Understanding a decision

The Actions tab is your to-do list of decisions: what your numbers say is worth deciding, in plain terms. You look at your data in Analytics, check the robots in AI traffic, and decide here. Each card answers five questions, in this order: what to decide, why now, on what evidence, how sure we are, and how you'll know whether it worked.

#Why a decision rather than a chart

A chart tells you a robot came 3,000 times. It doesn't tell you whether that's a problem, what you can do about it, or how to check that your action achieved anything. A decision answers all four.

These cards don't come out at random: the same numbers always give the same list, in the same order. No AI makes up a fact, a ranking or a threshold. An AI can reword a piece of evidence to make it readable; it cannot invent one.

#What each number is worth

This is what stops the screen from lying to you. Every number on screen carries a label that says exactly what it's worth:

The most useful row is the last one. "Common Crawl sent you no visitors" is not a reproach: the company behind this robot has nothing — no search, no assistant — that could send any. So the number reads "doesn't apply", never "0".

#How sure we are

Every decision tells you how sure we are — confident, fairly confident, or not very sure — and why: is it genuinely this robot (its official address matches), a lot or a little traffic, a full period to compare against or nothing to compare against, a full count or only a partial one (in which case the totals are minimums).

One thing not to confuse: when we write "genuinely this robot", it means its address matches the ones its vendor publishes officially. It's not a tamper-proof ID. The address that receives these hits is open to anyone: someone copying a published range would get the same "yes, it's this robot".

#Taking action: Snorklee never touches your site

When a decision proposes an access rule for robots, Snorklee prepares a robots.txt block for you to paste on your own site. It doesn't put it live, doesn't push it anywhere, changes nothing on your site. The "I've done it" button is simply a statement: you tell us the change is in place, and we start measuring its effect.

The reminder that comes with every proposal: a robots.txt file is a "no entry" sign, not a lock. A robot can walk right past it — and Snorklee shows you which ones do.

#Checking that it worked

The moment you say you've made the change, Snorklee freezes a starting point: the number to watch (the one named on the card), its value that day, and how many days it was measured over. Without that marker, comparing 28 days against 7 would make the number fall for a simple reason of arithmetic, not because anything moved.

Once the measuring window has passed, we bring the starting point back to the same number of days as the current period, and we look at whether the gap is sharp enough to mean something — a small count moves on its own, and we allow for that. Five answers are possible:

• Improved — the gap goes the right way and is sharp enough to trust;
• Unchanged — the gap stays within what moves anyway;
• Worse — the gap goes the wrong way and is sharp enough;
• Too little data to conclude — there aren't enough hits for a difference to mean anything. A "−100%" on 12 hits is not a result;
• We can no longer measure it — the number to watch is gone, or the period asked for falls outside the 90 days we keep in detail.

The last two are not Snorklee failing. They're the answers that stop you believing in an effect nobody actually saw.

#What no decision will ever tell you

• A saving, a cost, a weight in bytes, a return on investment. We don't measure the weight of the pages served to robots: the Snorklee code runs in your visitors' browsers, never inside the robot, and it reports the hit before the response finishes. Until that is measured end to end, no screen will put a number on a saving.
• The path of one particular robot. Nothing links what a robot reads to a visit later on: by choice, we keep no identifier that could tie the two together. That kind of link is always shown as likely, never as certain.
• Agents driving a real browser. We can't spot them yet, so Snorklee answers "no way to know" rather than counting them as zero.

#The Clone watch: what if your site were copied?

The Snorklee code on your site can end up running somewhere that isn't yours: on a copy of your site that kept the code, or behind a middleman serving your pages from another domain — the classic shape of a phishing site. When a visitor opens such a page, their browser tells us, by itself, the real address where the page runs; Snorklee compares it with your domain and sets aside the known harmless cases (caches, translated pages, archives, your own declared domains).

Two levels of truth, never mixed up:

• report — requests say one of your pages opened somewhere else. The address that receives these reports is open to anyone: on its own, an alert can be faked. Low confidence.
• confirmed clone — we went to look at the suspect page ourselves and found the Snorklee code on it: that domain really serves a copy of your site. A dated fact, high confidence, and an email goes to the site's owners (once per domain).

The card offers you three things: copy the evidence file (to paste into a report to the domain's registrar, Google Safe Browsing or Phishing Initiative), declare "this is one of my domains" (preview, mirror — never an alert for that site again), and track the effect: after a takedown, requests from that domain should die out, and that is exactly what we measure.

The limits, said plainly: when we couldn't confirm, it doesn't mean the site is clean — phishing kits know how to hide from checkers. A clone that removed the Snorklee code stays invisible here. And the absence of an alert has never meant there is no clone. Snorklee blocks nothing and takes nothing down: it spots, it proves, and it gives you what you need to act.

The Clone watch tab in the dashboard shows the full list: every domain spotted — including the ones that will never alert (caches, translated pages, your previews) —, the domains you've recognised as yours (removable at any time) and the watch's status. You look there; you still decide in Actions.

When a clone is confirmed, the served page is kept as-is (dated HTML copy, SHA-256 fingerprint): downloadable from the card, it's the exhibit to attach to your report — and it stays valid even if the clone later disappears.