Where the network stands

Part one — in plain English

What we have, and what it is doing

We own 426 websites. 263 of them are hooked up to the marketing engine, which means the engine has read the site, understands roughly what the business does, and can write for it.

Here is the honest picture, stage by stage. Each row is a smaller group than the one above it.

Stage	Sites	What it means
Websites we own	426	every property in the network
Set up in the engine	263	audited, with a file of its own
Having work written for them	253	the engine has drafted something
Matched to an industry	165	so the writing sounds like the trade
Switched on by an owner	85	someone said yes to it running
Able to tell if the phone rang	92	a tracked form or number is fitted
Actually published	90	the work reached the live website

253 → 90

Sites the engine has written work for, versus sites where that work actually reached the website.

That gap is the whole story. The machine is built, it runs every day, and it produces real work. Almost none of that work has been put on a website yet. It has been going into a queue that nobody was emptying.

The three questions Tim asked

We scored all 263 sites against them. Out of 99:

Does it work from end to end? — 56 out of 99. Most sites are set up and producing. What they are missing is the last step, publishing, and in some cases an account we do not have.
Did it make the phone ring? — 0. Not a low score. Zero. Across every one of the 263 sites, there is not one recorded phone tap, not one form filled in, not one campaign run. We have measurement fitted on 92 sites, so if it happens we will see it. It has not happened.
What still needs a person? — 3. Only 90 sites have ever had the engine publish something without a human doing it.

Why the second number is zero, plainly. Not because the engine is broken. Because nobody has spent a penny or sent a single message yet. We have built the shop and never opened the door. Nothing can ring a phone that was never pointed at a customer.

What we decided about the magnet sites

We have eleven websites about magnets. Every one of them explains what a coded magnet is. None of them sells anything, and none of them mentions any manufacturer other than Polymagnet.

The decision: turn them into distributor storefronts covering all brands, not just one. The hub becomes wholemagnetics.com with the full catalogue; the others each get a job feeding it, rather than eleven copies of the same thing.

We already had the product data: 220 real products with real prices, organised by what the magnet does — attach, align, twist, latch, spring. That is the right way to sell to an engineer, because an engineer arrives knowing what they need the magnet to do, not what size it is. Those became 103 product pages, with the different sizes listed on each page instead of scattered across separate ones.

What we need from Paul before this earns anything

An Amazon Associates account, and its tag. This is the one thing holding up real money. To earn a commission on a product, we need to be signed up to somebody's affiliate programme. We checked what we already have: 37 active partnerships, and only one is relevant to any of this — RVezy, for the RV park sites. There is nothing for magnets, plumbing, septic or building. Amazon is the one programme that certainly covers magnet products. Paul has to open the account; the assistant does not create accounts or handle payment details. Until a real programme exists, every product on the site offers an enquiry form rather than a buy button — because a shop that takes an order we cannot fill is the one mistake worth designing against.

What comes after magnets

Six more industries, in this order: real estate, septic, plumbing, construction, new home building, RV parks. One warning worth giving now: four of those six are services, not products. Nobody earns a commission on a drain field. Septic, plumbing, construction and new home building cannot be shops — they get enquiry forms and outreach instead. Real estate and RV parks are the two that can genuinely be shops today, because those are the two where we already belong to a programme.

Part two — in full detail

The arithmetic, and what was verified

The scoring, and why it is auditable

Every site is scored 0–99 on five axes, and the formula for each is printed on the worksheet itself. A ranking nobody can check is an opinion with a number attached, which is the thing this whole exercise exists to avoid.

Q1, end to end — in the engine 12, department on 15, knows its industry 18, work produced 15, something published 20, measurement fitted 19.
Q2, phone rang — a tap on a tracked number 40, a form completed 40, a campaign ever run 19. Deliberately harsh: no evidence scores zero, and it currently does on 263 of 263 sites.
Q3, unaided — shipped ÷ (shipped + items awaiting a human), × 99.
POT, potential — contactable 26, enough words to rank 22, industry known 22, more than five pages 15, sitemap 14.
PRO, profit — the page asks for a transaction 40, a real business behind it 25, contactable 20, catalogue depth 14.

Network averages: Q1 56, Q2 0, Q3 3, POT 72, PRO 56. Sites with a non-zero Q2: 2. With a non-zero Q3: 90. Highest combined potential and profit: buildercamp.com, hulloships.com, ofsthai.com, royallswindmill.com, smallhomevillage.com, magnetpoly.com.

The funnel, and one thing that was wrong

The stages are ordered strictly by descending count, and the build asserts that. The first version listed them in workflow order and the chart widened twice — "producing work" is not a subset of "activated", because the engine writes for a site whether or not an owner switched the department on, and "measured" is not a subset of "published". Every number in it was correct and it still read as a mistake.

The taxonomy

78 knowledge bases are built; 19 are in use, leaving 59 waiting for a site in that industry. 98 sites are in the engine with no industry matched to them at all — that is the largest single group and the clearest backlog.

Knowledge base	Covers	Sites
Realty	Estate agents, listings, property marketing	32
Reach	Marketing and agency work — the engine's own trade	26
Homebuild	Homebuilders and new construction	14
Ecommerce	Online shops and product catalogues	13
Events	Conferences, festivals, ticketing and venues	13
Media	Publishers, film, news and streaming	13

The catalogue work, and two things it caught

Source is a genuine Shopify export, magnetpoly-import.csv, 223 product handles.

185 of 223 had a bare part number where the name should be. The plan said drop them. The name turned out to be in the body copy in one of a few recognisable shapes, with the behaviour in the tags column, so it became a recovery job: 220 recovered, 3 genuinely unnameable. Deleting would have destroyed five sixths of the catalogue.
Eight products came back named "Alignment Polymagnet" with identical descriptions, differing only in price, weight and which SKU they pair with. Eight pages would have been eight near-duplicates. Grouped into 103 families with variant tables; the largest is Conventional Polymagnet with 29 variants at $2.00–$19.42.
Five products priced at $999.99, which is a "call us" sentinel rather than a price. Flagged quote-only; they route to the enquiry form. The $4,999 Polyvision is real — corroborated independently by multipolemag.com's own catalogue page, which lists it at $4,999/y.

The affiliate position, stated exactly

Impact's API returns only joined campaigns; the marketplace browse endpoint refuses the read-only credentials in /root/.secrets/impact.env. Of 37 active partnerships, RVezy is the only one touching these verticals and SmartMove the only real-estate-adjacent one. No Amazon Associates tag exists anywhere under /var/www. Which magnet manufacturers run their own programmes is unverified — the session's web search budget was exhausted, and guessing would be worse than saying so.

Where the code lives

/opt/autoengine/gen_wr_alldomains.py → the scored worksheet, and wr-scores.json, which this page reads so the two cannot disagree
/opt/autoengine/gen_wr_taxonomy.py → the funnel and taxonomy block on the front page, injected between wt:taxonomy markers and idempotent on re-run
/opt/wt-storefront/catalog.py → products.json; the storefront service sits beside the engine and never touches server.py, which fronts ~190 sites

Verified rather than assumed

Worksheet and front page return 200 from outside the droplet.
The taxonomy generator was run twice; the second run replaced its block rather than stacking a copy, and the marker count confirms one pair.
Funnel order asserted monotonic in the build.
Every new huddle item was confirmed present on the rendered page, not merely in the ledger.
The front page was backed up before injection.