Stop GA4 crediting sign-ups to login and payment pages, and count each sign-up once

My GA4 property is [GA4 property] and the site loads GA4 through GTM container [GTM container]. People sign in with [sign-in methods, e.g. Google, Meta, Microsoft] and pay through [payment provider]. Use the last 90 days. If the property holds less history than that, tell me how many days it actually has before you compare anything. 1. Run a GA4 report by session source, and a second one by landing page split by channel, for the same period. List every source that is really a login, consent or payment step: for example accounts.google.com, login.microsoftonline.com, the payment provider's domain, or facebook.com and linkedin.com when the landing page is an app, dashboard or callback page. For each one show sessions, users and key events. 2. Run a GA4 report by event name. For each key event, divide the event count by the number of users who triggered it. A sign-up that should happen once per person must stay close to 1. Flag every key event above 1.2 and say how many extra events that means. 3. Read the GTM tags, triggers and variables. Find the tag that sends the sign-up key event, the trigger that fires it, and any guard that stops it firing twice. Say where that guard lives: only in the current tab (sessionStorage), in the browser (cookie or localStorage), or on the server. 4. Check whether the GA4 Google tag already sends ignore_referrer for login and payment returns. Deliver: 1. a table of the false sources from step 1 with the key events they receive today, 2. the over-counting table from step 2, 3. a short diagnosis that keeps the two problems apart: wrong source and repeated counting, 4. the fix. For the wrong source, propose a Custom JavaScript variable that returns 'true' only when the referrer is a login or payment domain, or a social network landing on an app or callback path, and add it to the GA4 Google tag as the ignore_referrer parameter. Show the full JavaScript and test it in your answer against at least six referrer and path pairs, including an ordinary social media click that must stay a real referral. For repeated counting, say exactly where a once-per-person guard should live and what it should check. Do not change GTM or GA4 until I approve. After my approval, create the variable and update the tag with confirm=false first, show me the preview, then publish with a clear version name and tell me which version to roll back to if something looks wrong. Also remind me that GA4's own 'unwanted referrals' list can only be edited in the GA4 interface.