Process · Content and distribution

How the audience gets built

The second engine. Most of what a product needs after it works is a machine that finds the people it is for, and that machine is the same loop as the product one, pointed at words instead of software.

Four movements, and the same shape as the product loop. The expensive part here is the writing rather than the code, so the whole system is arranged so that the writing happens once.

Select a step to see how it works.

Pay for the writing once

The main path
Context, feedback, or a rule
A shared store
What the loop produces
The step described below

Selected step

Search demand

What it is

What people are actually looking for, used as the input that decides what gets written next.

How I run it

The pages are built on the queries that exist, like a trade in a state or a figure's birth chart, rather than on what I find interesting. The engine writes toward demand, and the registry means it only writes each thing once.

Read the full section ↓

Pay for the writing once

Deterministic sourcefederal wage data, or chart math computed locally: reproducible, never invented
What it is
A body of facts a machine can compute or look up the same way every time, rather than a model's recollection of them.
How I run it
On RoadHand it is federal data the government already publishes: BLS wages, OSHA enforcement, Davis-Bacon prevailing wages, GSA per diem rates, compiled into something a worker can use. On AstroSense it is the chart itself, computed locally with real astronomy, so the same birth data always produces the same chart.
Read the full section ↓
Groundingevery piece written against the exact computed facts
What it is
Writing generated against retrieved facts rather than from the model's own memory, so the output can be checked against something.
How I run it
Every piece of content is written against the exact computed placements or the exact wage numbers. That is the difference between a content engine and a content mill: the second pipes a prompt into a model and prints whatever comes back, and it reads like it.
Read the full section ↓
Generate oncen8n and the Claude API, on workflows a person signed off
What it is
Authoring treated as a capital cost rather than a per-request one: each piece is written once, stored, and reused instead of being generated again on every request.
How I run it
n8n workflows and the Claude API do the writing; the result is stored and served read-only after that. The pages themselves are incrementally regenerated against the current dataset, so a revised federal wage figure reaches them without a rebuild. What is paid for once is the prose, not the numbers.
Read the full section ↓
Content registryPostgres; stops the same thing being written twice
What it is
A record of what has already been produced, so the pipeline can tell the difference between a gap and something it has already filled.
How I run it
A Postgres registry preventing duplicate generation. Without it an automated engine happily writes the same article twice, and the second one costs exactly as much as the first.
Read the full section ↓
Owned surfacesserver-rendered pages, the daily library, the free tools
What it is
Destinations you control and that accrue value over time, meaning your own pages, as against rented attention on somebody else's platform.
How I run it
A server-rendered wage page for every trade in every state on RoadHand, built on the federal wage and per-diem dataset. On AstroSense, the free tools and the celebrity chart library, which doubles as the SEO engine.
Read the full section ↓
Social distributiondaily carousels pushed to the channels on a schedule
What it is
Pushing what was published into the channels where people already are, on a schedule, rather than hoping they arrive.
How I run it
Daily Instagram and Facebook carousels, generated from the same grounded content and posted automatically. The post is a by-product of something that already exists, which is why the cadence is sustainable for one person.
Read the full section ↓
The funnelfree surface earns the visit; one paid object holds the depth
What it is
The path from a stranger's first visit to a paying customer, and the decision about how much to give away before asking.
How I run it
A lot of genuinely free surface area around a single paid object. The free tools earn the visit and the trust, a deterministic teaser proves the engine is real, and the paid report is the one place the depth lives.
Read the full section ↓
Paid conversiona one-time report, or a subscription
What it is
The point where attention becomes revenue: a purchase or a subscription.
How I run it
One paid Snapshot Report on AstroSense, bought once. Stripe subscriptions on RoadHand. What converts is the strongest signal available about which content was worth producing, so it goes back into the source.
Read the full section ↓

↺ What converted goes back to the source, and search demand decides what gets written next. It is a loop, not a line.

Underneath all of it

Search demand

What people are actually looking for, used as the input that decides what gets written next.

Read the full section ↓

One

Compute

Start from facts a machine can reproduce

The government already publishes most of what a skilled-trades worker needs to make a decision: wages, enforcement records, prevailing rates, per diem, and nobody had compiled it into something a worker could actually use. That gap is the product, and the dataset is the content engine's raw material.

On AstroSense the same principle applies to a different domain: the chart is computed locally with real astronomy, deterministically, so the same birth data always produces the same chart. Most apps in that category are a thin wrapper over a paid ephemeris API or a mill piping prompts into a model. Computing it yourself is what makes everything downstream cheap.

Ground the writing in the computed facts

Every piece of written content is generated against the exact numbers or the exact placements, never against the model's own recollection of them.

This is the whole difference between a content engine and a content mill. The mill is cheap and generic and reads like it. Grounding costs you an architecture and buys you output that is specific, checkable, and worth indexing.

Two

Generate

Pay for the writing once

Writing is the scarce resource, so the system is designed so that I pay for each piece once and reuse it. n8n workflows and the Claude API do the authoring; the result is stored and served read-only after that.

That is what lets a solo project publish a fresh horoscope for all twelve signs every day, a growing library of full birth-chart reports, and daily social carousels, with no per-request AI bill. Treat generation as a per-request cost instead and the same surface area is unaffordable.

The honest caveat sits inside the promise rather than four sections below it, and the answer is that "once" applies to two different things. A chart is fixed once the birth data is, so that writing happens once and never again. Federal wage figures are revised a few times a year, and a page built on a stale number is simply wrong.

So the split is deliberate. The data-driven pages are incrementally regenerated: they re-render against the current dataset on a revalidation interval, which means a BLS or Davis-Bacon revision reaches the page without anybody rebuilding anything. What is written once is the generated prose, and that is what the registry tracks. The expensive thing happens once; the cheap thing happens continuously.

Refreshes on its own: The numbers. When a wage figure is revised, RoadHand's page re-renders against the current dataset on its next revalidation, with no rebuild.
Does not refresh: The sentences around the numbers. A paragraph generated against the old figure stays as it was, and nothing flags it today: no check ties a piece of prose to the facts it was written from.
Who catches it: A person reading the page. A step that marks prose for rewriting when its source figure changes is the obvious next piece, and it is not built.

A registry, so nothing is written twice

A Postgres content registry records what already exists. An automated engine without one will happily generate the same article a second time, and the duplicate costs exactly what the original did while being worth less than nothing to a search engine.

Tools: n8n runs the authoring workflows on a schedule
Claude writes each piece through the API, against the computed facts
PostgreSQL holds the content registry
Output: Generated content, stored and reused; the figures inside it refresh separately
Check: The registry: a piece that already exists is not generated again
Example: RoadHand's wage pages. The prose for each trade in each state is written once, and the figures in it re-render from the current federal dataset.

Search demand decides what comes next

The pages are built against queries that already exist, like a trade in a state or a specific figure's chart, rather than against what I find interesting. The engine writes toward demand, and the registry makes sure it writes each thing once.

Three

Distribute

Owned surfaces first

A server-rendered wage page for every trade in every state on RoadHand, built on the federal wage and per-diem dataset and rendered for search. On AstroSense, the free tools and a growing public celebrity chart library that doubles as the SEO engine.

Owned surfaces compound and rented attention does not. A page indexed today is still working in a year; a post is not.

Then push it where people already are

Daily Instagram and Facebook carousels, generated from the same grounded content and posted on a schedule.

The important part is that the post is a by-product of something that already exists rather than a separate act of creation. That is the only reason a daily cadence is sustainable for one person.

A person signs off: The workflows and the prompts they run, the page templates, and anything that opens a new public channel or carries launch copy.
Not reviewed one by one: Individual pages and scheduled posts. They come out of workflows that were approved, rather than past a person who read each one, and the limits below say what that costs.
Gated one at a time: One-off public work, such as a first newsletter send or new launch copy. That is a work packet rather than a recurring workflow, and it waits for a person at a gate, as the operations engine describes.

Four

Convert

The funnel is the design

A lot of genuinely free surface area around a single paid object. The free tools earn the visit and the trust, the deterministic teaser proves the engine is real, and the paid report is the one place the depth lives.

That is a design decision before it is a marketing one. It determines what every screen is for.

What converts goes back into the source

One paid report bought once on AstroSense; Stripe subscriptions on RoadHand. Conversion is the strongest available signal about which content was worth producing, so it feeds back to the front of the loop rather than sitting in a dashboard.

Where this breaks

What it does not do

An engine that writes toward search demand writes what people are already asking for. It will never produce the piece nobody knew they wanted, and that piece is often the one that matters.
Volume is not distribution. Hundreds of indexed pages are worth nothing until something links to them, and the engine does not solve that part.
This is built for one person operating at leverage. Some of it, particularly generate-once, is a constraint I designed around rather than a universal recommendation.

Seeing it run

RoadHand is this engine on federal data: the ETL, a server-rendered wage page for every trade in every state, and the n8n content pipeline behind them. AstroSense is the same pattern on a deterministic chart engine, with the funnel and the daily social cadence attached.

The other two engines are how the work gets made, the same shape pointed at software, and how the team keeps its memory, which turns meetings, user research and survey findings into the record all three read.

Half an hour, no deck. Where your current loop leaks, which stage would pay back first, where automation would actually help, and what still needs a person to look at it before it ships.