version: "1.0.0"
schema: "reasonkit-proofguard-deep-research-v1"
created: "2025-12-11"
last_updated: "2025-12-11"
license: "Apache-2.0"
metadata:
id: "PROT-PG-DEEP-001"
name: "ProofGuard Deep Research Protocol"
shortcode: "pg-deep"
thinktool: "ProofGuard"
priority: 1
enforcement: "on_demand"
triggers:
explicit:
- "deep research"
- "ProofGuard analysis"
- "verify claims"
- "triangulate sources"
implicit:
- "uncertainty > 0.5 on technical claim"
- "high-stakes decision pending"
- "multiple conflicting sources"
description: |
A rigorous research protocol that combines multi-source triangulation,
structured synthesis, and explicit confidence intervals to produce
high-quality, auditable technical assessments.
phase_1_context_loading:
name: "Context Loading"
objective: "Establish baseline understanding before research"
duration_estimate: "2-5 minutes"
steps:
- id: "1.1"
action: "Search episodic memory for prior work on topic"
tool: "mcp__plugin_episodic-memory_episodic-memory__search"
fallback: "Proceed without if unavailable"
- id: "1.2"
action: "Read relevant project documentation"
tools: ["Read", "Glob"]
targets:
- "CLAUDE.md / ORCHESTRATOR.md"
- "ARCHITECTURE.md"
- "Existing implementation files"
- id: "1.3"
action: "Identify existing decisions and constraints"
output: "Context summary with known constraints"
checklist:
- "[ ] Prior conversations searched"
- "[ ] Project docs reviewed"
- "[ ] Existing code examined"
- "[ ] Constraints identified"
phase_2_parallel_search:
name: "Parallel Search Initiation"
objective: "Cast wide net across multiple source types"
duration_estimate: "3-8 minutes"
search_strategy:
minimum_queries: 3
parallel_execution: true
query_diversity_required: true
query_templates:
technical_specs: "{topic} technical specifications benchmarks {year}"
academic_papers: "{topic} research paper arxiv {year}"
comparisons: "{topic_a} vs {topic_b} benchmark comparison {year}"
implementations: "{topic} implementation {language} integration"
source_tiers:
tier_1_authoritative:
description: "Official docs, GitHub repos, peer-reviewed papers"
examples:
- "HuggingFace model cards"
- "arXiv papers"
- "Official documentation"
- "Nature/Science publications"
weight: 1.0
tier_2_secondary:
description: "Reputable tech blogs, industry analysis"
examples:
- "NVIDIA technical docs"
- "Major framework docs (LangChain, LlamaIndex)"
- "Research lab blogs"
weight: 0.8
tier_3_independent:
description: "Independent validation, community implementations"
examples:
- "Academic course projects"
- "Independent benchmarks"
- "Community tutorials"
weight: 0.6
checklist:
- "[ ] Minimum 3 parallel searches executed"
- "[ ] Multiple source tiers targeted"
- "[ ] Current year included in queries"
phase_3_source_verification:
name: "Source Verification (ProofGuard Core)"
objective: "Triangulate every major claim with 3+ sources"
duration_estimate: "5-15 minutes"
triangulation_rule:
mandatory: true
minimum_sources: 3
requirement: |
EVERY factual claim MUST be verified by:
- Source A (Primary): Authoritative/official source
- Source B (Secondary): Different domain, same conclusion
- Source C (Independent): Different author, same timeframe
verification_levels:
"VERIFIED":
symbol: "✓"
criteria: "3+ sources agree, all URLs accessed"
confidence_boost: "+15%"
"LIKELY":
symbol: "~"
criteria: "2 sources agree, 1 indirect"
confidence_boost: "+5%"
"UNVERIFIED":
symbol: "?"
criteria: "Single source or conflicting info"
confidence_boost: "0%"
action: "MUST disclose in output"
"CONTRADICTED":
symbol: "✗"
criteria: "Sources disagree"
confidence_boost: "-20%"
action: "MUST investigate further"
output_format:
triangulation_table:
required: true
template: |
| Finding | Source A (Primary) | Source B (Secondary) | Source C (Independent) | Consensus |
|---------|-------------------|---------------------|----------------------|-----------|
| [Claim] | [URL/Citation] | [URL/Citation] | [URL/Citation] | [Status] |
anti_patterns:
- "Claiming verification without 3 sources"
- "Using same domain for all 3 sources"
- "Not actually accessing URLs"
- "Ignoring contradicting evidence"
checklist:
- "[ ] Each major claim has 3+ sources"
- "[ ] Sources are from different domains/authors"
- "[ ] All URLs were actually accessed"
- "[ ] Contradictions investigated"
- "[ ] Triangulation table created"
phase_4_structured_synthesis:
name: "Structured Synthesis"
objective: "Transform verified data into structured, actionable output"
duration_estimate: "5-10 minutes"
required_output_elements:
technical_specifications_table:
required: true
purpose: "Consolidate verified specs"
template: |
| Property | Value | Source |
|----------|-------|--------|
| [Spec] | [Val] | [Cite] |
benchmark_comparison_table:
required: "when benchmarks exist"
purpose: "Quantitative performance comparison"
template: |
| Benchmark | Subject Score | Baseline | Improvement |
|-----------|---------------|----------|-------------|
| [Name] | [Score] | [Base] | [Delta] |
fit_analysis_matrix:
required: "when evaluating adoption"
purpose: "Structured decision support"
template: |
| Criterion | Score | Rationale |
|-----------|-------|-----------|
| [Factor] | [✅/⚠️/❌] | [Why] |
architecture_diagram:
required: "when systems involved"
purpose: "Visual spatial reasoning"
format: "ASCII art within code blocks"
implementation_roadmap:
required: "when action items exist"
purpose: "Actionable next steps"
template: |
PRIORITY 1 (Immediate):
├── [ ] Task 1
└── [ ] Task 2
PRIORITY 2 (Next):
└── [ ] Task 3
confidence_requirements:
every_recommendation: true
format: "Confidence: XX%"
calibration_guidance: |
95%+ = Multiple sources agree, tested personally
85-94% = Strong triangulation, no contradictions
70-84% = Good evidence, some gaps
50-69% = Partial evidence, significant uncertainty
<50% = Should not recommend without disclaimer
checklist:
- "[ ] Technical specs table created"
- "[ ] Benchmarks compared (if applicable)"
- "[ ] Fit analysis matrix built"
- "[ ] Architecture diagram included"
- "[ ] Implementation roadmap provided"
- "[ ] All recommendations have confidence %"
phase_5_output:
name: "Output with Explicit Confidence"
objective: "Deliver auditable, actionable report"
duration_estimate: "3-5 minutes"
required_sections:
proofguard_verification_summary:
position: "top"
required: true
purpose: "Immediate audit trail"
detailed_analysis:
position: "middle"
required: true
subsections:
- "Technical Specifications"
- "Benchmark Results"
- "Integration Assessment"
- "Production Considerations"
decision_matrix:
position: "before conclusion"
required: true
purpose: "At-a-glance decision support"
template: |
| Component | Project A | Project B | Confidence |
|-----------|-----------|-----------|------------|
| [Item] | [Y/N/Defer] | [Y/N/Defer] | [%] |
executive_summary:
position: "end"
required: true
purpose: "Quick reference for decisions"
max_length: "5-7 rows in table format"
sources_section:
position: "end"
required: true
tiers:
- "Primary Sources (Tier 1)"
- "Secondary Sources (Tier 2)"
- "Independent Validation (Tier 3)"
quality_gates:
- gate: "All claims triangulated"
required: true
- gate: "All recommendations have confidence"
required: true
- gate: "All sources hyperlinked"
required: true
- gate: "Executive summary present"
required: true
- gate: "Actionable roadmap included"
required: true
objective_measures:
completeness_metrics:
- metric: "Triangulation Coverage"
formula: "(claims_with_3_sources / total_claims) * 100"
target: ">= 90%"
- metric: "Source Diversity"
formula: "unique_domains / total_sources"
target: ">= 0.7"
- metric: "Confidence Explicitness"
formula: "(recommendations_with_confidence / total_recommendations) * 100"
target: "100%"
quality_metrics:
- metric: "Actionability Score"
criteria: "Does output include implementation roadmap with priorities?"
scoring: "0-10 scale"
target: ">= 8"
- metric: "Auditability Score"
criteria: "Can every claim be traced to source?"
scoring: "0-10 scale"
target: ">= 9"
- metric: "Decision Support Score"
criteria: "Does output enable yes/no decision?"
scoring: "0-10 scale"
target: ">= 8"
efficiency_metrics:
- metric: "Research Depth vs Time"
formula: "sources_consulted / minutes_elapsed"
benchmark: ">= 2 sources/minute"
- metric: "Output Density"
formula: "structured_tables / total_output_sections"
target: ">= 0.5"
thinktool_integration:
activation:
shortcut: "pg-deep"
full_command: "--profile paranoid --thinktool ProofGuard"
automatic_triggers:
- "Research request with technical claims"
- "Evaluation of new technology"
- "High-stakes architectural decision"
profile_mapping:
paranoid:
modules:
["GigaThink", "LaserLogic", "BedRock", "ProofGuard", "BrutalHonesty"]
confidence_target: 95%
this_protocol: "REQUIRED"
scientific:
modules: ["GigaThink", "LaserLogic", "BedRock", "ProofGuard"]
confidence_target: 85%
this_protocol: "RECOMMENDED"
deep:
modules:
["GigaThink", "LaserLogic", "BedRock", "ProofGuard", "BrutalHonesty"]
confidence_target: 85%
this_protocol: "RECOMMENDED"
output_template_injection:
when: "ProofGuard Deep Research activated"
inject:
- "Triangulation table at top"
- "Decision matrix before conclusion"
- "Executive summary at end"
- "Sources section with tiers"
example_output_structure: |
# DEEP RESEARCH REPORT: [Topic]
## ProofGuard Verification Summary
| Finding | Source A (Primary) | Source B (Secondary) | Source C (Independent) | Consensus |
|---------|-------------------|---------------------|----------------------|-----------|
| [Claim] | [URL] | [URL] | [URL] | VERIFIED ✓|
---
## 1. [TOPIC A] ANALYSIS
### Technical Specifications (VERIFIED)
| Property | Value | Source |
|----------|-------|--------|
### Benchmark Performance (VERIFIED)
| Benchmark | Score | Baseline | Improvement |
|-----------|-------|----------|-------------|
---
## 2. [TOPIC B] ANALYSIS
[Same structure...]
---
## 3. FIT ANALYSIS FOR [PROJECT]
### [Project Name] Assessment
| Criterion | Score | Rationale |
|-----------|-------|-----------|
**RECOMMENDATION:** [HIGH/MEDIUM/LOW PRIORITY]
```
Confidence: XX%
Implementation Path:
1. Step 1
2. Step 2
Estimated Effort: [Low/Medium/High]
Risk: [Low/Medium/High]
```
---
## 4. SYNTHESIS & RECOMMENDATIONS
### Decision Matrix
| Component | Project A | Project B | Confidence |
|-----------|-----------|-----------|------------|
### Implementation Priority
```
PRIORITY 1 (Immediate):
├── [ ] Task 1
└── [ ] Task 2
```
---
## 5. SOURCES
### Primary Sources (Tier 1)
- [URL 1]
- [URL 2]
### Secondary Sources (Tier 2)
- [URL 3]
### Independent Validation (Tier 3)
- [URL 4]
---
## EXECUTIVE SUMMARY
| Topic | Recommendation | Project Fit | Confidence |
|-------|----------------|-------------|------------|
changelog:
- version: "1.0.0"
date: "2025-12-11"
changes:
- "Initial protocol derived from BGE-M3/RAPTOR deep research session"
- "Defined 5-phase workflow"
- "Established triangulation requirements"
- "Created structured output templates"
- "Added objective measures for testing"
- "Integrated with ThinkTool profiles"