Mission: Find Zava's Hidden Product-Quality Crisis Using SQL MCP

Mission/open goal Description

I selected the Data Analyst mission: Find Zava's Hidden Product-Quality Crisis.

The goal was to investigate the PromptathonDb SQL database and identify the strongest evidence-backed product risk cluster by combining sales impact, support-ticket burden, customer satisfaction, recurring complaint themes, and vector-similarity evidence.

The agent performed a read-only investigation across SalesOrderLines, Products, SupportTickets, SupportChats, and Docs. It identified the Premium Short Sleeve Men's Top, SKU ZCPTM-SS-M-BW, as the strongest risk cluster because it combined meaningful revenue with the highest SKU-specific ticket volume, a satisfaction score of 1, and repeated complaints about connectivity and smart-fabric failure after washing.

The final business recommendation was to pause additional shipments of this SKU and begin a focused quality investigation into wash durability and embedded connectivity electronics.

Harness and model

Harness: GitHub Copilot Chat in Agent mode, running inside GitHub Codespaces. Model: Raptor mini, as displayed in the Copilot session. Tools and environment: The preconfigured sqlMcpServer MCP server was used to investigate the PromptathonDb SQL Server database. The agent used read-only SQL MCP tools including describe_entities, aggregate_records, read_records, and FindSimilarDocsByDocId.

Turn-by-turn journey

1. I asked the agent to read missions/data-analyst.md and SQL_DATABASE.md, create an investigation plan, and perform the mission in read-only mode without modifying any data.

2. The agent first called describe_entities to inspect the configured database entities, fields, parameters, and permissions.

3. It aggregated SalesOrderLines to compare revenue and quantity sold across product categories and individual products. The Premium category generated $190,553.87 from 2,207 units. SKU ZCPTM-SS-M-BW generated $19,825.05 from 228 units.

4. It analyzed SupportTickets by RelatedSKU, priority, status, and satisfaction score. The selected SKU had 9 linked tickets, including 1 Critical, 2 High, 3 Medium, and 3 Low-priority tickets. Six tickets were Closed and three were Resolved. Its average SatisfactionScore was 1.

5. It read SupportChats and parsed MessagesJson to identify recurring complaint themes. The support signal covered English, Spanish, and French conversations.

6. It analyzed Docs connected to the SKU. Fourteen documents were associated with it, with repeated complaints about wash durability, sensor connectivity, pairing failure, and smart features stopping after washing.

7. The agent selected negative document DocId 39 and executed FindSimilarDocsByDocId with TopN 5. The returned documents formed a coherent cluster around connectivity and smart-fabric failure after washing.

8. During the investigation, the agent initially used the singular entity name Product, which caused an EntityNotFound error. I redirected it to the configured plural entity Products.

9. The agent also attempted to switch to the MSSQL extension after the error. I instructed it to continue only with the required sqlMcpServer MCP tools.

10. A response was cleared because of a possible public-code match. The agent automatically retried with a modified response and continued the evidence-based investigation.

11. I asked the agent to verify its numeric comparisons and customer-segment claims before finalizing the report. It corrected and strengthened the executive brief using the exact SQL MCP results.

12. The completed investigation was saved as data-analyst-executive-brief.md and pushed to my GitHub fork.

Completion

• Yes, the agent completed the mission or goal.
• No, the agent did not complete the mission or goal.

Bonus work

I went beyond the basic final response by:

• Creating a reusable Markdown executive brief containing the risk cluster, quantitative evidence, complaint themes, vector-similarity findings, recommendation, confidence level, and limitations.
• Recording the exact SQL MCP tool calls used during the investigation.
• Asking the agent to re-check important counts and customer-segment conclusions before accepting the final report.
• Documenting important agent errors and corrections, including the incorrect Product entity name, attempted MSSQL-extension switch, and public-code-match retry.
• Creating a submission-ready Markdown document with a Mermaid architecture and data-flow diagram.
• Committing and pushing the completed artifacts to my GitHub fork.

Working artifact:
https://github.com/tanusomani7004-stack/sql-ai-promptathon/blob/main/data-analyst-executive-brief.md

Submission documentation:
https://github.com/tanusomani7004-stack/sql-ai-promptathon/blob/main/promptathon-submission.md

Start with missions/data-analyst.md, SQL_DATABASE.md, and the linked data-analyst-executive-brief.md to understand the completed investigation and its SQL MCP evidence. The issue describes a finished mission and pushed artifacts but does not identify a remaining documentation change or implementation target, so completion criteria for a new contributor are not defined.