# PKA Whitepaper — Rework Recommendation **From:** Vox **To:** Larry (review) → Jimmie **Date:** 2026-05-19 **Re:** PKA AI Team — Complete Synopsis v6 → v7 --- ## TL;DR **Rework, not additive.** v6 is an excellent *internal* spec. It will read like a wiki page to an outside reader. If this is going to anchor a course, a business, or a piece you publish under your name, it needs a different opening, a different voice, and a different shape. I am proposing v7 as a structural rework. Today's accomplishments are folded in, but they are not the headline — they are an example of the pattern the rework is built around. A first draft of v7 is in the same folder as this doc: `2026-05-19 - Vox - PKA Whitepaper v7 (first draft).md`. --- ## What v6 Does Well - The team roster table is the single best artifact in the doc. Keep it almost as-is. - "What It Has Actually Done" is the credibility section. External readers will need this earlier and louder. - "What Makes This Different" (the It's Not / It Is table) is the clearest articulation of the thesis. It should move forward. - The numbers section is a strong closer. Keep it. ## What v6 Does Not Do (for an external audience) 1. **No opening narrative.** It begins with a definition ("PKA is..."), not a tension. An outside reader needs a reason to read past line one. Why did this get built? What was broken? 2. **Voice is product-spec, not story.** Lines like "single 16-table schema covering documents, clients, projects..." read as documentation, not as a piece a stranger would forward. Internal readers want the spec. External readers want the *why* behind the spec. 3. **Lessons are buried.** The most teachable parts (how skills are built, how memory is structured, what gets delegated vs. killed) are described as features, not principles. Course material lives in the principles. 4. **It assumes the reader knows you.** "The owner is Jimmie Needles" lands as a credit line, not as a hook. For external IP, your story is the hook, not a footnote. 5. **No tension or failure.** Every section is a win. Outside readers trust pieces more when the author names what did not work. v6 has no Sage/Finn retirement story, no Maya/Lane/Lux consolidation story told as a lesson. It is mentioned in a footer cell. 6. **Tech stack reads as proof, not narrative.** A long table of tools makes you look credentialed but does not make a reader care. The tools belong in an appendix or sidebar. ## Recommended Structural Changes for v7 | v6 Section | v7 Treatment | |------------|--------------| | "What It Is" (definition opener) | **Replaced with a 3-paragraph cold open**: the friction that started it, the moment it clicked, the shape of what it became. | | "How It Works" (mechanics) | Reframed as **"The Four Design Choices That Made It Work"** — orchestrator, skills-as-team-members, persistent memory, file-based deliverables. Each one as a principle, not a feature. | | "The Atlas Stack" (deep tech) | Compressed into a sidebar called **"The One App You Have To Build"** — keeps the credibility, drops the schema dump. | | "What It Has Actually Done" | Promoted to **third section**, opens with the single most impressive beat (the 565-check VIP Play push) before listing the rest. Readers stop scrolling at proof, not at description. | | "What Makes This Different" | Kept almost verbatim. It works. | | "The Stack" table | Moved to an **appendix**. | | "By the Numbers" | Kept as closer. | | *(new)* | **"Lessons I Did Not Expect"** — the retirements (Sage, Finn, Maya/Lane/Lux), the why-it-failed first, then what replaced it. This is the section that makes the piece courseworthy instead of a brag sheet. | | *(new)* | **"What This Is Not"** — a single paragraph that closes off the "is this just a chatbot?" question for a skeptical reader. | ## Voice Rules Applied to v7 - First person where it earns its keep. v6 narrates in third person ("The owner is Jimmie Needles"). v7 lets you speak in places where the I matters. - No jargon without a translation. v6 says "tsvector full-text search" once with no glossing. v7 either explains it in plain English or cuts it. - Show the work, then name the pattern. Every claim is followed by a specific example (a client, a dollar amount, a session). - No em dashes (brand rule). Used ellipses, commas, or sentence breaks instead. - Short paragraphs. Two to four sentences. Easier to skim, easier to quote. ## What Today (2026-05-19) Got Folded Into v7 Pulled from Atlas's `project_events` table for today's working day: 1. **Atlas Day 2 shipped** — POST `/api/project-events` ingest endpoint with Bearer auth, idempotency, and an inline Haiku classifier; PUT `/api/project-events/[id]/assign` re-tag endpoint; `/projects/unclassified` bucket UI with confidence display and one-click tagging. All 5 endpoint tests green. *This is the most courseworthy beat today: the AI team now logs its own work to a shared event stream that another AI auto-classifies into the right project bucket. It is the "continuity system" that makes a multi-agent workforce possible without a human reading every log.* 2. **Afton Electric QBO OAuth handshake done** (realm_id=1331299185). Ledger's pilot client is now live in the QBO integration. 3. **Loom delivered the Prospect Dashboard scoping doc** with a strawman SIPOC, an Atlas-vs-CRM tradeoff, and 7 audit questions before any build starts. Demonstrates the "map before automating" discipline that is now central to how this team operates. 4. **Two new team members on the roster**: Tally (AP Specialist, hired 2026-05-19) and Loom (Workflow Architect). Team is now 16+ specialists, not 14 as v6 says. Today's beats are not the headline of v7. They are example material woven into the structural sections that already need to exist. --- ## Open Questions for Jimmie Before I Lock v7 Three judgment calls only you can make. v7 ships as draft until you answer. 1. **Audience.** Is the primary external reader (a) other bookkeepers / accounting firm owners considering this for their practice, (b) a general business / "future of work" audience, or (c) people who would buy a course on how to do this themselves? The structure I drafted leans toward (c) but works for (a) with light edits. (b) needs a different rewrite. 2. **Author voice.** Is this published under your name as a personal piece, or as a J2 Bookkeeping artifact? The first warrants more "I" and personality. The second stays more institutional. 3. **Public scope.** Some of v6 names specific dollars, specific clients (contractor with $144K payroll, VIP Play 565 checks). Do those stay in for the external version, or do they get anonymized? I have left them in v7 with placeholders flagged in inline comments. --- ## Next Steps If You Approve 1. You answer the three questions above. 2. I lock v7, run it past Larry for a final review pass. 3. v7 becomes the new canonical doc, replacing v6 in `Owner Inbox/`. 4. Loom's daily-feed mechanism (separate doc in this folder) starts piping curated daily beats to me, and I publish v7.1, v7.2, etc. on a cadence we agree on. My recommendation: weekly minor revisions, monthly major. This piece has real legs. Worth doing it right.