Create evaluation matrix mapping approaches to requirements

[olaservo]

Background

Approaches documents six approaches with community input, and experimental findings captures implementation results. But there is no structured way to compare approaches against a shared set of requirements. This makes it difficult to identify where approaches genuinely diverge, where they overlap, and which gaps remain.

This issue depends on having a requirements document (#46) but can be started in parallel using draft requirements.

Scope

Create a section in docs/approaches.md (or a standalone docs/evaluation-matrix.md) containing:

1. A matrix table with:

   • Rows: Requirements (from docs/requirements.md)
   • Columns: Approaches 1–6
   • Cells: ✅ (satisfies), ⚠️ (partially / with caveats), ❌ (does not satisfy), ❓ (untested)

2. Notes per cell explaining the rating, especially for ⚠️ and ❌ — what's missing or what caveat applies.

3. A summary section identifying:

   • Requirements satisfied by all approaches (consensus areas — these can be convention-ized now)
   • Requirements where approaches diverge (these are the real decision points)
   • Requirements no approach currently satisfies (gaps that need new work)

4. Mapping to experimental evidence — where findings from Compare skill delivery mechanisms: file-based vs MCP-based #37, Survey client resource-loading support across major MCP clients #38, Build client-side reference implementation for model-driven resource loading #40, Build server-side reference implementation for skills-as-resources #41, Test skills-via-sampling approach (Approach 3 variant) #42 provide data, reference them.

Acceptance Criteria

• Matrix covering at least 10 requirements × 6 approaches
• Each ⚠️ or ❌ cell has a brief rationale
• Summary section identifies consensus areas, divergence points, and gaps
• Reviewed by at least 2 IG facilitators
• Referenced from the convention draft (Draft MCP Skills Convention v0.1 #43) as decision support

References

• Approaches
• Requirements document: Extract testable requirements from use cases and open questions #46
• Related research issues: Compare skill delivery mechanisms: file-based vs MCP-based #37, Survey client resource-loading support across major MCP clients #38, Build client-side reference implementation for model-driven resource loading #40, Build server-side reference implementation for skills-as-resources #41, Test skills-via-sampling approach (Approach 3 variant) #42