The analytical narrative and project close

Step 1: Structure the analytical narrative

An analytical narrative is a professional argument. Not a list of findings, not a chronological account of what you analyzed in what order. The structure matters because the reader -- Siobhan taking this to her board -- needs to follow an argument to a conclusion, not reconstruct your analytical journey.

The structure:

1. Conclusion first. Verdant Packaging's food container line appears profitable on revenue but is the least profitable when material costs and batch failure waste are included. PLA resin costs and a 22% batch failure rate erode the margin.
2. Supporting evidence. Data from four sources, cross-source profitability analysis, the OEE hierarchy showing quality as the bottleneck on LINE-A.
3. Alternative explanations considered and addressed. Could the food container line's lower profitability be seasonal? Could the waste numbers be inflated by a data quality issue in the lab reports? Address each and explain why the evidence supports the primary conclusion.
4. Forward-looking recommendation. Based on the leading indicator relationship between PLA moisture and batch failures, what data should Siobhan collect next, and how much data would be needed to confirm the relationship with confidence?

This is argument order, not chronological order. AI generates findings in the sequence you analyzed them -- "first I loaded the data, then I profiled it, then I defined metrics." The board does not care about your process. They care about the conclusion and whether it holds up.

Step 2: Direct AI to draft the narrative

Direct AI to produce a first draft:

Write an analytical narrative for Siobhan Murray's board at Verdant Packaging. Structure: conclusion first, then supporting evidence from the four data sources, then alternative explanations addressed, then forward-looking recommendations. The key finding is that the food container line is least profitable due to PLA resin costs and batch failure waste. Include the OEE hierarchy breakdown showing quality as the bottleneck.

Review what AI produces. AI will likely generate findings in chronological order -- the order the analysis was performed rather than the order that builds an argument. Look for these common problems:

• Findings presented as a flat list instead of building toward a conclusion
• Alternative explanations missing entirely
• The conclusion buried at the end instead of stated up front
• Uncertainty not acknowledged -- claims presented as certainties without noting data freshness gaps or sample size limitations

Restructure the draft. Move the conclusion to the front. Organize evidence by argument strength, not by when you discovered it. Add the alternative explanations section if AI omitted it.

Step 3: Power analysis

Power analysis answers a forward-looking question: how much data would you need to confirm the PLA moisture-batch failure relationship with confidence?

You observed that when PLA moisture exceeds 4%, the food container batch failure rate roughly triples. But how much data would you need to detect this effect reliably? Power analysis tells you.

Direct AI to calculate:

Run a power analysis for the relationship between PLA moisture content (above vs below 4%) and food container batch failure rate. Assume the observed effect size from the current data. What sample size would be needed to detect this effect with 80% power at a significance level of 0.05?

The power analysis becomes part of the recommendation. You are not just reporting what happened -- you are telling Siobhan how much data to collect going forward to confirm the pattern. This is a professional contribution that AI does not offer unless explicitly directed. AI analyzes the data you give it. It does not spontaneously recommend future data collection.

Include the power analysis recommendation in the narrative: "To confirm the PLA moisture-batch failure relationship with statistical confidence, Verdant Packaging should collect [N] additional batches of production data with moisture readings. At the current production rate, this would take approximately [X] weeks."

Step 4: Final cross-model review

Run cross-model review on the complete narrative, not just the profitability numbers. Provide the full argument to a fresh AI session:

Review this analytical narrative for Verdant Packaging's board. [Paste the full narrative.] Does the argument hold together? Are there alternative explanations I haven't considered? Is the evidence sufficient for the conclusions drawn? Flag any claims that are not well-supported by the data.

Compare the second AI's assessment against the first's. The narrative is the final deliverable -- cross-model review at this stage catches argument-level problems that line-level review in Unit 4 would not. Does the second AI agree with the conclusion about the food container line? Does it flag any alternative explanations the first AI missed? Does it question the power analysis assumptions?

Document any changes you make based on the cross-model review.

Step 5: Deliver to Siobhan

Send the narrative and dashboard link to Siobhan. She reads the argument. She evaluates: does the conclusion match her operational intuition? Are the recommendations actionable? Is the uncertainty communicated honestly?

If the narrative is well-structured with alternative explanations addressed, Siobhan responds: "This is exactly what I needed -- I can take this to the board. The food container line needs work and now I can see why."

The metric hierarchy documentation is part of the deliverable. Siobhan's team needs to maintain these definitions going forward. OEE, its three components, the leading and lagging indicators, the cascade relationships -- all documented so the next person who touches these metrics knows what depends on what.

Step 6: Project close and GitHub push

Commit your work and push to GitHub. The repository should contain:

• The Jupyter notebooks with your analysis
• The DuckDB views powering the dashboard
• The metric hierarchy documentation
• The analytical narrative
• The CLAUDE.md governance file with all the rules and definitions accumulated across the project
• A README describing what was built, what was found, and what data was used

git add -A && git commit -m "P7 complete: multi-source quality assessment with metric hierarchies and analytical narrative"

git push origin main

The project is complete. You pulled together four data sources in four formats, designed a metric hierarchy where changing one definition cascades to others, built a dashboard that displays the hierarchy for operational decisions, and delivered an analytical narrative structured as a professional argument.