Ga naar de inhoud

Deze pagina is nog niet in uw taal. U leest de Engelse tekst.

Risk signals

Three warnings from your own log

Your workspace already writes down who signed in and which profiles opened. Three rules read those rows and speak up when they take a shape that tends to come before a lost account.
3
Rules over your rows
30 days, newest 5000 entries
Window the rules read
4 and up
Profiles on one address
5 and up
Starts inside one minute

What the rules read

Nothing new is collected for this. Two of the rules read sign-ins and profile starts the log already holds, over the last thirty days.

The third reads today's profile list instead, because a cluster on one address is a fact about how profiles are set up right now rather than an event that happened.

  • Sign-ins

    Who signed in and when, with the address the server itself saw and the country, where a country is known at all.

  • Profile starts

    An opening that took the lock on the server. A window opened while the server was out of reach leaves no entry.

  • Today's profiles

    The name and the exit address of each profile that has not been deleted.

The History screen in the app: workspace rows with a date, an action and who did it
The rows the rules are arithmetic over, read in the app and in the dashboard.

A sign-in from a new country

Either the person travelled or the password is in somebody else's hands. Both are worth the same day rather than the week when profiles start vanishing.

The comparison is per person, not per team. One colleague living in Poland does not make an Argentine sign-in normal for another, and a first sign-in is never an alarm: a newcomer has no usual country yet.

The country comes from the edge

The server keeps no address-to-country table of its own. The country comes only from the header a trusted edge in front of it sets, and without such an edge the country stays unknown and this rule says nothing rather than guessing.

Four profiles behind one address

Platforms take a cluster of accounts on one address together, and they need no simultaneous logins to find it. Three profiles is still work; four on one exit is where the line sits.

The three heaviest clusters are named with profile names, so thirty identical lines never bury the one worth reading.

How the address got thereWhat the rule counts
Linked from the shared libraryOne cluster per library entry
Typed into the profile by handA cluster of its own: protocol, host, port and login make the key
One address saved twice in the libraryTwo clusters: here the library entry is the identity
No proxy on the profile at allNot this rule: those leave from your own address and get their own line
A profile in the recycle binNothing: warning about an address nobody opens teaches you to ignore the panel

Five openings inside one minute

Accounts signing in together read as one script behind all of them, whatever the names on them say. Five starts inside a minute is the count that raises the line.

The count alone is not enough, so the rule looks at density too. A pack is a burst only when the smallest gap between two neighbouring starts falls under two and a half seconds.

Five starts 300 ms apart
One line, naming the pack with the most starts in it
Ten starts 3 seconds apart
Silent: that spacing is the app's own queue
Ten starts 10 minutes apart
Silent: this is how a morning should look

The bulk launcher waits a random 3 to 14 seconds before every opening after the first, and the density threshold sits below that minimum on purpose. A warning seen every morning is a warning nobody reads by Friday.

Where a warning shows up

The panel sits at the top of your dashboard, and it is absent entirely when there is nothing to say.

A line cannot be dismissed. The cluster line goes when the profiles are spread apart; the other two are read out of the log itself and leave when their entries fall out of the window.

  • Most urgent first

    A possible stolen password stands above the housekeeping lines, and equally urgent lines stand newest first.

  • Admin and owner only

    Reading the log is their action. A manager, an operator or a member is refused, and the panel stays quiet rather than breaking.

  • Through the interface too

    One call returns the same lines to a program key allowed to read logs, and both toolkits carry it.

From a warning to the rows

A line tells you the shape. What actually happened is in the log itself, read three ways over one set of rows.

  1. 1

    Everything

    The whole workspace newest first, with the email of the person instead of an internal id.

  2. 2

    Openings

    Only the starts and stops of browser windows: the view behind a burst and behind who was working when.

  3. 3

    Sign-ins and addresses

    Sign-ins and machine registrations with the address, the country and the machine name.

What each line asks of you

Each of the three points at a different repair, and no browser setting does any of them for you.

  1. 1

    A new country

    Change the password, turn on the second sign-in step and look through the linked machines.

  2. 2

    A cluster on one address

    Spread those profiles over more addresses, or keep the cluster knowingly for accounts meant to sit together.

  3. 3

    A burst of openings

    Open through the queue or a schedule, so a morning arrives spread out instead of in a row.

What this does not do

  • It does not see the platform's side. These rows are your own workspace's, so nothing here can say an account is limited, under review or in a shadow ban.
  • It is three rules, not a model. Nothing is scored, predicted or learned, and the same rows read the same way every time.
  • It blocks nothing. A warning is a line to read: profiles keep opening and no launch is held back because of it.
  • The cluster rule counts only profiles with an address set. A profile with none leaves through your own connection, which is a separate line in the same panel.

How to check

The rules are arithmetic in one file, and the stand runs that file rather than reading it.

The thresholds, the window and the per-person country comparison
apps/server/src/logs/suspicion.ts · apps/server/test/подозрительная-активность.js
The spacing a bulk launch really uses, which the burst threshold sits below
apps/desktop/src/manager/renderer.js
The address taken from the server, the country only from a trusted edge
apps/server/src/common/request/client-net.ts
Reading these rules starts at admin
apps/server/src/logs/logs.controller.ts
The panel shows what the server counted and survives a refusal
apps/web/components/dashboard/AttentionPanel.tsx · apps/web/test/attention.js

Questions

Does a warning mean an account is already lost?
No. These are shapes that come before losses, read from your own rows. Whether a platform has acted is visible there and nowhere else.
Why is my morning batch never flagged?
Because the app spaces its own openings by three to fourteen seconds inside one window of the app, and a burst needs a gap under two and a half seconds. Two machines starting at once are not spaced by anything.
Why is the country column empty for me?
Because the country is only as good as the edge in front of the server. Without that header the field stays empty and the rule stays quiet.
Who can see the panel?
The panel itself shows up for anyone, but these three lines are read out of the log, which is open to the owner and an admin only. What they read covers the whole workspace rather than their own profiles.
Can I change the thresholds?
Not today. Four profiles, five starts and thirty days are constants in one file, identical in every workspace.

Read your own rows

Install ARMANOS, sign in, and the history fills from the first profile you open.