Comparative Matrix

Synthesizes analysis outputs into structured decision frameworks.

Process

1. Collect analysis outputs from multiple frameworks
2. Normalize findings to comparable dimensions
3. Generate comparison matrix
4. Apply decision heuristics
5. Document recommendations with rationale

Comparison Dimensions

Core Dimensions (Always Include)

| Dimension | What to Compare | Decision Criteria | |-----------|-----------------|-------------------| | Typing | Strict (Pydantic) vs Loose (dicts) | Team preference, runtime safety needs | | Async | Native async vs sync-with-wrappers | Scalability requirements | | State | Immutable vs mutable | Concurrency safety, debugging | | Config | Code-first vs config-first | Flexibility vs discoverability | | Extensibility | Composition vs inheritance | Maintainability, learning curve |

Domain-Specific Dimensions

| Dimension | When to Include | |-----------|-----------------| | Reasoning Pattern | Comparing agent frameworks | | Memory Strategy | Long-running agents | | Multi-Agent | Orchestration systems | | Observability | Production deployments | | Tool Interface | Custom tool development |

Matrix Template

## Best-of-Breed Matrix: [Analysis Title]

| Dimension | Framework A | Framework B | Framework C | **Recommendation** |
|:----------|:------------|:------------|:------------|:-------------------|
| **Typing** | Pydantic V1, deep nesting | TypedDict, flat | Loose dicts | *Pydantic V2, flat structures* |
| **Async** | Sync core, async wrapper | Native async | Mixed | *Native async required* |
| **State** | Mutable, in-place | Immutable copy | Hybrid | *Immutable preferred* |
| **Config** | YAML + Python | Pure Python | JSON | *Python for type safety* |
| **Extensibility** | Deep inheritance (6 layers) | Composition | Protocols | *Composition + Protocols* |

### Dimension Details

#### Typing
- **Framework A**: Uses Pydantic V1 with deeply nested models (Message → Content → Block → ...)
  - Pro: Full validation at boundaries
  - Con: Difficult to extend, version migration pain
- **Framework B**: TypedDict with flat structure
  - Pro: Simple, fast, IDE support
  - Con: No runtime validation
- **Recommendation**: Adopt Pydantic V2 with intentionally flat structures. Use TypedDict for internal types.

[Continue for each dimension...]

Decision Heuristics

Apply these heuristics when recommendations aren't obvious:

Scalability-First

IF high_concurrency_expected:
    PREFER native_async
    PREFER immutable_state
    PREFER stateless_tools

DX-First (Developer Experience)

IF team_is_small OR rapid_iteration:
    PREFER simple_inheritance_over_protocols
    PREFER code_first_config
    PREFER explicit_over_magic

Production-First

IF mission_critical:
    PREFER strict_typing
    PREFER comprehensive_observability
    PREFER explicit_error_boundaries

Output Artifacts

1. Summary Matrix - Single-page comparison table
2. Detailed Analysis - Per-dimension breakdown with evidence
3. Recommendation Document - Actionable decisions with rationale
4. Trade-off Log - Documented compromises and their justification

Example Output Structure

comparative-analysis/
├── matrix.md              # Summary comparison table
├── dimensions/
│   ├── typing.md          # Detailed typing analysis
│   ├── async.md           # Concurrency model analysis
│   └── ...
├── recommendations.md     # Final decisions
└── tradeoffs.md          # Documented compromises

Integration

• Inputs from: All Phase 1 & 2 analysis skills
• Outputs to: antipattern-catalog, architecture-synthesis