PODCAST_RESONANCE_TRAINING.md
1 # Podcast Resonance Training Procedure 2 3 ## Overview 4 5 This procedure trains the Theory of Mind system by predicting which podcast ideas will resonate with Rick, then comparing predictions to actual selections. 6 7 **Why this matters:** The delta between prediction and actual reveals gaps in the cognitive fingerprint model. Each training session improves the system's ability to surface relevant content before Rick explicitly requests it. 8 9 --- 10 11 ## The Pipeline 12 13 ``` 14 ┌─────────────────────────────────────────────────────────────────┐ 15 │ PODCAST TRAINING PIPELINE │ 16 ├─────────────────────────────────────────────────────────────────┤ 17 │ │ 18 │ 1. CAPTURE 2. TRANSCRIBE 3. PREDICT │ 19 │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ 20 │ │ Overcast │ → │ Podwise │ → │ Cognitive│ │ 21 │ │ OPML │ │ Markdown │ │Fingerprint│ │ 22 │ │ Export │ │ Transcript│ │ Analysis │ │ 23 │ └──────────┘ └──────────┘ └──────────┘ │ 24 │ ↓ ↓ ↓ │ 25 │ Recently played Key ideas Resonance │ 26 │ episodes extracted predictions │ 27 │ │ 28 ├─────────────────────────────────────────────────────────────────┤ 29 │ │ 30 │ 4. PRESENT 5. REVEAL 6. TRAIN │ 31 │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ 32 │ │ Obsidian │ → │ Rick │ → │ Compare │ │ 33 │ │ Note │ │ Reorders │ │ & Update │ │ 34 │ │ Template │ │ & Comments│ │Fingerprint│ │ 35 │ └──────────┘ └──────────┘ └──────────┘ │ 36 │ ↓ ↓ ↓ │ 37 │ Prediction table Ground truth New gravity │ 38 │ for review revealed wells, patterns │ 39 │ │ 40 └─────────────────────────────────────────────────────────────────┘ 41 ``` 42 43 --- 44 45 ## Step-by-Step Procedure 46 47 ### Step 1: CAPTURE - Export Overcast Data 48 49 **What:** Export your podcast listening history from Overcast. 50 51 **How:** 52 1. Open Overcast app 53 2. Settings → Export OPML 54 3. Save to `~/Downloads/overcast.opml` 55 56 **Or ask Claude:** 57 > "What have I listened to recently?" 58 59 Claude will parse the OPML and show recent listens. 60 61 --- 62 63 ### Step 2: TRANSCRIBE - Get Podcast Transcript 64 65 **What:** Obtain a transcript of the episode you want to train on. 66 67 **Options:** 68 1. **Podwise** (preferred) - AI-generated transcripts with highlights 69 - Download markdown export to `~/Downloads/` 70 2. **Manual transcript** - From podcast show notes 71 3. **Whisper transcription** - Generate locally from audio 72 73 **Convention:** Name transcript files with podcast/episode name for matching. 74 75 --- 76 77 ### Step 3: PREDICT - Claude Analyzes and Predicts 78 79 **What:** Claude reads the transcript and predicts which ideas will resonate. 80 81 **Ask Claude:** 82 > "I listened to [podcast episode]. Can you analyze the transcript at [path] and predict which ideas resonated with me?" 83 84 **What Claude does:** 85 1. Parses transcript into discrete ideas 86 2. Scores each idea against cognitive fingerprint: 87 - Gravity well proximity 88 - Pattern preference match 89 - Altitude alignment 90 - Resonance trigger detection 91 3. Ranks ideas by predicted resonance 92 4. Outputs prediction table with confidence scores 93 94 --- 95 96 ### Step 4: PRESENT - Create Training Note 97 98 **What:** Claude creates an Obsidian note with predictions for Rick to review. 99 100 **Output location:** `/Users/rcerf/repos/Sovereign_Estate/training/` 101 102 **Note structure:** 103 ```markdown 104 # Resonance Prediction Training: [Episode Title] 105 106 ## Claude's Predictions 107 | Rank | Idea | Resonance Factors | Confidence | 108 |------|------|-------------------|------------| 109 | 1 | ... | ... | 0.92 | 110 111 ## Rick's Actual Selections 112 | Rank | Idea | Commentary | 113 |------|------|------------| 114 | 1 | | | 115 116 ## Ideas Claude Missed 117 | Idea | Why it resonated | 118 |------|------------------| 119 120 ## Ideas Claude Over-Predicted 121 | Idea | Why it didn't land | 122 |------|-------------------| 123 ``` 124 125 --- 126 127 ### Step 5: REVEAL - Rick Annotates 128 129 **What:** Rick fills in actual selections with commentary. 130 131 **Instructions:** 132 1. Open the training note in Obsidian 133 2. In "Rick's Actual Selections": 134 - Copy ideas that actually resonated 135 - Rank them by strength of resonance 136 - Add commentary explaining WHY they resonated 137 3. In "Ideas Claude Missed": 138 - Add any ideas that resonated but weren't in Claude's top predictions 139 - Explain what made them resonate 140 4. In "Ideas Claude Over-Predicted": 141 - Note any high-confidence predictions that didn't actually resonate 142 - Explain why they didn't land 143 144 **The commentary is crucial** - it reveals the cognitive patterns that the fingerprint model should learn. 145 146 --- 147 148 ### Step 6: TRAIN - Update Cognitive Fingerprint 149 150 **What:** Claude reads Rick's annotations and updates the model. 151 152 **Ask Claude:** 153 > "I've filled in my selections in [training note]. Run the training analysis." 154 155 **What Claude does:** 156 1. Computes metrics: 157 - Precision (% of predictions that were correct) 158 - Recall (% of actual resonances that were predicted) 159 - F1 Score 160 2. Identifies new gravity wells from surprising selections 161 3. Adjusts pattern preferences based on what actually resonated 162 4. Updates altitude distribution 163 5. Records training session in cognitive fingerprint history 164 165 --- 166 167 ## Quick Commands 168 169 **List recent listens:** 170 ``` 171 "Show me what podcasts I've listened to in the last 2 weeks" 172 ``` 173 174 **Start training session:** 175 ``` 176 "Create a resonance training session for [episode name]" 177 ``` 178 179 **Run training analysis:** 180 ``` 181 "Analyze my training annotations in [note path]" 182 ``` 183 184 **Show cognitive fingerprint:** 185 ``` 186 "What does my current cognitive fingerprint look like?" 187 ``` 188 189 --- 190 191 ## File Locations 192 193 | File | Path | 194 |------|------| 195 | Overcast OPML | `~/Downloads/overcast.opml` | 196 | Podwise transcripts | `~/Downloads/*.md` | 197 | Training notes | `/Users/rcerf/repos/Sovereign_Estate/training/` | 198 | Cognitive fingerprint | (stored in God Database) | 199 200 --- 201 202 ## Training Session Checklist 203 204 - [ ] Export fresh Overcast OPML 205 - [ ] Identify episode for training 206 - [ ] Obtain transcript (Podwise or other) 207 - [ ] Claude generates predictions 208 - [ ] Review predictions in Obsidian 209 - [ ] Fill in actual selections with commentary 210 - [ ] Claude runs training analysis 211 - [ ] Review cognitive fingerprint updates 212 213 --- 214 215 ## Example Training Session 216 217 **Input:** 218 > "I just listened to the Acquired Playbook episode. The transcript is at ~/Downloads/The Playbook Lessons from 200+ Company Stories.md. Create a training session." 219 220 **Output:** 221 - Training note at: `Sovereign_Estate/training/2026-01-13-acquired-playbook-resonance.md` 222 - Claude's top 10 predictions with confidence scores 223 - Empty tables for Rick's annotations 224 225 **After annotation:** 226 > "I've filled in my selections. Run training." 227 228 **Output:** 229 - Precision: 60%, Recall: 80%, F1: 0.69 230 - New gravity wells: "reflexivity", "duration" 231 - Pattern adjustment: +0.1 to FIRST_PRINCIPLES 232 233 --- 234 235 *Procedure Version: 1.0* 236 *Sovereign OS Theory of Mind Pipeline*