@DeRonin_: This article made me rewrite 4 sections of my automated content engine tonight content engine v2: https://x.com/DeRonin…

This article made me rewrite 4 sections of my automated content engine tonight content engine v2: https://x.com/DeRonin_/status/2048078915520901242… what I'm pulling from his setup into mine: [ 1. the writer context packet template ]: was dumping full audience DNA + 15 KOL profiles + viral signals into every draft prompt switched to per-post packets, 400-900 tokens, only the slices THIS post actually needs small packet beats giant context window every time drafts come back sharper on the first try [ 2. the viral postmortem prompt ]: added as the final verifier before anything queues forces the model to point at EXACT lines that did the work, not generic praise most drafts pass my normal verifier and fail this one that gap is where the real edits live [ 3. content object as a folder, not a database row ]: was tracking each post as a row in SQLite with state columns switched to per-post run folders with idea.md, brief.md, draft.md, feedback.md simpler impossible to skip lifecycle steps each post carries its own state [ 4. bookmarkability rubric scoring ]: every draft now scored 0-2 on 6 criteria before it queues below 8/12 goes back to the brief, not the trash most "bad" drafts are good drafts that skipped one row of the rubric [ what my pipeline looks like now ]: SCRAPERS (9 sources) ↓ SIGNAL LAYER (external) + KNOWLEDGE GRAPH (audience DNA, content DNA, voice rules, KOL profiles, proof bank) ↓ IDEA GATE routes to: original / repurpose / rewrite / research ↓ RUN FOLDER one per post, carries full lifecycle ↓ WRITER CONTEXT PACKET per-post, 400-900 tokens only ↓ DRAFT ↓ VIRAL POSTMORTEM + AVOID-SLOP PASS ↓ BOOKMARKABILITY RUBRIC (8/12 minimum to ship) ↓ POSTIZ SCHEDULER ↓ FEEDBACK LOOP (24h / 72h) ↓ UPDATES STORES: audience DNA, voice rules, banned patterns, hooks skipping his two-model split (opus + gpt-5.5) for now one model + a tight skill graph still handles my volume article is worth reading in full these four patterns are what I'd implement first if you're building any kind of content system