Spreadsheets and Data Visualisation with WorkBuddy

Last checked and updated: September 3, 2026

WorkBuddy can read spreadsheets, combine files, calculate metrics, generate charts, and lay out reports. The official data-analysis practice shows one prompt producing research, structured data, visualisation, and a report. That breadth is useful, but a single hidden chain also makes errors harder to locate.

The safer sequence is profile → define → clean → calculate → reconcile → visualise → explain.

Practice dataset

Create a copied CSV with these columns:

date,channel,leads,qualified,spend
2026-08-01,Search,42,12,840
2026-08-01,Newsletter,18,9,120
2026-08-08,Search,51,15,1020
2026-08-08,Newsletter,20,11,140

Save it as campaign-copy.csv in a practice workspace.

Step 1: profile without editing

Inspect campaign-copy.csv in Ask mode. Do not edit or create files.
Report row count, columns, inferred types, missing values, duplicates, invalid negatives,
date range, category values, and any ambiguity that affects calculation.

Manually confirm row count and totals. A four-row sample is intentionally easy enough to audit.

Step 2: define the metrics

Do not ask for “insights” before defining the maths:

Use these definitions only:
qualification_rate = qualified / leads
cost_per_lead = spend / leads
cost_per_qualified = spend / qualified
For any zero denominator, return blank and flag it instead of dividing.
Round displayed currency to 2 decimals, but calculate totals using unrounded values.

Metric definitions belong in the output. Otherwise two people can read the same label and calculate different things.

Step 3: plan the outputs

Ask Plan mode for:

• campaign-clean.csv preserving all source columns.
• campaign-summary.csv grouped by channel.
• campaign-analysis.md with methods, findings, limitations, and source filename.
• One chart comparing qualification rate and cost per qualified lead by channel.
• cleaning-log.md listing every changed or rejected row.

Require it to keep campaign-copy.csv unchanged.

Step 4: reconcile before charting

Check these identities:

sum(source leads) = sum(clean leads) + sum(rejected leads)
sum(source qualified) = sum(clean qualified) + sum(rejected qualified)
sum(source spend) = sum(clean spend) + sum(rejected spend)

For the sample, total leads are 131, qualified are 47, and spend is 2,120. If WorkBuddy’s result disagrees, do not continue to the chart.

Step 5: review the chart

Ask:

• Does the chart use the defined aggregation?
• Are percentages and currency distinguished?
• Is the scale misleading?
• Are missing values visibly different from zero?
• Can every plotted value be found in campaign-summary.csv?

A dual-axis chart may make unrelated changes look correlated. Prefer two small charts when the scales or meanings differ.

Prompt for a manager-ready report

Using only campaign-summary.csv and the documented metric definitions,
write a one-page manager brief.
Separate Observations from Hypotheses.
Do not call a channel “better” unless you name the metric and trade-off.
Include the exact date range, row count, total spend, and three limitations.
Place the source filename beneath the chart.

Model and credit discipline

The official task guide currently recommends MiniMax for Excel, data analysis, and PPT work. Treat that as a starting suggestion. Test correctness and credits on a sample before sending a large workbook. Trim irrelevant sheets, name the required ranges, and split profiling from final generation so an error does not force a complete rerun.

Success check

The workflow passes when the cleaned data reconciles to the source, every formula is defined, chart values trace to the summary table, observations are separated from hypotheses, and the original file remains unchanged.

FAQ

Can WorkBuddy handle Excel as well as CSV?

Official documentation lists Excel among supported office artefacts and describes spreadsheet skills. CSV is easier for a first audit because the raw rows are transparent. Move to Excel after the workflow passes on a sample.

Should I let it “fix” missing values automatically?

Only with an explicit rule and cleaning log. Blank, zero, unknown, not applicable, and not collected mean different things.

Can I trust a total shown in a chart?

Trace it to the summary table and recalculate a sample. Rendering correctness is separate from arithmetic correctness.

Quick self-check

Q1. Why calculate with unrounded values and only round the display?

Answer: Repeated rounding can make totals and ratios drift from the source data.

Official sources

• Data analysis and visualisation practice
• Task-bar model guidance
• Results and artefacts

What’s next

Build a research brief whose claims preserve their sources in L07: Research, Sources, and Knowledge.