Phase 1a: Publish GitHub MCP case study at /research/github-mcp-case-study/

[mickdarling]

Part of #32 (Epic: MCPAQL 2026 product launch). Phase 1a deliverable.

Depends on:

• feat(case-studies): publish GitHub MCP measured comparison #31 (existing case-study draft) being merged
• The LLM-correctness benchmark numbers being available (tracked in MCPAQL/tools)

What

Publish the GitHub MCP measured-comparison case study at mcpaql.org/research/github-mcp-case-study/. Combines:

• The existing case-study draft from PR feat(case-studies): publish GitHub MCP measured comparison #31 (functional parity, 89 operations mapped, token-reduction numbers)
• The new LLM-correctness benchmark numbers (first-call success, turns, tokens-to-completion, recovery) from the parity-runner extension
• Methodology documentation
• Reproducibility notes (link to harness, harness invocation, environment)

Why

This is the first piece of measured-evidence marketing for MCPAQL. Two numbers live or die here:

1. Token reduction (96% — already attested in ORG-README.md)
2. LLM correctness delta (new, from this case study)

Once published, the new .com homepage (Phase 1b) cites this URL as the proof page. Without the case study live, the homepage is making claims it can't substantiate.

How

• Merge feat(case-studies): publish GitHub MCP measured comparison #31 (case-study draft) to develop
• Layer in the LLM-correctness numbers once the benchmark run completes
• Verify the page renders at /research/github-mcp-case-study/ on the .org build
• Add cross-references from /spec/* pages and the future homepage placeholder
• CI green: Lighthouse, HTMLHint, markdownlint, lychee

Phase

1a (this week). Blocking for the Phase 1b homepage launch.