Add GraphRAG corpus visibility diagnostics for internal-fallback indexing/query mismatch

[Copilot]

• Inspect failing CI runs and identify actionable pipeline errors on this PR branch.
• Fix the MCP config validation script so the default config path matches the repository’s current appsettings format.
• Fix the markdownlint failure in the touched documentation content.
• Run targeted validation for the script and markdown lint/build paths.
• Run code review/security checks, then reply to the PR comment with the addressing commit.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>C64CityBuilder GraphRAG local-text visibility diagnosis</issue_title>
<issue_description>Observed on March 11, 2026.

This issue captures diagnosis only. It intentionally excludes fixes, remediation, and implementation steps.

Workspace under test

• E:\github\C64CityBuilder

Summary of symptoms

• Local staged PRG TXT files were present under E:\github\C64CityBuilder\mcp-data\graphrag\input\docs\prg.
• Commodore_64_Programmers_Reference_Guide.txt contained verified exact phrases, but exact filename and exact snippet queries returned no local hits from MCP context or GraphRAG.

Evidence

• Local staged input confirmed: E:\github\C64CityBuilder\mcp-data\graphrag\input\docs\prg\Commodore_64_Programmers_Reference_Guide.txt
• Local staged input confirmed: E:\github\C64CityBuilder\mcp-data\graphrag\input\docs\prg\Mapping_the_Commodore_64.txt
• Local inspection confirmed exact phrases in Commodore_64_Programmers_Reference_Guide.txt including: Video Bank Selection; MCS6510 Microprocessor Instruction Set; KERNAL Power.Up Activities; The COMMODORE 64 PROGRAMMER'S REFERENCE GUIDE has been.
• Exact MCP context and GraphRAG queries returned no hits for Commodore_64_Programmers_Reference_Guide.txt, The COMMODORE 64 PROGRAMMER'S REFERENCE GUIDE has been, Mapping_the_Commodore_64.txt, and helpyouusethePEEKandPOKEinstructionstoextendyour.

Observed server behavior

• MCP GraphRAG status reported enabled=true, state=ready, isIndexed=true, backend=internal-fallback, lastIndexedDocumentCount=43, and lastIndexDurationMs approximately 1 to 4 ms.
• Forcing POST /mcpserver/graphrag/index updated lastIndexedAtUtc but still reported 43 documents and an approximately 4 ms duration.
• An unrelated query like Color RAM returned an external C64-Wiki result instead of the local book content.

Why this appears to be an indexing or query visibility problem and not only a source-content problem

• A locally staged TXT with verified clean exact phrases remained undiscoverable.
• GraphRAG continued to report ready/indexed state before and after forced reindex.
• The reported document count remained fixed at 43.
• Index duration remained near-instant while local content stayed invisible.
• Missing OCR dependencies can explain some PDF extraction limitations, but they do not by themselves explain why already-staged local TXT content was not query-visible.

Separate source text quality note

• Mapping_the_Commodore_64.txt also has source text quality issues, including noisy or gibberish text near the beginning and collapsed spacing/run-on text such as helpyouusethePEEKandPOKEinstructionstoextendyour.

Impact on users and workspaces

• Users and agents can receive false confidence from ready/indexed status while workspace-local staged material remains undiscoverable.
• External sources may dominate answers even when local documents are present.

Open diagnostic questions

• Are TXT files under mcp-data/graphrag/input/docs/prg included in the effective corpus for this workspace?
• What corpus produced the fixed 43-document count?
• Does the internal-fallback backend read the staged input tree or some other store?
• Are exact-filename and exact-snippet queries executed against the same corpus that status reports as indexed?
• Is workspace scoping or source-key registration excluding these local documents from visibility?

Client machine dependency note

• On the client machine used for ingestion, OCR dependencies were missing: tesseract missing and pdftoppm missing.</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes C64CityBuilder GraphRAG local-text visibility diagnosis #28

---

🔒 GitHub Advanced Security automatically protects Copilot coding agent pull requests. You can protect all pull requests by enabling Advanced Security for your repositories. Learn more about Advanced Security.

[sharpninja]

@copilot Resolve pipeline errors

[Copilot]

@copilot Resolve pipeline errors

Resolved the current PR pipeline blockers in baeb997. The config validation script now auto-detects the YAML appsettings file used in this repo, and the USER-GUIDE markdownlint failure is fixed.

Warning

Firewall rules blocked me from connecting to one or more addresses (expand for details)

I tried to connect to the following addresses, but was blocked by firewall rules:

• example.com
  • Triggering command: /home/REDACTED/work/McpServer/McpServer/tests/McpServer.Support.Mcp.Tests/bin/Release/net9.0/McpServer.Support.Mcp.Tests /home/REDACTED/work/McpServer/McpServer/tests/McpServer.Support.Mcp.Tests/bin/Release/net9.0/McpServer.Support.Mcp.Tests @@ /tmp/tmpDL7XJC.tmp (dns block)

If you need me to access, download, or install something from one of these locations, you can either:

• Configure Actions setup steps to set up my environment, which run before the firewall is enabled
• Add the appropriate URLs or hosts to the custom allowlist in this repository's Copilot coding agent settings (admins only)