Token Max Factory

token-max-factory — Universal Point-and-Shoot Content Factory

Drives the token-max-site-factory engine through Archon. Most programmatic SEO fails as thin duplicate content or hallucinated local "facts" — this factory attacks both with hard validation gates and a packet system where the packet is the ONLY fact source.

Before generation, require an owner-intent decision. Use the repository's prompt-packs/dataforseo-intelligence/07-owner-intent-map.md when live search evidence is available. Every cluster must be upgrade, create, consolidate, or hold; TokenMax only receives approved create owners and bounded upgrades.

Hard boundary (never violate): the factory NEVER deploys, touches DNS, Search Console, sitemaps, or indexing. Generation ends at a clean validation report in the target repo's worktree. Shipping is a separate human-approved lane. A preflight node re-verifies this contract on every run.

Prerequisites (one-time)

# Archon CLI (free, MIT) — macOS/Linux:
curl -fsSL https://archon.diy/install | bash
# Windows PowerShell:  irm https://archon.diy/install.ps1 | iex

# The engine (free, MIT):
git clone https://github.com/TheSmokeDev/token-max-site-factory
# or, via the Archon marketplace:  archon workflow install token-max-site-factory

The writer runs on YOUR coding-agent subscription: codex at xhigh reasoning by default; claude via explicit install --allow-claude. No SEO APIs, no paid data vendors.

Quality contract (defaults, per page)

2,800-3,400 words (hard fail <2,700) · pairwise + cross-corpus shingle/Jaccard overlap ≤0.10 · ≥8 H2 · ≥4 AI-citable blockquotes · ≥5 FAQ questions · text-to-HTML ≥0.15 · unique title/meta · packet = only fact source · vertical prohibited-claim regexes.

Point-and-shoot: onboard a new site

cd token-max-site-factory

python engine/token_max_site_factory.py new-site --site <id> --target-repo <path>   # 1. scaffold config
python engine/token_max_site_factory.py scan --site <id> --allow-network            # 2. scan live site + local repo (the ONLY network step, manual)
# 3. edit sites/<id>/site.yaml — start from sites/example/site.yaml; see references/site-config.md
python engine/token_max_site_factory.py install --site <id> --run-input "pilot-10"  # 4. stamp workflow shim + marker into target repo
cd <target-repo> && archon workflow run token-max-site-factory-<id> --no-worktree   # 5. run

Prove one gold-standard owner page, then a representative 10-page pilot BEFORE any tranche. Tranches ≤300 pages (pairwise overlap is O(n²)). Review the worktree diff + report, merge, deploy through the site's own lane.

Choose program.scale in site.yaml:

• starter: single operator/site and bounded pilot;
• growth: requires audience, conversion goal, evidence owner, success metrics, primary query, and canonical route owner;
• enterprise: additionally requires named reviewers and authority sources.

Scale adds governance and receipts, not permission to weaken gates. Community or Reddit research may justify one reviewed owner page, never an automated modifier page family.

Running on Archon 0.5.0 — CRITICAL gotchas

1. Run input is BAKED at install time (install --run-input "..."). 0.5.0 bash nodes do NOT receive the user message (regression; fixed on Archon's dev branch). To change what a run does, re-run install with a new --run-input. Input grammar: pilot-10, <phase>, <phase>-tranche-300, --topics a,b --entities x,y --limit 20, --dry-run.
2. Never put dollar-sign shell variables in workflow YAML — 0.5.0 blanks EVERY dollar token in inline node scripts. On-disk scripts (engine/tmsf.sh) are safe. The template is already dollar-free; keep it that way.
3. Archon's bash may be WSL bash — the template's if/else picks the Windows vs /mnt/c/... path form automatically.
4. Prefer --detach for long runs: archon workflow run <name> --no-worktree --detach — survives the launching shell. Track via the batch ledger, not the console.
5. Stale "running" rows block relaunch ("Workflow already active on this path"). Fix: archon workflow status → archon workflow abandon <run-id> (DB-only; safe when the process is dead).
6. until_bash exit-1 "error" log lines during loops are normal — that's the "not done yet" signal; the loop is fine.

Resume & recovery (state is on disk, always resume-safe)

• Resume anything: re-run the same archon workflow run command. prepare --resume --resume-existing-outputs folds in every page already on disk. Never delete <artifacts>/state/batch.json.
• Run died after generation finished? Finish deterministically, no Archon:

  cd <target-repo>
  python <factory>/engine/token_max_site_factory.py validate --site <id>   # hard gate
  python <factory>/engine/token_max_site_factory.py emit --site <id>       # html sites; noop for markdown
  python <factory>/engine/token_max_site_factory.py report --site <id>
  

• Ledger check: remaining --site <id> or read <artifacts>/state/batch.json statuses (pending_generation → generated/regenerated/held_back).

Verification checklist (after every batch)

1. <artifacts>/reports/validation.json → ok: true, 0 held back, 0 overlaps.
2. <artifacts>/reports/batch-report.md — word counts in band and public states false. Check the sibling batch-report.receipts.json for SHA-256 config, prompt, batch, validation, source, and output receipts.
3. HTML sites: spot-open an emitted page; confirm canonical + FAQPage JSON-LD; if the site has its own head-baking tooling, prove it changes NOTHING on factory pages (emitted heads carry rel=canonical → skipped).
4. Confirm pre-existing content lanes untouched (git status on the live tree).

Prompt profiles

local-service-seo, multi-location-enterprise, saas-b2b, ecommerce-category, and regulated-insurance (California insurance rules as the worked regulated example). New vertical = new profile in prompts/profiles/ + site claim_rules_md.

No images are generated anywhere — text-forward by design (GEO extraction favors extractable text).

References

• references/site-config.md — the full site.yaml schema, annotated
• Engine README.md — runbook + troubleshooting
• Engine WORKFLOW.md — the Archon marketplace package doc