Write the findings and deliver

Step 1: Open the narrative template

The numbers are verified. The charts are built. None of it matters until Naledi can read it and make a decision.

Open materials/narrative-template.md. This is the structure for the deliverable — the document Naledi will actually use. It has four sections: an executive summary (2-3 sentences with the headline findings), key findings (3-4 bullet points with specific numbers), the charts, and a recommendation grounded in the data.

Read it through before directing AI to write anything. The template is not a form to fill in mechanically — it is a communication structure. The executive summary leads with the most important finding. The key findings support it with evidence. The recommendation tells Naledi what to do and why. Every number in this document must match what you verified in Unit 3.

Step 2: Direct AI to draft the narrative

This is where prompt context matters. If you tell Claude "fill in the narrative template," it will produce something generic — plausible sentences with round numbers pulled from its training data, not from Naledi's verified metrics. The prompt needs to carry the context that makes the output accurate.

Direct Claude with something like:

Using the narrative template in materials/narrative-template.md, draft the analysis narrative for Naledi. Use these verified values: net revenue by channel (retail ZAR 5,326,500, commission ZAR 3,761,500, online ZAR 2,486,000), profit margins (retail 28.2%, commission 65.5%, online 59.2%), and revenue per workshop-week (shelving ZAR 8,622, dining tables ZAR 6,880). The executive summary should lead with the most profitable channel and the most efficient product type. Key findings should include specific ZAR amounts. Refer to the metric definitions in materials/metric-definitions.md for how each metric is defined.

Notice what the prompt does: it gives Claude the specific numbers, names the template structure, and points to the metric definitions. A prompt that says "write the narrative" leaves Claude to guess the numbers. A prompt that includes the numbers constrains the output to verified values.

Step 3: Review AI's draft for accuracy

Read what Claude produced. This is a review, not a skim. Check three things:

Does the executive summary lead with the right finding? The most profitable channel (commission at 65.5% margin) and the most efficient product type by workshop-week (shelving at ZAR 8,622/week) should be named explicitly. If the summary buries these behind generic statements like "the analysis revealed several insights," that is not useful to Naledi.

Do the key findings use the correct numbers? Every ZAR amount and percentage should match the verified values from Unit 3. AI sometimes rounds differently, swaps metrics between findings, or generates numbers that look right but are not the ones you verified. Check each one.

Does the recommendation follow from the data? If the recommendation says "focus on online sales" but the data shows online has the lowest total revenue and moderate margins, the recommendation contradicts the evidence. The recommendation should connect to the numbers above it — not to what sounds like good business advice in general.

Step 4: Fix and finalize

If anything failed the review, direct Claude to fix the specific issue. Be precise:

The executive summary says "retail is the strongest channel" but the data shows commission has the highest margin at 65.5%. Rewrite the executive summary to lead with commission as the most profitable channel and shelving as the most efficient product type per workshop-week.

This is the pattern for every analytical deliverable going forward: AI drafts, you review against the verified numbers, you direct corrections. The deliverable is not done when AI says it is done. It is done when every number matches and every claim follows from the data.

Once the narrative is accurate, embed the four charts from Unit 4 in the Charts section of the notebook. The complete notebook should read as a single document — summary, evidence, charts, recommendation — that Naledi can follow from start to finish.

Step 5: Deliver to Naledi

This is the project's delivery moment. Send the completed analysis to Naledi through the chat. On this platform, "delivery" means pasting the executive summary into the chat panel and attaching the notebook file -- the same way you would send a deliverable to a real client over email or Slack. Include the executive summary in your message so she can see the headline findings immediately, and let her know the full notebook with charts and detailed findings is attached.

Naledi responds warmly. She is impressed to see the numbers laid out clearly — this is exactly what she needed for planning next year. She reacts specifically to the revenue-per-workshop-week finding: "I never thought about it that way — I always looked at the sale price, not what it costs in workshop time. The shelving numbers are eye-opening."

Then she asks a small follow-up: "Could you also check whether there's a seasonal pattern? December felt slow last year but I'm not sure."

This is a natural business question — the kind that comes up in every client conversation once someone sees data for the first time. You can address it briefly using the monthly trend chart from Unit 4, or note it as future work. Either way, the original scope is complete. The analysis she asked for is delivered.

Step 6: Push to GitHub

The analysis is done and delivered. Now close the project properly. Write a README for the project repository that describes what was analyzed and what was found. The README should cover:

• What the project is (sales analysis for Khumalo & Grain, a Cape Town furniture workshop)
• What was analyzed (twelve months of sales data across three channels)
• Key findings (which channel is most profitable, which product type is most efficient per workshop-week)
• Tools used (Python, DuckDB, Jupyter, matplotlib)

Direct Claude:

Write a README.md for this project. Summarize what was analyzed, the key findings, and the tools used. Keep it concise — someone reading it should understand what this project does and what it found.

Review the README the same way you reviewed the narrative — do the findings match the verified numbers? Then commit and push:

git add . git commit -m "T5: complete analysis narrative and project delivery" git push origin main

The project is delivered. Naledi has her analysis. The work is documented and stored. This is what closing a project looks like — the client has what they need, and the repository is a complete record of the work.