{"project":"youtube-scripting","memory":"{\"id\":\"m_20251112_0001\",\"type\":\"goal\",\"body\":\"Deliver world class packaging before any script body is written.\",\"tags\":[\"lever_packaging_first\",\"outcome_high_retention\"],\"source\":\"chat:2025-11-12T18:00:00Z#u\",\"created_at\":\"2025-11-12T18:00:00Z\",\"updated_at\":\"2025-11-12T18:00:00Z\",\"supersedes\":null} {\"id\":\"m_20251112_0002\",\"type\":\"definition\",\"body\":\"Packaging means title, thumbnail text, hook, and intro.\",\"tags\":[\"concept_packaging\",\"concept_title_thumbnail_hook_intro\"],\"source\":\"chat:2025-11-12T18:00:01Z#u\",\"created_at\":\"2025-11-12T18:00:01Z\",\"updated_at\":\"2025-11-12T18:00:01Z\",\"supersedes\":null} {\"id\":\"m_20251112_0003\",\"type\":\"decision\",\"body\":\"Use Twitter first for research and web sources second for corroboration.\",\"tags\":[\"lever_research_order\",\"subcap_twitter_first\"],\"source\":\"chat:2025-11-12T18:00:02Z#u\",\"created_at\":\"2025-11-12T18:00:02Z\",\"updated_at\":\"2025-11-12T18:00:02Z\",\"supersedes\":null} {\"id\":\"m_20251112_0004\",\"type\":\"constraint\",\"body\":\"Stop after producing exactly twelve packaging items.\",\"tags\":[\"guardrail_hard_stop\",\"lever_scope_control\"],\"source\":\"chat:2025-11-12T18:00:03Z#u\",\"created_at\":\"2025-11-12T18:00:03Z\",\"updated_at\":\"2025-11-12T18:00:03Z\",\"supersedes\":null} {\"id\":\"m_20251112_0005\",\"type\":\"project\",\"body\":\"Run one project per workflow to keep scope and comparisons clean.\",\"tags\":[\"lever_project_scoping\",\"flow_one_project_per_workflow\"],\"source\":\"chat:2025-11-12T18:00:04Z#u\",\"created_at\":\"2025-11-12T18:00:04Z\",\"updated_at\":\"2025-11-12T18:00:04Z\",\"supersedes\":null} {\"id\":\"m_20251112_0006\",\"type\":\"resource\",\"body\":\"Load YouTube Scriptwriter.md to build the body after packaging is locked.\",\"tags\":[\"concept_framework_body\",\"concept_youtube_scriptwriter_doc\"],\"source\":\"chat:2025-11-12T18:00:05Z#u\",\"created_at\":\"2025-11-12T18:00:05Z\",\"updated_at\":\"2025-11-12T18:00:05Z\",\"supersedes\":null} {\"id\":\"m_20251112_0007\",\"type\":\"resource\",\"body\":\"Load YouTube Synthetic Data.md as the synthetic source of truth with a visible date.\",\"tags\":[\"concept_synthetic_data_doc\",\"guardrail_claim_verification\"],\"source\":\"chat:2025-11-12T18:00:06Z#u\",\"created_at\":\"2025-11-12T18:00:06Z\",\"updated_at\":\"2025-11-12T18:00:06Z\",\"supersedes\":null} {\"id\":\"m_20251112_0008\",\"type\":\"decision\",\"body\":\"Ask once for audience, outcome, and constraints if they are missing.\",\"tags\":[\"lever_require_context\",\"subcap_intake_questions\"],\"source\":\"chat:2025-11-12T18:00:07Z#u\",\"created_at\":\"2025-11-12T18:00:07Z\",\"updated_at\":\"2025-11-12T18:00:07Z\",\"supersedes\":null} {\"id\":\"m_20251112_0009\",\"type\":\"decision\",\"body\":\"Label every claim as CONFIRMED, PROVISIONAL, or UNVERIFIED.\",\"tags\":[\"subcap_claim_labels\",\"lever_evidence_discipline\"],\"source\":\"chat:2025-11-12T18:00:08Z#u\",\"created_at\":\"2025-11-12T18:00:08Z\",\"updated_at\":\"2025-11-12T18:00:08Z\",\"supersedes\":null} {\"id\":\"m_20251112_0010\",\"type\":\"constraint\",\"body\":\"Exclude unverified claims from packaging or frame them as a question.\",\"tags\":[\"guardrail_claim_quality\",\"lever_evidence_threshold\"],\"source\":\"chat:2025-11-12T18:00:09Z#u\",\"created_at\":\"2025-11-12T18:00:09Z\",\"updated_at\":\"2025-11-12T18:00:09Z\",\"supersedes\":null} {\"id\":\"m_20251112_0011\",\"type\":\"constraint\",\"body\":\"Pause after packaging and wait for the selection marker.\",\"tags\":[\"guardrail_system_pause\",\"flow_two_stage_packaging\"],\"source\":\"chat:2025-11-12T18:00:10Z#u\",\"created_at\":\"2025-11-12T18:00:10Z\",\"updated_at\":\"2025-11-12T18:00:10Z\",\"supersedes\":null} {\"id\":\"m_20251112_0012\",\"type\":\"decision\",\"body\":\"Audit the chosen packaging against synthetic rules before writing the body.\",\"tags\":[\"lever_quality_control\",\"subcap_packaging_audit\"],\"source\":\"chat:2025-11-12T18:00:11Z#u\",\"created_at\":\"2025-11-12T18:00:11Z\",\"updated_at\":\"2025-11-12T18:00:11Z\",\"supersedes\":null} {\"id\":\"m_20251112_0013\",\"type\":\"decision\",\"body\":\"Write the script body only after packaging is locked.\",\"tags\":[\"lever_sequencing\",\"flow_packaging_before_body\"],\"source\":\"chat:2025-11-12T18:00:12Z#u\",\"created_at\":\"2025-11-12T18:00:12Z\",\"updated_at\":\"2025-11-12T18:00:12Z\",\"supersedes\":null} {\"id\":\"m_20251112_0014\",\"type\":\"research\",\"body\":\"Capture handle, url, date, and a one line paraphrase for each Twitter item.\",\"tags\":[\"subcap_research_metadata\",\"lever_traceability\"],\"source\":\"chat:2025-11-12T18:00:13Z#u\",\"created_at\":\"2025-11-12T18:00:13Z\",\"updated_at\":\"2025-11-12T18:00:13Z\",\"supersedes\":null} {\"id\":\"m_20251112_0015\",\"type\":\"synthesis\",\"body\":\"Keep the output order predictable from header through audits.\",\"tags\":[\"flow_output_order\",\"lever_process_discipline\"],\"source\":\"chat:2025-11-12T18:00:14Z#u\",\"created_at\":\"2025-11-12T18:00:14Z\",\"updated_at\":\"2025-11-12T18:00:14Z\",\"supersedes\":null} {\"id\":\"m_20251112_0016\",\"type\":\"constraint\",\"body\":\"Use grade five clarity with short sentences and simple words.\",\"tags\":[\"guardrail_reading_level\",\"lever_clarity\"],\"source\":\"chat:2025-11-12T18:00:15Z#u\",\"created_at\":\"2025-11-12T18:00:15Z\",\"updated_at\":\"2025-11-12T18:00:15Z\",\"supersedes\":null} {\"id\":\"m_20251112_0017\",\"type\":\"constraint\",\"body\":\"Keep the voice direct and respectful with no emojis or profanity.\",\"tags\":[\"guardrail_style\",\"lever_tone\"],\"source\":\"chat:2025-11-12T18:00:16Z#u\",\"created_at\":\"2025-11-12T18:00:16Z\",\"updated_at\":\"2025-11-12T18:00:16Z\",\"supersedes\":null} {\"id\":\"m_20251112_0018\",\"type\":\"decision\",\"body\":\"Optionally use project only memory so improvements compound over runs.\",\"tags\":[\"lever_memory_scope\",\"flow_compound_improvements\"],\"source\":\"chat:2025-11-12T18:00:17Z#u\",\"created_at\":\"2025-11-12T18:00:17Z\",\"updated_at\":\"2025-11-12T18:00:17Z\",\"supersedes\":null}","submittedAt":"2025-11-12T18:01:05.838Z","formMode":"production"}
{"project":"youtube-scripting","memory":"{\"date\":\"2025-11-15\",\"fact\":\"I’m building a YouTube video that shows a 3-level AI memory system for ChatGPT, Claude, Grok, etc.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 1 uses one project-only memory file per topic (e.g., YouTube, fitness, client, Bible study).\"} {\"date\":\"2025-11-15\",\"fact\":\"In Level 1 I turn OFF reference past chats and turn ON project-only memory to keep recall precise.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 1 workflow: talk normally, then use commands like 'Save this memory' or 'Save this conversation' to update memory.md.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 1 updates work by downloading the new memory file, re-uploading it to the project, and deleting the old one.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 2 moves the same memory idea onto a Hostinger n8n VPS so memories live on a server I control.\"} {\"date\":\"2025-11-15\",\"fact\":\"In Level 2 the AI prints clean JSONL memory lines in a code block that I copy into an n8n form.\"} {\"date\":\"2025-11-15\",\"fact\":\"The n8n form appends those JSONL lines into per-project memory files on the VPS (e.g., youtube-memory.jsonl).\"} {\"date\":\"2025-11-15\",\"fact\":\"Because memory lives on the VPS, I can point multiple AIs at the same memory link instead of being locked into one platform.\"} {\"date\":\"2025-11-15\",\"fact\":\"I’m recommending Hostinger’s n8n KVM2 plan and coupon code COREY for ~2 vCPU, 8GB RAM and a strong discount.\"} {\"date\":\"2025-11-15\",\"fact\":\"Hostinger’s Kodee AI assistant can help manage the VPS (health checks, security, backups) via natural language.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 3 adds RTK (Relational Tag Kernel) to give each memory tags and scores so recall feels like 'genius mode'.\"} {\"date\":\"2025-11-15\",\"fact\":\"RTK runs as a custom GPT that takes my project docs, then outputs a memory tag list and an edges file.\"} {\"date\":\"2025-11-15\",\"fact\":\"The RTK edges file encodes how tags relate (very low → very strong) so the AI can prioritize the most relevant memories.\"} {\"date\":\"2025-11-15\",\"fact\":\"Live tagging in Level 3 adds hidden chips under important replies so saves and retrieval can reuse those tags automatically.\"} {\"date\":\"2025-11-15\",\"fact\":\"I created three handbooks (Method 1, Method 2, Method 3) as the download pack so viewers can follow detailed setup later.\"} {\"date\":\"2025-11-15\",\"fact\":\"The video packaging is locked: big promise about changing AI with memory, plus 'Install Genius Mode Now' style thumbnail text.\"} {\"date\":\"2025-11-15\",\"fact\":\"I’m editing in Descript and using simple diagrams and B-roll instead of full-length tutorials to keep retention high.\"}","submittedAt":"2025-11-16T04:08:29.834Z","formMode":"test"}
{"project":"youtube-scripting","memory":"{\"date\":\"2025-11-15\",\"fact\":\"I’m building a YouTube video that shows a 3-level AI memory system for ChatGPT, Claude, Grok, etc.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 1 uses one project-only memory file per topic (e.g., YouTube, fitness, client, Bible study).\"} {\"date\":\"2025-11-15\",\"fact\":\"In Level 1 I turn OFF reference past chats and turn ON project-only memory to keep recall precise.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 1 workflow: talk normally, then use commands like 'Save this memory' or 'Save this conversation' to update memory.md.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 1 updates work by downloading the new memory file, re-uploading it to the project, and deleting the old one.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 2 moves the same memory idea onto a Hostinger n8n VPS so memories live on a server I control.\"} {\"date\":\"2025-11-15\",\"fact\":\"In Level 2 the AI prints clean JSONL memory lines in a code block that I copy into an n8n form.\"} {\"date\":\"2025-11-15\",\"fact\":\"The n8n form appends those JSONL lines into per-project memory files on the VPS (e.g., youtube-memory.jsonl).\"} {\"date\":\"2025-11-15\",\"fact\":\"Because memory lives on the VPS, I can point multiple AIs at the same memory link instead of being locked into one platform.\"} {\"date\":\"2025-11-15\",\"fact\":\"I’m recommending Hostinger’s n8n KVM2 plan and coupon code COREY for ~2 vCPU, 8GB RAM and a strong discount.\"} {\"date\":\"2025-11-15\",\"fact\":\"Hostinger’s Kodee AI assistant can help manage the VPS (health checks, security, backups) via natural language.\"} {\"date\":\"2025-11-15\",\"fact\":\"Level 3 adds RTK (Relational Tag Kernel) to give each memory tags and scores so recall feels like 'genius mode'.\"} {\"date\":\"2025-11-15\",\"fact\":\"RTK runs as a custom GPT that takes my project docs, then outputs a memory tag list and an edges file.\"} {\"date\":\"2025-11-15\",\"fact\":\"The RTK edges file encodes how tags relate (very low → very strong) so the AI can prioritize the most relevant memories.\"} {\"date\":\"2025-11-15\",\"fact\":\"Live tagging in Level 3 adds hidden chips under important replies so saves and retrieval can reuse those tags automatically.\"} {\"date\":\"2025-11-15\",\"fact\":\"I created three handbooks (Method 1, Method 2, Method 3) as the download pack so viewers can follow detailed setup later.\"} {\"date\":\"2025-11-15\",\"fact\":\"The video packaging is locked: big promise about changing AI with memory, plus 'Install Genius Mode Now' style thumbnail text.\"} {\"date\":\"2025-11-15\",\"fact\":\"I’m editing in Descript and using simple diagrams and B-roll instead of full-length tutorials to keep retention high.\"}","submittedAt":"2025-11-16T04:28:40.684Z","formMode":"production"}