Tech Eval Orchestrator

Orchestrates the full 3-phase technology evaluation pipeline: Research (9 artifacts), Visualize (4-6 JSX dashboards), Deepen (15-25 ideation prompts).

Core Capabilities

1. Phase 1 — Research Pipeline

Coordinates parallel research agents to produce 9 markdown artifacts:

Artifact	Skill	Model	Parallel Group
1-2-3-quick-start.md	quick-start-guide-generator	Haiku	A
executive-summary.md	executive-summary-generator	Opus	B
coditect-impact.md	strategy-brief-generation	Opus	B
sdd.md	sdd-generator	Opus	C
tdd.md	tdd-generator	Opus	C
adrs/*.md	adr-decision-workflow	Sonnet	D
glossary.md	glossary-generator	Haiku	A
mermaid-diagrams.md	mermaid-architecture-generator	Sonnet	D
c4-architecture.md	c4-architecture-generator	Sonnet	D

Execution: Groups A-D execute in dependency order. Within each group, artifacts generate in parallel.

Quality Gate 1 (post-research): moe-judges — source diversity, recency, relevance (threshold: 85%)

2. Phase 2 — Visualization Pipeline

Generates React JSX dashboards from Phase 1 artifacts:

Dashboard	Skill	Model
tech-architecture-analyzer.jsx	jsx-dashboard-generator	Sonnet
strategic-fit-dashboard.jsx	jsx-dashboard-generator	Sonnet
coditect-integration-playbook.jsx	jsx-dashboard-generator	Opus
executive-decision-brief.jsx	jsx-dashboard-generator	Opus
competitive-comparison.jsx (ext)	jsx-dashboard-generator	Sonnet
implementation-planner.jsx (ext)	jsx-dashboard-generator	Sonnet

Quality Gate 2 (post-visualization): Valid JSX, Tailwind-only, WCAG AA contrast, no empty tabs.

3. Phase 3 — Ideation Pipeline

Generates categorized follow-up prompts:

Output	Skill	Model
ideation-prompts.md (15-25 prompts)	research-ideation-generator	Sonnet

Quality Gate 3 (post-ideation): All 6 categories present, prompts self-contained, no overlap.

4. Compliance Integration

Cross-cutting compliance analysis runs parallel to Phase 1:

Output	Skill	Model
compliance-analysis.md	compliance-surface-analyzer	Opus

5. Bundling

All artifacts bundled into distributable format:

Output	Skill	Model
bundle.zip (or pdf/docx)	artifact-bundler	Haiku

Orchestration Flow

INPUT: topic + options
  |
  v
[Phase 1: Research] ──parallel──> 9 artifacts
  |                                    |
  |                              [Compliance Analysis] (parallel)
  v
[Quality Gate 1: moe-judges] ── fail? retry with expanded search
  |
  v
[Cross-Reference: artifact-cross-linker]
  |
  v
[Quality Gate 2: council-review] ── fail? human review
  |
  v
[Phase 2: Visualize] ──parallel──> 4-6 JSX dashboards
  |
  v
[Phase 3: Deepen] ──────────────> 15-25 prompts
  |
  v
[Bundle: artifact-bundler] ─────> output bundle
  |
  v
OUTPUT: complete evaluation package

Token Budget Management

Phase	Budget	Model Mix
Research (Phase 1)	80,000	40% Opus, 40% Sonnet, 20% Haiku
Visualization (Phase 2)	40,000	30% Opus, 70% Sonnet
Ideation (Phase 3)	10,000	100% Sonnet
Quality Gates	15,000	50% Opus, 50% Sonnet
Bundling	5,000	100% Haiku
Total	150,000

Budget warnings at 80%, hard stop at 95%.

Command Interface

# Full 3-phase pipeline
/r2a "<topic>" --full

# Phase 1 only (default)
/r2a "<topic>" --artifacts all

# Phase 1 + 2
/r2a "<topic>" --artifacts all --visualize

# Phase 1 + 2 + 3
/r2a "<topic>" --full

# Phase 2 from existing artifacts
/r2a:visualize ./artifacts/

# Phase 3 from existing artifacts
/r2a:deepen ./artifacts/

# Seed folder with system prompt
/r2a "<topic>" --seed ./research/

# Extended visualization (6 dashboards)
/r2a "<topic>" --full --extended

Input Requirements

{
  "topic": "string (required)",
  "artifacts": "all | list of artifact names",
  "mode": "standard | competitive | technical",
  "phases": [1, 2, 3],
  "extended": false,
  "output_format": "zip | pdf | markdown | docx",
  "output_path": "./artifacts/",
  "seed": false,
  "review_after": [],
  "date_range": "2025-2026"
}

Error Handling

Error	Action
Research phase produces < 3 artifacts	Retry with expanded search scope
Quality gate fails twice	Escalate to human checkpoint
Token budget exceeded at 95%	Complete current artifact, skip remaining
Web search returns no results	Fall back to codebase-only analysis
JSX validation fails	Re-generate with stricter constraints
Mermaid syntax error	Auto-fix via mermaid-diagram-fixing skill

Skills Coordinated (18+)

Core Skills

• quick-start-guide-generator, executive-summary-generator, sdd-generator, tdd-generator
• glossary-generator, mermaid-architecture-generator, adr-decision-workflow
• c4-architecture-generator, jsx-dashboard-generator, research-ideation-generator
• compliance-surface-analyzer, artifact-bundler

Supporting Skills

• llm-research-patterns, competitive-analysis, strategy-brief-generation
• source-attribution-tracker, artifact-cross-linker, mermaid-diagram-fixing

Quality Gate Skills

• moe-judges, council-review, uncertainty-quantification

Related Components

Component	Relationship
strategy-brief-generator	Predecessor — handles strategy briefs
research-agent	Worker — executes research tasks
senior-architect	Worker — architecture decisions
competitive-analyst	Worker — competitive analysis
web-search-researcher	Worker — web research
/r2a command	Primary invocation path

---

Plugin: research-to-artifacts (R2A v2.0) ADR: ADR-157 Track: H (Framework)