--- type: reference skill: titan-research module: thinking-file phase: "Phase 0 (creation) through Phase 4 (completion)" date: 2026-04-08 source: SYSTEM_MOD_thinking-file_v0.1_2025-12-19.md --- # Thinking File: Claude's Reasoning Capture Protocol External reasoning workspace that keeps the chat stream clean while preserving Claude's meta-cognitive trail for audit and learning. ## When to Create **Activate immediately when:** - User requests research (any trigger phrase) - Phase 0 begins - This is the FIRST action, BEFORE brief creation **Do NOT activate when:** - Simple Q&A that doesn't need research - Express mode with explicit "skip scaffolding" - Continuing research in same session (append to existing file) ## File Naming ``` _THINKING_topic-kebab-case_v1_YYYY-MM-DD.md ``` - Prepend underscore sorts to top of directory - Topic matches the corresponding BRIEF file - v1 for first run; v2 for re-run from scratch; append for continuation ## File Location Same directory as the brief: `resources/research/briefs/` ## WRITE-ONLY Rule **Thinking files are never loaded back into context.** - Claude writes to the thinking file but does not read it during the same session - It is an artifact for Brandon (or future Claude) to review - This prevents token explosion and recursive loading problems - Enables transparent reasoning without performance cost ## Template ```markdown --- type: THINKING_PROCESS status: Active version: 1.0 date: YYYY-MM-DD associated_brief: BRIEF_topic_v1_date.md associated_report: [filled after report created] tags: [thinking-process, topic-keywords] --- # Thinking Process Log: [Topic] **Research Start:** [timestamp] **Research End:** [filled at completion] **Total Checkpoints:** [increments] --- ## Pre-Brief: First Impressions **User's Request:** [exact phrase] **What I think user actually needs:** - [Interpretation of real underlying need] **Possible blind spots in their framing:** - [Blind spot I suspect] **Initial Hypotheses:** - H1: [working theory] — If true: [implication]. If false: [alternative]. **Research Strategy Gut Check:** - Phase 1 approach (likely): [how I'll map problem space] - Biggest risk: [what I'm most likely to get wrong] --- ## Checkpoint 1: Phase 1 Architect **Key Insights:** 1. [Insight] — Confidence: [high/medium/low because X] **Assumptions I'm Making:** - [Assumption] — If wrong, this breaks: [consequence] **Blind Spots Check:** - What am I systematically missing? - What would a domain expert flag? --- ## Checkpoint 2: Phase 2 Miner **Triangulation Status:** - [Claim]: triangulated / single source / conflicting **Surprises & Pivots:** - Expected X, found Y **Updated Hypotheses:** - Original: [from brief]. Current: [validated/refuted/revised] --- ## Checkpoint 3: Phase 3 Critic **Pre-Mortem Red Team:** - If this research is wrong, why? - Data quality risk: [assessment] - Selection bias risk: [assessment] **Confidence Calibration:** - Where am I overconfident? - Where am I underconfident? --- ## Checkpoint 4: Phase 4 Synthesis **NOW/NEXT/NEVER Decisions:** - NOW: [immediate actions] - NEXT: [queued after validation] - NEVER: [anti-patterns to avoid] **Meta-Learning:** - What did this research teach me about research? - What would I do differently next time? --- ## Final Summary **Confidence in Final Output:** [%] **Primary Risk Factor:** [what could make this wrong] **Recommended Follow-Up:** [if any] ``` ## When to Update **Mandatory (at every phase transition):** - Phase 0 → 1: Initial Analysis (after brief complete) - Phase 1 → 2: Checkpoint 1 - Phase 2 → 3: Checkpoint 2 - Phase 3 → 4: Checkpoint 3 - Phase 4 complete: Checkpoint 4 **Optional (during phases):** - After major insight discovered - When assumption violated - When significant pivot made ## What to Log vs. Skip **Always log:** Assumptions, blind spot checks, confidence calibrations, red team challenges, surprising findings, hypothesis updates. **Skip:** Verbatim source content (link instead), redundant confirmations, generic observations without specifics. ## Writing Style Be candid: "I'm making a big assumption here that might be wrong." Use questions: "What if the opposite were true?" Be specific: Not "need to be careful" but "if assumption A is wrong, recommendation 1 breaks because B." ## Edge Cases **Simple research:** Still create file. Keep checkpoints brief (1-2 sentences each). **Research pivots midstream:** Add a PIVOT checkpoint between regular checkpoints. Document what changed and why. **Sequential research (chained):** Create NEW thinking file with incremented version. Reference previous thinking file in Initial Analysis section.