1 · Descriptive analytics

An agent will answer a question about your data in seconds. Your job is to decide whether that answer survives contact with the real distribution — then prove one of them by hand in a pivot table.

At a glance

Time	~65 minutes, in pairs
Work with	A partner
You need	An agent (Claude or Manus) + a spreadsheet (Google Sheets or Excel)
You hand in	Three verified insight sentences + one pivot table (Stage 3 Summary and Verification)
Graded on	Whether every average is labelled mean or median, every mean–median gap is explained, and every insight sentence names the evidence it stands on

The scenario

Kopi Kembali’s marketing lead has three questions about the social feed in kk-social-posts.csv — the same file as Unit 02, so your profiling notes carry over:

1. Which platform earns the best engagement?
2. Which content format performs best?
3. When is the best time to post?

An agent will answer all three in seconds. Your job is to decide whether the answers survive contact with the actual distribution, then rebuild one of them yourself in a pivot table.

Data dictionary

Column	Meaning
post_id	Unique post id
platform	Instagram, TikTok, Facebook, LinkedIn, Twitter or YouTube
post_date, post_time, day_of_week	When the post went out
content_type	video, carousel, static image, text, reel, etc.
likes, comments, shares	Engagement actions
views	Views or impressions for the post
follower_count	Platform followers at the time
engagement_rate_pct	A pre-calculated rate
hashtag_count	Hashtags used
sentiment	Sentiment tag applied by an earlier tool

Get your data

• Download kk-social-posts.csv

Step 1 — Direct the agent · 20 min

Ask the three questions below one at a time — wait for each answer before moving on. For every one, require the summary statistic to be named.

🔗 On your agent platform

Open Claude or Manus and upload kk-social-posts.csv. If you are working in Google Sheets, the built-in Gemini side panel also works natively, depending on your account tier.

Question 1 — engagement by platform:

Profile engagement in this dataset.
Group by platform.
For each platform, show:
- The number of posts
- The mean engagement rate
- The median engagement rate
- The 75th percentile engagement rate

Sort by median engagement rate, highest first.

Question 2 — format performance:

Compare content performance by type.
Group by content type (or format: video, carousel, static image, reel,
etc.).
For each content type, show:
- The count of posts
- The median likes per post
- The median shares per post
- The 75th percentile of engagement rate

Rank by median engagement rate.

Question 3 — best posting time:

Analyse posting times.
Extract the hour of day from the posting timestamp.
Group by hour.
For each hour, show:
- The count of posts
- The median engagement rate
- The mean engagement rate

Highlight the three hours with the highest median engagement.

Step 2 — Interrogate the summaries · 15 min

For every answer, ask:

• Is this a mean or a median — and does the gap between them tell you the distribution is skewed?
• Is one extreme post carrying an entire average?
• Does the pattern hold inside each platform, or does the pooled number blend groups that behave differently (a Simpson’s Paradox check)?
• Did the agent pull a raw column for “engagement rate,” or apply its own formula? Same question for any aggregation.
• Is a misleading denominator hiding in any of the three answers (a rate divided by the wrong base)?

Spot the distortion

Skew, outliers, aggregation, and a misleading denominator can each make a clean-looking summary wrong in a different way. A mean–median gap larger than double is a flag, not an explanation — you still have to say why.

Step 3 — Verify with a pivot · 20 min

Pick one of the three answers above and reproduce it by hand.

1. Open kk-social-posts.csv in Google Sheets or Excel. Select all the data (including headers) and insert a pivot table.
2. Build the pivot to match your chosen question:
   • Q1 (platform): platform to Rows; engagement_rate_pct to Values twice — once as Average, once as Median. Check the row count against the agent’s post count.
   • Q2 (content type): content_type to Rows; likes to Values (Median and Mean); shares to Values (Median); engagement_rate_pct to Values (Median).
   • Q3 (posting time): add a helper column extracting the hour from post_time; drag it to Rows; engagement_rate_pct to Values (Median and Mean); sort by median, descending.
3. Compare your pivot figure to the agent’s figure for each key metric. A difference is not a failure; an unexplained difference is.

If there’s a mismatch, investigate before you move on — the agent may have used mean instead of median, included a different date range, or defined the grouping column differently.

Step 4 — Write the sentences · 10 min

For each of the three questions, write one insight sentence — not a statistic, a decision someone could act on.

INSIGHT (Question ___: ______________________________)
What the agent found: _______________________________
Why it matters — who acts on this, and for what decision: ______
What I checked: pivot-verified / trusted the agent — any oddities: ___
Insight sentence (one sentence, actionable):
______________________________________________________
Verification decision: accepted / revised / rejected — because ___

Repeat for all three questions. “Carousels earn the highest typical engagement (state the median actions per post)” is actable. “Video is best” is not.

Verify & submit

• Every reported average is labelled mean or median
• Any mean–median gap larger than double is explained, not ignored
• The pivot figure and the agent figure reconcile, with the reason for any gap written down
• You checked whether the best posting time holds within platforms, not only overall
• Each insight sentence names the evidence it stands on
• Reflection questions completed: what surprised you, where verification changed your mind, and what this data does not answer

Then submit:

1. The pivot table — in your team’s Google Sheets Analysis Workbook or linked Drive evidence folder; link the exact tab.
2. The three insight sentences — add to your individual portfolio (Stage 3 Summary and Verification, evidence for Unit 5).
3. The checkpoint record — submit your individual Unit 5 xSiTe checkpoint, linking the exact Sheet tab and your personal contribution.

Capstone connection

In Stage 3 of your capstone, apply the same discipline to your own dataset: name every statistic as mean or median, check whether a pooled pattern survives a subgroup split, and verify at least one figure by hand before it enters your analysis. Document it in your orchestration log under Stage 3.

Unit 05 app · lesson plan