FoxTrust pre-alpha

How it works

FoxTrust gives every address a risk from 0 to 100, a level and the reasons behind them. This page explains how the reasons are found and how they add up. The numbers on it come from the scoring configuration 2026-11-12.1, the one this site runs with.

Two kinds of signal

Network categories say what kind of network an address belongs to: a hosting provider, a public cloud, a commercial VPN, a Tor exit relay, or address space that should not appear on the internet at all (a bogon). A category describes the network, not the people using it.

Observed behaviour says what the address was seen doing, for example SSH or login brute force, or belonging to a netblock hijacked by criminals. Behaviour is evidence about the address itself.

Some addresses are shared by many unrelated people, such as the addresses of Private Relay services. They carry the category private_relay, which adds no risk, and behaviour seen on them counts for half, so one person's abuse does not mark everyone behind the address.

Search crawler addresses carry the category search_crawler: those in the ranges that search engines publish for their crawlers, and those FoxTrust confirms by reverse DNS for a search engine without such a list. It adds no risk either; it lets a policy allow search crawlers before any other rule, because a crawler cannot solve a challenge.

Addresses that AI companies publish for their bots carry ai_training, ai_search or ai_user, by what the company says the bot is for. They add no risk and have no default rule; a policy decides what to do with them. The same holds for ads_crawler, a search engine's crawlers for ads, APIs and abuse scanning.

FoxTrust keeps the two apart all the way through: they are stored, scored and shown as different kinds of reason. Mixing "what it is" with "what it did" is the most common source of false alarms in IP reputation.

Decay

Categories do not fade: an address stays a Tor exit for as long as it is listed as one.

Behaviour fades with time. Each kind of behaviour has a half-life: after that many hours without a new sighting, it counts half as much. For SSH brute force the half-life is 72 hours, so a sighting from 72 hours ago counts half, and one from twice as long ago counts a quarter. An address that stops misbehaving drops out of the high levels on its own.

How reasons combine

Each reason has a weight (how strong that kind of evidence is), a confidence (how much the observation can be trusted) and, for behaviour, its decay. The reasons combine like independent pieces of evidence:

risk = 1 − Π(1 − weight · confidence · decay)

Two weak reasons give more risk than one, but the total never passes 100, and no single reason can push it there alone. The verdict then splits the risk back into the reasons, so each reason shows how much of the total it accounts for; the contributions add up to the risk.

Categories alone are never high

The part of the risk that comes from categories is capped at 69. A hosting or Tor address with no observed behaviour can be medium at most: being on a type of network is not misbehaviour. Only behaviour can take an address to high.

Levels

LevelRisk
lowbelow 30
mediumfrom 30
highfrom 70

What happens at each level is your decision, written as a policy: allow, challenge or block. FoxTrust scores; it does not block anyone by itself.

What a verdict shows

Every reason in a verdict has three fields: its code (what the address is or did), when it was last seen, and its contribution to the risk.

{
  "risk": 73,
  "level": "high",
  "categories": ["tor"],
  "reasons": [
    { "code": "ssh_bruteforce", "lastSeen": "2026-10-11T07:50:00Z", "contribution": 48.8 },
    { "code": "tor_exit", "lastSeen": "2026-10-11T10:50:00Z", "contribution": 24.1 }
  ]
}

A verdict never says who reported an observation, and never names the network range it matched. Those stay with FoxTrust for reviewing disputes.

Accuracy

Every release is checked against a list of reference addresses that must never be blocked: public resolvers, root servers, package mirrors and CDNs. A release that rates more of them medium or high than the one before is held until the change is explained.

If a verdict is wrong

If FoxTrust lists an address you control and the listing is wrong, dispute it. Disputes are answered within 5 working days, and a removed listing disappears in the next hourly update.