The door that opened on another product

A product is reachable when three separate things are true at once: an address that resolves over working HTTPS, a page that answers in a language the visitor reads, and a record that says which page produced the enquiry. Three different systems own them, they fail independently, and none of them is visible from the others.

Mechanicum 3D — the industrial 3D training product the group distributes — had the middle one.

What the address does

Measured again before publishing, from an ordinary machine with nothing special installed. The transcript is trimmed to the lines that matter:

$ curl -sSI https://mechanicum3d.com
curl: (60) SSL: certificate subject name 'test-self-signed.spaceweb.ru'
            does not match target hostname 'mechanicum3d.com'

$ openssl s_client -connect mechanicum3d.com:443 -servername mechanicum3d.com \
    | openssl x509 -noout -subject -issuer -dates
subject=C=RU, ST=SPb, L=SPb, O=Test Self SSL Company, OU=Test,
        CN=test-self-signed.spaceweb.ru, emailAddress=support@sweb.ru
issuer= C=RU, ST=SPb, L=SPb, O=Test Self SSL Company, OU=Test,
        CN=test-self-signed.spaceweb.ru, emailAddress=support@sweb.ru
notBefore=May 25 10:09:43 2016 GMT
notAfter =May 24 10:09:43 2016 GMT

$ curl -sSI http://mechanicum3d.com
HTTP/1.1 302 Moved Temporarily
Location: https://vardix.com/vargates-technics-app/

Issuer and subject are the same name: the certificate signs for itself, and it is the hosting provider's own placeholder, issued to the provider's test hostname rather than to this one. Its validity window closed the day before it opened — those dates are not a typo here, they are what the server presents. A browser does not negotiate with any of that; it stops. Over plain http the same address answers a redirect, and the redirect lands on the page of a different product — VARGATES Technics.

So the one address a distributor is most likely to type either refuses the connection or delivers them to something they did not ask for.

What was not broken

The item that opened this work said the product had no working page or form. Half of that was wrong, and the half matters. /mechanicum3d has been on this site since March, answers 200, and is in English; the product catalogue has listed it since the catalogue itself opened in May. Nothing had to be built from nothing.

What did not exist was a way for that page to produce a lead of its own.

A page that sent its visitors away

Three links on the page left the site for the product's own address at roqed.com — two in the hero, one in the closing card. That site works. But a visitor who leaves is not an enquiry anybody can answer, and the page that persuaded them is named nowhere afterwards.

There was one route to a form: a line of small print under the closing card offering to ask about licences and seats, pointing at the contacts page with this product preselected. That is real mail, and it is attributable — up to a point:

Only the last line tells the two routes apart. The subject prefix and the Interested in line belong to the enquiry register, not to any one form, so they say the same thing whichever page the visitor came from — the first two columns above are before and after in time, not one route against another. /contacts is written into the contacts form as a constant, because that is where that form lives: every enquiry made there arrives saying so, whichever page sent the visitor.

What changed

The page now carries the site's compact lead form, already standing on this product, with the first-screen call opening it in place instead of leaving. The secondary exit still goes to the product's own address, because someone who wants the product's own pages should get them — and that link is tagged, so a visitor who takes it is measured rather than lost.

Three files: the page, one line in the enquiry register, and a new test file — 164 lines added, 31 removed. The register line is the part that decides whether any of this can be counted:

-  { slug: "mechanicum3d", label: "Mechanicum3D", routedFor: INFO, subjectPrefix: "[Mechanicum3D]", intent: "product" },
+  { slug: "mechanicum3d", label: "Mechanicum 3D", routedFor: INFO, subjectPrefix: "[Mechanicum 3D form]", intent: "product" },

The same change deleted the closing card, and with it the small-print route to the contacts page. Before, that link was the one path producing an attributable enquiry for this product; now the page has a form of its own and no fallback behind it.

The six checks

The new spec does not test that the page looks right. It tests the things that would silently stop a lead from being attributable:

• no other of the 23 enquiry topics carries this label or this subject prefix — a topic added later that takes either of them turns the suite red;

• the mail a lead from this page will be filed by — built from the same register entry the page hands the form — carries all three lines above;

• the path answers 200, its <html lang> is English, its heading is in no Cyrillic or Kazakh letter, and no link in its main content points at the other product's page;

• the form is present and visible, the product is asked with a select rather than sent hidden and already stands on this product, the hidden source field carries this path, and the anchor the hero points at is the one holding the form;

• on a 375-pixel-wide screen the route to the form sits within the first screen, with nothing scrolled;

• the product catalogue still has a card pointing here.

The site's end-to-end suite ran green on the branch: 2 357 passing, 2 skipped. That is a receipt from one run on the branch worktree, not a measurement made for this material.

The one the form's design turns on

The product could have been sent as a hidden field. It is not, and the reason is written into the component: a form must never carry two controls with the same name. FormData.get returns the hidden one, and the visitor's own answer is the one discarded — silently, in the single place where a lead is recorded.

So the page asks which product, visibly, with this product already chosen; and the test asserts the select, not a hidden input. A check written against the hidden input would have passed on exactly the arrangement that loses the answer.

What none of this proves

Not one of the six checks submits the form. The register is checked, the mail is rendered from that register, the page is opened in a browser and the catalogue is walked — but nothing posts a lead and watches it arrive. The item's own acceptance criterion is that a submission creates an enquiry with this source; what shipped is everything up to the moment of sending, and the send itself is still covered only by the delivery path the rest of the site uses.

Two more edges are worth naming rather than discovering later. The rename applies to the register, so it is retroactive in effect: every enquiry already in the mailbox carries the old prefix, everything from the day of the change carries the new one, and any saved filter that matched the old one stops matching. And the product answers to three names across three surfaces — the catalogue card says ROQED Mechanicum, the enquiry register says Mechanicum 3D, the page's own heading says neither.

What this does not fix

The measure

Not the page. The item that asked for this named its own measure before any of it was written: ten deals from this form. A page that answers and a form that files correctly are preconditions — they make the question askable, and they answer nothing by themselves.

Which is also why the source lines had to change. The subject prefix says which product; Submitted from says which page asked. Without both, in a month there is no way to say whether the work paid for itself, and the honest report would have to be that we do not know.

The product: ROQED Mechanicum · its page on this site: /mechanicum3d.