Wiki
Wordnerds Knowledge Wiki

How the Wiki Works

Five layers, read top to bottom. Raw inputs enter at the top and flow down — synthesised into concepts, ratified into canon, then built into the website and other assets at the bottom. One-way flow throughout: inputs never overwrite canon, and the website is always a derivative.

Layer 1 — Foundation
Inputs
Raw material — every call, meeting, document, and piece of research. Two ingest lanes: the repo inbox and the Wiki Google Shared Drive.
Sources
Customer calls Prospect calls Wordnerds meetings Documents Research
Ingest lanes
Repo lane — inputs/inbox/ Drive lane — Wiki Shared Drive / Inbox
⚡ Auto-ingest · daily at 09:30
Humans contribute. Claude processes.
Repo lane: drop a file in inputs/inbox/ and it flows automatically. Drive lane: upload to the Wiki Shared Drive Inbox — wiki-ingest fetches via the Drive API, host-converts PDF/docx/images, and archives the original on success.
Daily (09:30): a LaunchAgent runs wiki-ingest --auto unattended. Clean files go end-to-end; anything needing judgement escalates to needs-review/.
↓ auto-synthesis on every ingest (wiki-ingest)
Layer 2 — Synthesis
Concept Pages
What we believe — distilled from inputs and kept current across 8 domains. Contradictions logged in decisions.md. Humans review; they don't gate every update.
systm aeo customers prospects competitors concepts internal wordnerds-os
Claude manages. Humans review.
Auto-synthesised from every input, with sources: traceability. Trend signals are written to trends-buffer.md — never auto-promoted without a PR.
Sensitivity gate (SCHEMA v0.10): class-scoped — public third-party scrapes bypass it; Wordnerds-generated or customer-touching material gets the full gate.
↓ trend promotion + canon-drift detection (via PRs)
Layer 3 — Ratified Knowledge
Canonical Assets
The single source of truth — the official Wordnerds documents. Humans ratify all changes via pull request; draft PRs auto-merge after 96 hours with no objection.
Strategic
Positioning Messaging House Jobs to be Done Personas Entity canon Questions by stage
Methodology
VoC ROI pyramid AEO playbook AEO content patterns AEO measurement
Claude proposes. Humans ratify.
wiki-canon-drift drafts updates where the wiki has moved ahead of canon. wiki-trend-digest promotes trends that hit the evidence threshold. wiki-steward (weekdays 08:30 UTC) lazy-consensus-merges PRs after 96h of silence.
Humans accept / reject / comment · auto-merge after 96h
↓ canonical assets power the brief (wiki-query retrieves on demand)
Layer 4 — Built From the Wiki
Derivative Assets
Everything built from the canonical foundation. The website (Layer 5) is the primary automated output; the rest are human-built with Claude assisting where coverage is solid.
Website ↓ Layer 5 Sales decks CX Corner Webinars Customer comms Social & blog
Claude assists. Humans ship.
Website pages follow the structured pipeline below. Other outputs draw on the same canon, with a human owning the final cut.
Claude assists · Humans ship
↓ human-reviewed build pipeline (wordnerds-site)
Layer 5 — Website Build Pipeline
From Wiki to Published Page
A 7-stage pipeline in the wordnerds-site repo, ending in a review gate (stage 7) that QAs the rendered page for strategy, copy, AEO, SEO, conversion/UX balance and fact-check before a human deploys. Claude drives each stage; the renderer turns a JSON schema into self-contained HTML — no Astro yet (that's Phase 6).
1
Brief
Content Strategist
briefs/
2
Layout
Design Orchestrator
.layout-spec
3
AEO
AEO Strategist
.aeo-spec
4
Style
Style Director
.styled-spec
5
Copy
Copy Composer
04-page-copy
6
Schema
Schema Composer → build
.schema.json
7
Review
Review Page · QA gate
reviews/
↗
Deploy
human → Cloudflare
deploy:<slug>
Claude drafts. Humans review & ship.
wiki-query retrieves canon for the brief. Each stage emits a file; the Schema Composer builds the HTML, then review-page grades it (BLOCKER / WARNING / NOTE) — any BLOCKER means "not ready". Pete deploys from the staging preview only once it's clean.
Staging: preview.wordnerds.ai → live