Master plan

Status: draft plan for review · Branch: docs/account-manager-training · Owner: Raymond Last updated: 2026-07-04 This is the build plan for a searchable, video-accompanied documentation site. It is the control document for the ~10-hour autonomous build that follows. Read the Open Decisions at the bottom first — a couple of forks are yours to call before I start.

---

​

1. Goal & audiences

Two jobs, one site: The site is written so a single page serves both where possible, with audience tags (For AMs / For self-serve) where the guidance diverges — rather than maintaining two sites. Audience model (resolved): self-serve subscribers can touch everything — the difference is not capability, it’s discipline and standardization. Agency/internal AMs run the rigorous, repeatable version at scale (checklists, quality gates, cadence); self-serve users get the same tools with a lighter-weight path. So audience tags mostly mean “here’s the disciplined version (AM)” vs “here’s the quick version (self-serve)”, not “you can’t do this.” Never mention credits/credit units anywhere in the docs. We are moving off that model; describe throughput and cadence in concrete terms (pages, posts, weeks) with no credit language.

​

2. The three parts (the user’s structure)

1. Part 1 — Theory & the Synscribe way of SEO. Not a regurgitation of generic SEO advice. It codifies what we do differently and why: programmatic SEO + GEO (AI-answer-engine optimization) for SaaS, SERP-cliff-first keyword selection, BOFU-first, speed-to-rank, and agent-operated execution. This is the layer that makes the how-tos make sense.
2. Part 2 — Doing it on the platform. Step-by-step micro-guides, one task per page, each with a video where one exists (or is planned). The literal button-by-button.
3. Part 3 — SOPs for running accounts at scale. The operating procedures for internal AMs and agency partners: onboarding, monthly/weekly cadence, quality gates, client reviews, roles & handoffs, capacity/credit planning.

​

3. Design principles for the site

• One page = one micro-guide. Small, searchable, linkable. A page should be readable in 2–4 minutes and answer exactly one “how do I…” or “why do we…”.
• Video-first where it helps. Every task page has a slot for a video. Some already exist (Loom library + 2 YouTube); the rest go on a shot-list for Raymond to record.
• Real prompts, real examples. The platform is agent-operated, so the “how” is often a prompt you send Pi. We lift real operator prompts from production chats (via the prod-context skill) rather than inventing them — copy-paste-ready, lightly anonymized.
• Grounded, not aspirational. If a feature isn’t live, it’s not taught as if it is (e.g. the guardrail’s real publish-blocking vs. its stale settings copy). Code + prod win over docs.
• Progressive disclosure. Part 1 gives the model; Part 2 the mechanics; Part 3 the cadence. Cross-linked so a reader can drop from “why” straight to “how.”

---

​

4. Full site map

Legend for per-page status (filled during the build): ✍️ draftable now · 🎥 video exists (Loom/YT to embed) · 🎬 needs video (shot-list) · ❓ needs Raymond (knowledge gap) · 🔬 mine from prod chats

​

Part 1 — Theory & the Synscribe Way

​

Part 2 — Doing it on the platform (micro-guides)

2.0 Orientation 2.1 Account setup 2.2 Keywords & planning 2.3 Landing pages 2.4 Blog content 2.5 Publish & distribute 2.6 Analytics & review

​

Part 3 — SOPs for running accounts at scale

---

​

5. Source ledger (what feeds the build)

• Workshop Notes (Notion, 69k chars) — Part 1 philosophy backbone. To be mined in Phase 1.
• SOP database (Notion) — Part 3 backbone; several are already high quality (Onboarding Month 1, Organise Keyword). Port + normalize, don’t rewrite what’s already sharp.
• Loom library (~20 videos, catalogued from the SOP page) + 2 YouTube how-tos — embed map.
• Client dossiers (Wonderchat, Hyperbound) — real worked examples for Part 1/3 case studies and Part 2 illustrations (pillar design, competitor mining, BOFU strategy).
• Production operator chats via prod-context (factory session/messages) — real prompts and longitudinal patterns (how the same operator handles a cycle over months). Feeds the copy-paste prompt blocks throughout Part 2 & the “what good looks like” in Part 3.
• Codebase (already mapped) — surfaces, routes, skills, integrations; keeps the how-tos honest.
• Existing repo training docs (docs/training/account-manager/*) — my earlier product map; fold in / supersede.

---

​

6. The 10-hour phase plan

Each phase has a mini-goal (the acceptance check) and a concrete deliverable. Phases are ordered so the site is usable even if we stop early (index + Part 3 SOPs land first, since those are what next week’s AMs need most).

​

Phase 0 — Scaffold & decide (≈0.75h)

• Stand up the doc-site skeleton (tooling per Open Decision #1), nav for the 3 parts, search, a video-embed component, and the audience-tag convention.
• Mini-goal: site builds locally; a placeholder page renders with an embedded video and is searchable.

​

Phase 1 — Mine the sources (≈2h) (runs as parallel subagents)

• Mine Workshop Notes → a distilled “Synscribe SEO principles” outline for Part 1.
• Mine prod-context operator chats (3–5 exemplary sessions across a couple of orgs, over time) → a real-prompt library tagged by task, + patterns for Part 3.
• Catalogue every Loom/YouTube → an embed map (video ↔ page).
• Confirm VFS filing conventions (folders, memory, soul).
• Mini-goal: a _sources/ folder with (a) principles outline, (b) prompt library, (c) video embed map, (d) VFS conventions note — every downstream page has its raw material staged.

​

Phase 2 — Part 3 SOPs (≈2h) (first, because next week needs it)

• Port + normalize the existing SOPs into micro-guides; fill the gaps I can; flag the rest.
• Mini-goal: all 16 Part 3 pages exist; the 4 already-strong SOPs are fully ported; ❓ pages have a clear “what I need from you” stub.

​

Phase 3 — Part 2 platform how-tos (≈2.5h)

• Write every task page: steps + real prompt block + embedded video (or 🎬 shot-list stub).
• Mini-goal: all Part 2 pages exist; every page with an existing video has it embedded; every video-less page has either full written steps or a shot-list entry.

​

Phase 4 — Part 1 theory (≈2h)

• Write the differentiated-SEO chapters from the mined principles + client examples.
• Mini-goal: all 12 Part 1 pages drafted; each states a distinct Synscribe stance (not generic SEO), with a real example; ❓ positioning nuances flagged for Raymond.

​

Phase 5 — Assemble, cross-link, gap register, shot-list (≈0.75h)

• Home page + learning paths (“New AM start here”, “Self-serve tour”); cross-links between parts; the consolidated knowledge-gap register and video shot-list as their own pages.
• Mini-goal: a coherent, navigable, searchable site; every gap and every needed video is itemized in one place for Raymond to action.

Buffer (≈0.25h): commit hygiene, a build check, and a written hand-off summary.

If a decision blocks me mid-run, I proceed on the recommended default, mark the spot with ❓, and keep moving — I won’t stall the whole run waiting.

---

​

7. Video strategy

• Embed what exists: ~20 Looms + 2 YouTube already cover a big chunk of Part 2. Phase 1 produces the exact video↔page map.
• Shot-list for the rest: every 🎬 needs video page gets an entry with: the one task it shows, a 3–5 beat outline, and est. length. Raymond records against the list; each page has the embed slot ready so dropping a URL in is a one-line edit.
• Self-serve priority: for audience B, prioritize recording 2.0.x (orientation) + 2.1.x (setup) — that’s where a lost subscriber needs to see it most.

​

8. Knowledge-gap register (mostly resolved 2026-07-04)

Resolved by Raymond:

1. Part 1 positioning — I own the draft; Raymond reviews. Mine Workshop Notes + client dossiers for the sharp version. (1.1, 1.7, 1.12)
2. Credits — do not mention at all. No credit language anywhere; describe cadence in pages/posts/weeks. (3.16)
3. Agency vs self-serve — self-serve can do everything; the delta is discipline/rigor/ standardization at scale, not gated capability. (3.15, audience tags)
4. VFS conventions — confirm the real folder structure from prod chats + code, then document.
5. prod-context access — key provided and verified working.
6. Exemplary chats/orgs — pick from the high-activity orgs (wonderchat, chalkie, frameshft, straitsx, rently-uae, bluente, bookaclinic, scheduling-wizard, …).

Remaining low-stakes gap, handled inline with ❓ where hit:

• Any org-specific claim I can’t verify against code/prod/chats gets a ❓ marker rather than a guess, and lands in the consolidated gap register page for Raymond’s pass.

​

The Definition of Done (what the goal-loop judge checks) — see GOAL.md

The autonomous build is driven by docs/training/GOAL.md, whose acceptance gates encode “a GREAT outcome,” not merely “all pages exist.”

---

​

9. Decisions (resolved 2026-07-04)

1. Format: portable Markdown in-repo under docs/training/ this run; pick the site generator (Starlight/Mintlify/etc.) in a later tooling pass. Everything is authored so it drops into a generator unchanged (clean front-matter, relative links, video-embed slots).
2. Audience: one site with audience tags — shared pages with For AMs / For self-serve callouts where guidance diverges; Part 3 is AM-only.
3. Prompt mining: live — FACTORY_API_KEY provided and verified working (curated + sql tiers; reads factory.agent_session/messages). Rich source confirmed (wonderchat, chalkie, frameshft, straitsx, rently-uae, bluente, bookaclinic, …). Mining routes through subagents to keep transcripts out of the main context.