Mission 1: ZCPTM-SS-M-BW product-quality investigation

[NobukoIT]

Promptathon Mission 1 — Submission Journey

1) 目的

Investigate product-quality risk for Zava and produce evidence-backed recommendation.

2) High-level workflow

• Use SQL MCP to enumerate entities and confirm schema.
• Use sqlcmd to run aggregation SQL against PromptathonDb.
• Inspect Docs and SupportChats for qualitative evidence.
• Run the MCP custom tool FindSimilarDocsByDocId for vector similarity.
• Consolidate quantitative + qualitative evidence and prepare executive brief.

3) Prompts and agent interactions (high-level)

• "List available SQL MCP entities." → used the MCP describe_entities call.
• "Show top product categories by revenue." → executed SQL via sqlcmd.
• "Which SKUs have most support tickets and low satisfaction?" → SQL aggregation on SupportTickets.
• "Show docs and run vector similarity for DocId=39." → executed find_similar_docs_by_doc_id.

4) MCP tool calls used

• describe_entities — to list Customer, Doc, Employee, FindSimilarDocsByDocId, Product, SalesOrder, SalesOrderLine, SupportChat, SupportTicket.
• read_records attempts were made but the working path used direct sqlcmd queries for full control and to capture raw outputs.
• find_similar_docs_by_doc_id executed with DocId=39, TopN=5 to validate similarity clustering.

5) SQL queries executed (examples)

• Category revenue:

SELECT TOP 15 ProductCategory, SUM(LineTotal) AS Revenue, SUM(Quantity) AS QuantitySold, COUNT(*) AS OrderLines
FROM dbo.SalesOrderLines
GROUP BY ProductCategory
ORDER BY SUM(LineTotal) DESC;

• Top SKUs:

SELECT TOP 20 SKU, ProductName, ProductCategory, SUM(LineTotal) AS Revenue, SUM(Quantity) AS QuantitySold, COUNT(DISTINCT OrderId) AS Orders
FROM dbo.SalesOrderLines
GROUP BY SKU, ProductName, ProductCategory
ORDER BY SUM(LineTotal) DESC;

• Support tickets by SKU:

SELECT COALESCE(RelatedSKU,'UNKNOWN') AS RelatedSKU, COUNT(*) AS TicketCount, AVG(CAST(SatisfactionScore AS FLOAT)) AS AvgSat,
SUM(CASE WHEN Priority='High' THEN 1 ELSE 0 END) AS HighPriority,
SUM(CASE WHEN Status='Open' THEN 1 ELSE 0 END) AS OpenCount
FROM dbo.SupportTickets
GROUP BY COALESCE(RelatedSKU,'UNKNOWN')
ORDER BY COUNT(*) DESC;

• Docs for SKU:

SELECT DocId, SourceType, SourceId, RelatedSKU, Title, LEFT(Body,200) AS BodySnippet, TagsJson
FROM dbo.Docs
WHERE RelatedSKU='ZCPTM-SS-M-BW'
ORDER BY DocId;

• Vector search (MCP stored procedure):

EXEC dbo.FindSimilarDocsByDocId @DocId = 39, @TopN = 5;

6) Errors, fixes, and notable troubleshooting

• Error: sqlcmd SSL/TLS certificate verify failed (self-signed certificate). Fix: add -C (Trust Server Certificate) option to sqlcmd and to the Python helper's trust flag.
• Attempted MCP read_records calls with certain entity names returned EntityNotFound errors in that path, so direct sqlcmd queries were used for robust, copyable outputs.
• Confirmed DB name PromptathonDb and table list in INFORMATION_SCHEMA.TABLES before running aggregations.

7) Key findings (final)

• Risk cluster: ZCPTM-SS-M-BW (Premium Short Sleeve Men's Top). Multiple lines of evidence:
  • Commercial: SKU is among top revenue-contributing SKUs (Revenue ~ $19.8k, QuantitySold=228, Orders=105).
  • Support: 9 support tickets for this SKU, average satisfaction ≈ 1.67 (low), includes high/critical priorities.
  • Qualitative: Multiple reviews and support chats in English, Spanish, French describing failure of smart fabric connectivity after washing.
  • Vector search: Representative negative review (DocId=39) yields similar negative reviews in top results.

8) Artifacts

• Notebook: artifacts/ZCPTM_issue_notebook.ipynb (contains queries, captured outputs, and executive brief).

9) Suggested submission contents

• The notebook artifact.
• This submission_journey.md describing prompts, tool calls, SQL commands, errors/fixes, and findings.
• A short executive brief (already embedded in the notebook).

10) Next steps

• (Optional) Add charts and failure-rate computations by matching shipped quantities (requires POS/shipments data).
• Finalize GitHub issue with the above attachments and a short narrative of the journey and decisions made.

[NobukoIT]

@microsoft-github-policy-service agree