner
You need	An agent (Manus, or Claude Code as your instructor confirms) + the dataset
You hand in	A screenshot/export of the agent’s output + this worksheet’s “What I Cannot Yet Verify” record
Graded on	Whether you noticed and recorded genuine uncertainties, not whether the answer is “correct”

The scenario

Kopi Kembali’s operations manager has exported one week of orders (Monday 3 to Sunday 9 August 2026) from six outlets and wants a quick read: how did the week trend, and which drinks carried it? The export is exactly as the till system produced it — nobody has looked at it yet. It is real data with real messiness: typos, inconsistent formats, a few missing values.

Data dictionary

Column	Meaning
order_id	Till receipt id
outlet	Which outlet took the order
order_date	Date of the order
drink	Drink name
size	Regular or Large
qty	Cups in the order
unit_price_sgd	Price per cup, SGD
payment_method	PayNow, Cash, Card or GrabPay
total_sgd	Order total as the till recorded it

Get your data

• Download kk-orders-week.csv

Part 1 — Prepare your task · 5 min

Do not clean the file yourself. The agent will do that. With your partner, choose one question to ask about the data:

• What is the overall trend over time?
• Which category has the highest total or average?
• Are there any obvious gaps or quality problems in the dataset?
• Rank the items by their highest value.
• What does a summary by one column look like?

Write your question in one clear sentence. Make sure you both understand what you are asking.

Part 2 — Run the task · 15 min

🔗 On your agent platform

Go to Manus (or Claude Code, if your instructor has directed you there) and start a new task. Upload or paste in kk-orders-week.csv.

Write a clear, specific instruction — profile first, clean what needs cleaning, then answer your question, showing the numbers. Two worked examples to adapt:

Analyse this social media dataset. First, clean it: fix any date format issues,
standardise category names, handle missing values sensibly. Then rank posts by
engagement rate and tell me the top three performers. Show me the numbers.

This is customer data. Profile it first: list all columns, note missing values,
flag any obvious errors. Then summarise total spending by customer segment.
Show the breakdown.

Run the task and let the agent work — it usually takes 30 seconds to a few minutes. Agents work faster when you are not watching, so step away rather than hovering.

Part 3 — Look hard at the output · 25 min

Read through everything the agent produced carefully. Professional analysts use three checks to validate agent output — apply them here.

Source verification

• Can you trace the answer back to the actual data? Ask the agent to show you the specific table, query, or calculation it used.
• Does the row count or row subset make sense? (If the agent reports 1000 results from a dataset you uploaded with far fewer rows, something is wrong.)
• Is the agent pulling from the right version of the data or the right time period?

Reasonableness check

• Do the numbers fall within logical business ranges for a kopi stall?
• Are there any sudden spikes or drops that might signal an error? A 10x change is a red flag worth investigating.
• Are there any suspiciously round or clean numbers? Real data is usually messy.

Beyond the numbers

If the agent cleaned data (filled missing values, standardised names, etc.), can you see what it changed and why? If it built a chart, does the chart match the underlying numbers, or does the visual mislead? Did it hedge appropriately — did it say “we found X” or “the data suggests X”? Did the agent make any choices, like dropping rows or filling values, that you would question? That is exactly what goes on your list.

Part 4 — Write your “What I Cannot Yet Verify” list · 25 min

This is the most important part. You are not trying to find the “correct” answer. You are learning to spot what you do not know. For each of three items, write the thing you are unsure about and briefly explain why — do not research it or ask the agent to explain, just notice and record it.

Thing 1:
What is one thing about the agent's output that you are not 100% sure is correct?


Why are you not sure?


Thing 2:
What is one other thing that looks a bit odd or that you did not expect?


Why does it make you uncertain?


Thing 3:
What is one more thing you would want to double-check if you had to stand
behind this output?


Why would you want to check this?

What a strong answer looks like

“The agent filled missing values with the average. I know the average is 42. But I do not know if filling with the average was the right choice, or if the agent should have dropped those rows. I cannot decide if the result is trustworthy.” Notice: strong answers do not require maths — they are careful reading and noticing.

Verify & submit

• I have filled in my question (Part 1)
• I have run the task and saved the output (Part 2)
• I have reviewed the agent’s output against the source-verification and reasonableness checks (Part 3)
• I have written three genuine things I am not sure about, with reasons (Part 4)
• I can explain to my partner why I am unsure about each one

Then submit to your individual portfolio:

1. A screenshot or export of the agent’s output — the cleaned data, the summary, the chart, or the text.
2. This completed worksheet, especially the “What I Cannot Yet Verify” list.

Unit 01 app · lesson plan