Signal

Signal is the safety product. It answers one question in four ways: what do we know about this account, this server, this link, and what should happen here. The shared registry is the centre of it, and your protection level decides what your server does with what it says.

Scam detection, OCR and Join Guard are how Signal finds things out. They are instruments rather than the point: they read what is posted, read the text inside images, and check people at the door, and everything they catch becomes evidence in the same place. You do not tune them one by one; you pick a protection level and Signal decides what that means for each of them.

Whatever catches something, it ends up in the same record. A scam message, scam text read out of a screenshot, a report a member sends in, a ban list a server shares, a server somebody names, and a partner submission all become the same kind of evidence, counted once per independent source, and gathered onto one case for a reviewer. Five servers catching the same scammer within an hour is one piece of work with five pieces of evidence.

The threshold is limited to 20 up to 70. Below 20 the gate catches nobody the shield has not already handled; above 70 it starts catching ordinary members. The default follows your protection level: 30 on Basic, 45 on Standard and Custom, 55 on Strict, 65 on Lockdown.

A gated member is always told their score, the threshold and how to contest it.

Every server that runs Signal reports its high-confidence catches. When the same indicator turns up in at least 3 unrelated servers inside 45 minutes, Radar opens a wave: that is a campaign, not one local spammer.

Servers in the wave are warned and temporarily raised one step up the ladder. The boost never overwrites the level you chose; it is a time-boxed flag that both the bot and the dashboard resolve when they read your settings, and it rolls back on its own.

On a Custom server a boost does not replace your config either. It tightens the threshold and makes sure joins and messages are checked, and it never raises a threshold you already set lower.

A confirmed case becomes evidence in the Signal Registry, credited to the server that saw it rather than to Modora. Scores count one contribution per independent source, so ten catches in one server stay one source and three servers seeing the same account is what adds up.

Modora counts as one source however many places a ban lands. That is deliberate: a shared record in which the operator can clear everybody’s threshold on its own is the record that gets accused of running the ecosystem.

The Signal Registry is a shared dossier on Discord accounts. Signal already judges behaviour as it happens: it reads a message, scores it and acts. The registry answers the other question, the one a server owner has before anybody has posted anything, which is whether this account is trouble somewhere else.

One account, one row. It holds the labels the account carries, how strongly each one is carried, how many independent parties are behind it and the date it runs out. What that means inside your server is your decision, not ours.

A label is a sum of evidence and nothing else. Each kind is worth a fixed amount out of a maximum of 100: a staff decision 80, a detection 35, a report 30, a shared ban list 25, a community link 20.

The sum counts one contribution per independent source, and only the strongest piece from that source. Ten reports from one server is one server disagreeing with the account; two reports from two servers is a pattern. That cap is what keeps a single angry party from driving a listing.

Withdraw the evidence and the label drops with it, in the same step. No listing outlives the reason for it.

A reported server is checked by a person before it means anything. Once confirmed, it gets a rung: recorded and nothing more, a warning anyone asking about it will be told, or invites to it being removed in servers that asked for that. No rung removes a member of it. That runs through the account dossier, with the evidence and the appeal that belong to it.

We keep what a server has called itself, because renaming is the cheapest way around a listing, and every invite we have seen, because rotating codes is the second cheapest. Both come from what Discord tells anybody who hovers the link.

Where our bot is already inside a confirmed server, it records who is there and how far in. Being a member is associate: a log line, and no setting of ours can make it more than that. Holding the permissions to remove people is a claim about conduct, and it becomes evidence for whatever that server was confirmed as. It is still one source, so one bad server is never a case on its own.

Evidence has a lifetime: a staff decision 730 days, a detection or a report 365, a community link 270, a shared ban list 180. When the last piece behind a label runs out, the label stops counting.

An account nobody reports again is swept hourly, so a listing actually lapses instead of sitting in the table forever. Only a person can take a single listing out of the clock, and that is a deliberate staff decision on that one account.

The registry stays off in a server until you switch it on. Each label gets its own action: nothing, a log line, a role, a kick or a ban. Labels you have not configured fall back to a log line.

Two thresholds guard every action: the confidence a label needs, and the number of independent sources behind it. They start at 70 and 2, so nothing is taken away from anybody on one party's word. A label that misses either bar still produces a log line, with the reason it was held back, because that is the message that makes the thresholds worth having.

Every enforcement carries the route to contest it, and the route stays open. A network-level action (global ban, blacklist, network quarantine) can only be lifted by Modora, so those always start at modora.gg/signal/appeal, which shows what is on the account and hands over to a ticket in the Modora server.

A Modora appeal is a conversation, not a form. A listing turns on evidence somebody else supplied, and the questions a reviewer has to ask about it - which server, what did you post, do you still have the message - do not fit in one box that gets read once. So it is a ticket: a person answers, the member can answer back, and both sides keep the record. Anyone too restricted to join a server can mail support@modora.gg instead.

A ban that a single server issued goes to that server's own appeal form when one is configured; otherwise it falls back to the Signal page. Members can also use /appeal in the server itself if the Appeals module is on there.

Your own listing is part of the trust check for your account at /signal/check: whether you are listed, what for, how many independent sources are behind it and the month it runs out. Dates are month precision only, because a timestamp to the day would tell the server that reported exactly which of its reports this was.

An upheld appeal clears the account. Cleared blocks enforcement outright and is sticky: it survives new evidence, so the next report cannot quietly undo an appeal somebody won. Reopening a cleared account is a separate staff decision. A rejected appeal changes nothing in the registry.

Every endpoint below is keyed and scoped. A key is issued per partner and names exactly what it may reach, so a key for reading a record cannot file one. Keys and their scopes are managed on dev.modora.gg.

Send your Signal API key as X-API-Key. The endpoint needs the signal.trust_check scope, the same one /lookup uses, and sits on the same rate limit as the rest of the v1 surface.

listed is the verdict. confidence and source_count are the two numbers your thresholds are set against. listed_month and expires_month are month precision; expires_month is null when the listing does not run out on its own, which only happens when a reviewer decided that.