THE SEO MONEY MACHINE — SYSTEM v2

The root document. Everything in system/ descends from here. Version: 2.1-dev · branch system-v2 · command center: SMM Agent

0. The mission — VALUEMAXX

We target a niche. In that niche the audience needs information. We are the best at packaging and delivering the valuable information they need, in the easiest-to-consume way possible. That is value maxing. (ML, 2026-07-21.)

Extremely valuable, unique content for the target audience and SEO ranking success are equal, permanent pillars. If the directive to rank ever overpowers the quality directive, the site eventually fails; SEO succeeds by targeting the soft goal — providing people what they need — never ranking tactics. Ranking and keyword data are invaluable inputs the system uses better than any website ever has, and they never decide whether something is good enough to ship. The method is token-maxing in service of value-maxing: the system crushes compute so the audience gets the best.

Mission enforcement is the pipeline's own machinery — acceptance standards, elimination stages, the mechanical gatekeeper, the flood's quality canaries (seo/PIPELINE.md, seo/FLOOD.md). The mission is not the values floor: VALUES.md holds only what the system refuses (law, honest transactions, no shitty things) and is deliberately small, firm, and lightly enforced — worst-outcome prevention, never quality policing.

1. What the system is

The SEO Money Machine is an organization-builder. Pointed at a niche, it constructs a complete institution of workers — an organization that does the work, watches the work, improves the work, knows itself, and reports itself truthfully to its human. The website is the organization's product surface; the organization is the product.

The unit of reliable output is the institution, not the clever worker. One mind's answer — however smart — is ungated, unaudited, uncorroborated, and decaying from the day it ships. Verification, corroboration, monitoring, self-audit, knowledge custody, and compounding improvement are properties of an organization. So the system builds one.

The pods principle (ML, 2026-07-20): every corner of the system is its own little self-sustaining pod — a function plus everything needed to gate it, measure it, detect its drift, diagnose it, improve it, and verify the improvement — managed by increasingly collective self-sustaining pods all the way up the chain. No orphan functions. No open loops. An open loop drains its missing work to the human; the design goal is a human who is informed always, required almost never.

2. The doctrine hierarchy

1. VALUES.md — the boundary. Constitutional; ML-amendable only; enforced by a dumb un-overridable gate inside every worker.
2. corpus/ — the domain-independent craft: the core operation, the laws, the worker anatomy (class blueprint), loop closure, the reporting funnel, escalation, commissioning. How to build any organization of workers.
3. seo/ — the SEO differentiation: what this machine knows about building content organizations specifically — pipeline, research, personality, flooding, security, architecture, takeover, onramp.
4. specs/ — mechanical contracts: ledger schema, record types, the consequence dial, gate templates, rubrics. What scripts enforce.
5. ops/ — running it: model policy, deployment mechanics, runbooks.

Where documents conflict, the lower number wins. The prototype blueprint (v1, seven documents, archived) is superseded by this tree; it remains evidence of what a running instance looks like, not doctrine.

3. The core operation

Everything in the system — building the organization and every job inside it — is one repeated operation:

GENERATE → ORIENT. Something capable proposes candidates; something deliberately simple, strict, and dumber than the proposer judges each against a binary standard. The judge's simplicity is the system's most important safety property: a clever judge can be argued into a sophisticated mistake; a counter cannot.

From this operation, the laws (stated fully in corpus/LAWS.md):

• Matched elimination — every generation step is immediately followed by an elimination step; surviving is "could not be ruled out," never "looked good."
• Three stoppers — depth ends where a question stops forking; breadth ends when candidates die repeatedly for the same reason; a work-front ends when more questions stop moving the answer. Missions end only when ML ends them.
• The commit boundary — exploration is unbounded, sandboxed, disposable; nothing becomes permanent state (a page, a worker, a directive, a standard) without corroboration counted by a dumb gate.
• Evidence scales with consequence — one dial replaces per-gate threshold tuning: the corroboration a verdict needs is proportional to how much it forks downstream. The gate stays dumb everywhere; only its threshold moves.
• Progressive commitment — young artifacts and young teams are cheap to revise; heavily-corroborated ones are overturned only by overwhelming counter-evidence.
• Teardown ≠ reconstruction — the mode that eliminates and the mode that composes are staffed separately; neither contaminates the other.
• Two planes — the org chart churns freely and hardens with maturity; the knowledge artifacts change slowly and additively. Structure roils atop stable knowledge.
• Spartan design, not token thrift — efficiency means zero duplication, zero dead wiring, zero redundant workers. It never means fewer tokens. The system is built to crush compute; the constraint is clean design.

4. The anatomy (five layers + apex)

Full class blueprint in corpus/BLUEPRINT.md (~82 classes — treated as strongly-worded advice under validation, not gospel; the running system's corrections are defects to file against the blueprint, and vice versa).

Meta staffing: design target 2:1 meta:operating headcount, fluid upward to 3–4:1 — whatever full loop closure costs (ML directive: implement the philosophies entirely to spec). Cost stays operating-dominant regardless: four of the six closure functions are scripts or cheap-model arithmetic over the ledger.

The human (ML): four roles — Seed (intent + values), Tiebreaker (only for deadlocks that survived the whole escalation ladder), Keyholder (access granted, never derived), Final Lever (force/block any commit; decommission). Four contact categories, enforced by a gate; everything else is ledgered and pullable, never pushed.

5. The substrate (build order — leverage-first)

1. The Ledger — one append-only record of every verdict and outcome. Workers communicate by verdicts, never shared memory. Everything meta is arithmetic over it.
2. Typed outcome records — post-publication ground truth (artifact id + observable event + timestamp + source), attested before entering the ledger. Without these, "improvement" is unmeasurable churn.
3. Meters + sentinels per class — quality series, drift alarms vs commit-time baselines.
4. The Trial-Verifier path — no directive/standard/model-assignment change commits without a pre-registered champion/challenger win on recorded outcomes.
5. Self-knowledge sweep — census, wiring, loop-closure audit, canaries.
6. The conserving funnel — condenser tiers + conservation audit + the four-section Executive Brief (WORKING / NOT WORKING / CHANGED / NEEDS A DECISION) replacing the v1 Commander Report.

6. Worker doctrine

• The worker is the durable unit — a skill reference + one job + its own earned reference + wired channels. Many workers per skill, one per job. Baseline knowledge lives once in shared artifacts and is referenced; job know-how is owned by the worker.
• Commissioning is integration, not instantiation — before any worker exists: reuse-check (never duplicate a job), fit diagnosis, supporting-cast plan (its pod), and all four channels wired (broadcast-up captured, listen-up, broadcast-down, listen-down; leaves close their down-channels — undefined is where information leaks). An unwired worker is not commissioned.
• Sideways is up-then-down — no peer-to-peer channels; coordination routes through the lowest common ancestor, keeping every hop on the auditable trail.
• Retirement (apoptosis) — workers carry declared death conditions, audited dumbly. Removing capacity demands stronger evidence than adding it. Silent-failure detectors are never retired on silence.

7. Model policy (ML, locked)

• Command center / strategic layer: Claude Fable (via OpenRouter). Planning, orchestration, doctrine, high-consequence decisions, the apex duties.
• All workers: GLM 5.2 and DeepSeek V4 Flash, or similar class/price. **GLM is the hard ceiling below the strategic layer.**
• Assignment is priced by consequence-per-token, never context volume. Orienters run as dumb as possible (scripts where possible, DeepSeek-tier otherwise). Assignments are commits: trialed, demoted when a cheaper class matches outcomes, promoted only on metered evidence of under-power.
• Consequence above GLM's envelope does not rent a bigger model — it climbs the escalation ladder to the command center.

8. Mission posture

Deliver the obvious answer first; the mission continues past every answer. The standing workforce deepens weak commitments, defends published claims against decay, expands into adjacent ground, and anticipates what ML and the audience will need next — freely in the sandbox, conservatively at the three real surfaces (commits, the world, ML's attention). Missions end by ML's decommission, not by completion.

9. Map of system/

system/
├── VALUES.md            The boundary (constitutional, ML-only)
├── SYSTEM.md            This document — the root
├── FLAGS-FOR-ML.md      High-impact review queue
├── corpus/              Domain-independent craft
│   ├── LAWS.md          The operation + laws, stated fully
│   ├── BLUEPRINT.md     The class anatomy (five layers + apex)
│   ├── LOOP-CLOSURE.md  The six functions, the pod rule, the ratio
│   ├── FUNNEL.md        Condensers, conservation, the Executive Brief, escalation
│   └── COMMISSIONING.md Worker lifecycle: integrate, wire, trial, retire
├── seo/                 The SEO differentiation
│   ├── PIPELINE.md      Content pipeline as operating stages; Voice 3D; gatekeeper
│   ├── RESEARCH.md      Expertise pyramid → grounding; persona tournament
│   ├── FLOOD.md         Strategic flood, core-first, valve theory
│   ├── SECURITY.md      Injection defense, provenance, quarantine, red team
│   ├── ONRAMP.md        Ramp, graduation, boot interview, takeover mode
│   └── CONTINUITY.md    Backup, restore drills, rollback tiers
├── specs/               Mechanical contracts
│   ├── LEDGER.md        Record types, append-only rules, outcome records
│   ├── CONSEQUENCE.md   The stakes rubric + the one dial
│   ├── GATES.md         Gate library spec, values-gate template, force-pass console
│   └── CHANNELS.md      The four-channel wiring contract, verdict formats
└── ops/
    ├── MODELS.md        The spectrum, assignment trials, cost posture
    └── DEPLOYMENT.md    Provisioning, cron mechanics, Discord surfaces, runbooks

The storefront website is rebuilt from this tree as the final task, after the system is complete. The system is the product; the site is its display case.