Data collection without code

Settle this before you plan a job. A value read on page forty is there for step forty-one, and once the run ends it is a line in a log rather than a row in a file.

Screenshots are different. Those are real files in a folder only your own account can read.

The builder holds sixteen kinds of step and four of them read rather than act. Each runs inside the profile's own browser, with that profile's fingerprint, cookies and proxy.

A fifth keeps evidence instead of text. The screenshot step saves the page as it looked.

• Extract text

  Takes the first element a selector matches and keeps its text, trimmed and capped at 4000 characters.

• For each element

  Runs its body once per matched element, ten by default and up to a thousand, re-reading the list each pass.

• Run JavaScript

  Runs your own code in the page and keeps what it returns, under a sixty second limit.

• Wait for request

  Holds until the page asks for an address you name, so a late list is read after it lands.

A broken selector fails loudly and is fixed in a minute. What ends a job quietly is the shape of the traffic: one browser, one cookie jar and one address reading a thousand pages an hour.

A profile per source and a proxy per profile is the part this product is for, and no selector fixes it.

The addresses are yours to bring. ARMANOS does not sell proxies and runs no exit network of its own: the store is a directory of 8 providers, and an order there is only a recorded request with no payment.

A schedule belongs to a flow and names the profiles, the time and how many run at once. Daily, on chosen weekdays, or every 1 to 24 hours, in your own local time.

It fires only while ARMANOS is open. A slot more than thirty minutes late is walked forward and counted as missed, and missed slots are never made up.

That last part is deliberate. Thirty profiles waking at once at an hour they never work is the burst a site reads as a robot.

The honest answer about export is that collecting extensions do it and we do not. The catalog inside the app carries two, both point and click, both writing CSV themselves.

They install into the profile you pick, so the collection runs behind that profile's fingerprint and address.

They are other people's work. We keep their store identity so a page sees the extension it expects.

Rules come before technique. Terms of service, database rights and personal data law decide the question before any step kind gets to it.

One platform names it outright. LinkedIn asks for one account per person under a real name and acts on automated collection of data from it, so we say that plainly and sell nothing that gets around it.

Where a site carries no such rule, the problem is an engineering one again.

Nothing in the product answers a challenge and nothing connects to a service that would. A challenge is yours to answer in the window.

A run does not stop at one either. The step that found nothing is written down as failed, and the steps after it keep going against a page still asking.

A profile launched with no window has nowhere to show the question, so work that can raise one belongs in a launch you can see.