--- name: titan-research description: > Deep research protocol for comprehensive, multi-phase investigations. Use when: user invokes /titan, "deep research", "deep dive", "systematic research", "research protocol", "TITAN mode", "comprehensive research", "investigate this thoroughly", "I need a thorough analysis", "research this topic deeply". Enforces structured intake, multi-source synthesis, and rigorous output formatting. Produces decision-ready intelligence, not reference documents. user-invocable: true argument-hint: "[research topic or question]" --- # TITAN Research Protocol v2.0 Intelligence amplification system. Transforms research from "Ask > Search > Summarize" into recursive, cumulative knowledge building. Optimizes for Return on Attention: maximum insight density, minimum noise. **Behavioral constraints (always active):** - TONE: Objective, high-bandwidth. "McKinsey Partner" meets "Chief Engineer." - NO FLUFF: Every sentence must reduce uncertainty. - NO LAZINESS: Do not suggest "further research" unless the web is literally empty. - CITATION: [Source Name] for every claim. ## Platform Dispatch Detect execution environment and adapt: **Claude Code** (Agent tool available): - Use subagent orchestration for parallel research - Spawn research agents per pillar (up to 6 parallel agents) - Each agent: web search per assigned topics, structured findings - Orchestrator: synthesize agent outputs into final report - Save all artifacts to filesystem with naming convention - Reports go to `resources/research/reports/` - Briefs go to `resources/research/briefs/` - **Multi-phase orchestration is Claude-Code-native.** When the brief structures research as 2+ dependent phases (Phase N+1 pillar scopes depend on Phase N findings), Claude Code is the right platform. See `references/multi-phase-protocol.md`. Claude.ai extended search is single-phase by design. **Claude.ai** (extended search available): - Use extended search tool for deep pillar research - Sequential pillar execution within single context - Each pillar: 100+ links via extended search - Produce artifact with naming convention - Store as project file or conversation artifact ## Phase 0: Intake Gatekeeper (Mandatory) Before researching, validate the request. **Reject if:** - Too vague ("research AI" without scope) - Better answered with a quick search - No clear decision or action the research will inform **If rejected:** Ask for 3 targeted Constraint Questions to narrow scope. **If accepted:** 1. Create Thinking File IMMEDIATELY (see `references/thinking-file.md`) - `_THINKING_topic_v1_YYYY-MM-DD.md` in `resources/research/briefs/` - Log first impressions, initial hypotheses, blind spots - This is WRITE-ONLY; never loaded back into context 2. Enter Brief Building Mode ### Brief Building Mode Load `references/input-dor.md` for the full facilitation protocol. Two intake paths: **Path A — Dialogue** (default): Interactive facilitation through targeted questions. **Path B — Ingest** (when user provides raw inputs): Extract from transcripts, notes, or documents first, then ask only about gaps. See "Ingest Mode" in input-dor.md. Core flow (both paths): 1. **Create brief file**: `BRIEF_topic_DRAFT_v1_YYYY-MM-DD.md` (See `references/brief-scaffolding.md` for template and protocol) 2. **Build brief content**: Via dialogue (Path A) or extraction + gap-fill (Path B) - Core question (specific, answerable) - Success criteria (what decision/action this informs) - Scope (IN/OUT boundaries) - Constraints (time, budget, capability) - Hypothesis (if any) - Anti-patterns (what to avoid) - Evidence bar (Standard / High / Paranoid) 3. **Validate**: Run quality gates on completed brief 4. **Generate Research Execution Plan**: Mandatory. Decompose the brief into explicit research pillars with structured allocation. This bridges the brief to execution. See `references/agent-allocation.md` for pillar object schema. ``` ## Research Execution Plan ### Pillar Allocation | Pillar | Name | Priority | Weight | Depth | Topics | Source Strategy | Validation | |--------|------|----------|--------|-------|--------|-----------------|------------| | P1 | [name] | high | 30% | deep | [topics] | [strategy] | high | | P2 | [name] | high | 25% | balanced | [topics] | [strategy] | standard | | ... ### Emergence Allowance [X]% reserved (based on topic novelty: well-understood=10%, moderate=25%, novel=50%) ### Execution Notes - Platform recommendation (Claude Code subagents vs Claude.ai extended search) - Sequencing if running sequentially (priority order) - Batching guidance (max 5-6 pillars before timeout risk on Claude.ai) - Key dependencies between pillars ``` 5. **Confirm**: Present full brief + execution plan, get user approval to launch **Express mode**: If user says "skip brief" or "express mode", capture core question only. Still generate a minimal 3-pillar execution plan before launching. ## Phase 0.5: Socratic Stress Test Before launching research, challenge the brief: - Are we solving the right problem? - What assumption, if wrong, would invalidate this entire inquiry? - What domains/perspectives are we not considering? - Is the scope too broad to produce actionable results? This is a brief quality check, NOT a completeness guarantee. Emergence will surface what we couldn't anticipate. See `references/emergence-protocol.md`. ## Phase 1: Landscape Scan (The Architect) Before searching, map the territory: 1. **Deconstruct**: Break request into First Principles (economics, psychology, physics) 2. **MECE Tree**: Create research outline that is Mutually Exclusive, Collectively Exhaustive 3. **Gap Analysis**: Identify Known Unknowns and Unknown Unknowns 4. **Pillar Allocation**: Map research pillars to agents/search streams (See `references/agent-allocation.md` for structured allocation protocol) Key players, frameworks, schools of thought. 3-5 primary tension points. Flag areas where conventional wisdom may be wrong. ## Phase 2: Deep Investigation (The Miner) Execute research with Adversarial Filtering: - **Triangulation**: A fact is only true if confirmed by 3 distinct source types (e.g., Academic Paper + Industry Report + User Forum) - **Divergence Hunting**: Specifically search for critiques, failures, competitors - **Provenance**: Prioritize primary data (10-K, API Docs, GitHub Issues) over secondary interpretations (Blogs, News) - **Emergence awareness**: Flag discoveries that don't fit the brief structure. These may be the most important findings. See `references/emergence-protocol.md`. Track confidence levels on each finding. ## Phase 3: Synthesis (The Critic) Run a silent Red Team simulation before finalizing: - **Attack**: "If this report is wrong, why?" - **Refine**: Patch weak points. Flag structural data gaps as Critical Blind Spots. - **Pattern detection**: Identify convergent signals across sources - **Contradiction resolution**: Explain why contradictions exist, not just that they do Distinguish "known" from "likely" from "speculative". ## Phase 4: Output (Chain of Density) Structure all research output using the standard report schema. See `references/report-standardization.md` for full protocol. ### Report Structure ``` # [Topic]: Strategic Deep Dive ## 1. Executive BLUF (2 sentences max) The Alpha: single most high-leverage insight. Confidence Score: (0-100%) & Primary Risk Factor. ## 2. Decision Matrix | Option | Evidence Strength | Risk | Effort vs Impact | The "Catch" | ## 3. Key Findings (Executive Summary) Bullet points. No fluff. What matters. ## 4. [Content Sections - adapted to topic] Full analysis with numbered citations [1][2][3]. ## 5. Blind Spots & Conflicts Consensus vs Reality. The "Undiscussables." ## 6. Compounding Heuristics 1-3 "Golden Rules" from this research domain. ## 7. Research Success Report & Path Navigation - BLUF assessment vs original brief - Remaining questions & hypotheses - Emergent research targets (new discoveries) - Suggested next actions (prioritized by leverage) ## Sources & Citations [1] Source Title (URL) [2] Source Title (URL) ``` ### Report Filing **File naming**: `RESEARCH_REPORT_topic-kebab-case_v1_YYYY-MM-DD.md` **Location**: `resources/research/reports/` (canonical home) **YAML front matter** (required): ```yaml --- type: RESEARCH_REPORT status: Active topic: topic-kebab-case topic_area: [spirituality | business | technology | operations | career | finance | health] destination: [path to wiki-link stub] date: YYYY-MM-DD version: 1.0 associated_brief: BRIEF_topic_v1_YYYY-MM-DD.md tags: [tag1, tag2] source: titan-research --- ``` **Post-report**: Update associated brief status to Research-Complete. See `references/report-standardization.md` Phase 4. ### Wikilink Discipline When writing markdown files that reference OTHER markdown files inside an Obsidian vault, use Obsidian wikilink form `[[basename]]` or `[[path/to/basename]]`, NOT `[text](path.md)`. Obsidian vaults with `alwaysUpdateLinks: true` keep wikilinks intact across file renames; markdown-style links break. This applies only when both the writing file and its target live inside the same Obsidian vault. Exceptions: external URLs (`https://...`), targets outside the vault, and explicitly-broken targets (a broken `[[wikilink]]` is invisible failure; keep the markdown form so the broken intent is visible). Applies at write-time, not post-write. Subagent prompts must include this discipline (see `references/agent-allocation.md` Platform-Specific Mapping → Claude Code → Wikilink Discipline). ## Phase 5: Meta-Check Before delivering: - Did I actually answer the question asked? - Am I presenting the most effective answer or the most obvious one? - Where would this analysis fail? - What assumption, if wrong, would invalidate the conclusion? - Did emergence surface anything that should restructure the report? ## Post-Research: Sequential Chaining When a report identifies remaining questions or emergent targets, and the user wants to continue researching: 1. Load `references/sequential-chaining.md` 2. Harvest gaps/targets from Report N's Research Success Report 3. Auto-generate Brief N+1 with inherited context and narrowed scope 4. Execute Run N+1 with standard Titan phases Chain depth > 3 requires explicit user confirmation (diminishing returns likely). ## Module Reference Load these on-demand when the corresponding phase activates: | Module | Load When | Reference File | |--------|-----------|----------------| | Thinking File | Phase 0 (FIRST action, before brief) | `references/thinking-file.md` | | Input DOR | Phase 0 (brief building) | `references/input-dor.md` | | Brief Scaffolding | Phase 0 (file creation) | `references/brief-scaffolding.md` | | Agent Allocation | Phase 0 (pillar decomposition) + Phase 1 | `references/agent-allocation.md` | | Emergence Protocol | Phase 0.5, 2, 4 (discovery) | `references/emergence-protocol.md` | | Report Standardization | Phase 4 (output formatting) | `references/report-standardization.md` | | Sequential Chaining | Post-research (multi-pass) | `references/sequential-chaining.md` | | Meta-Analysis | Post-research (multi-report synthesis) | `references/meta-analysis.md` | | Discussion Documents | Post-report (extended Q&A) | `references/discussion-documents.md` | | Definitions & Concepts | Any phase (terminology) | `references/definitions-and-concepts.md` | ## Usage Notes - This protocol is for genuine deep dives, not quick questions - Output should be "ready for implementation", not "here's what I found" - Every research run compounds previous knowledge - Use `assets/research-brief-template.md` for blank brief skeleton - Use `assets/report-output-template.md` for blank report skeleton