How Risk Radar decides what counts as risk

Most tools that promise to track risk end up producing a number: a score out of 100, a colour, a dial pointing at "elevated". The number feels authoritative and is usually invented, a weighting someone chose that no reader can inspect. Risk Radar deliberately does not produce one, and this is what it does instead.

What Risk Radar actually is

Risk Radar takes the same live headlines as the main news desk and sorts them into four standing views, called lenses: regulation, counterparty, stablecoins and protocol. Each lens shows the four most recent matching headlines and one number, a count of how many matching stories landed in the last 48 hours. That figure shows the volume of activity, not the level of risk. It tells us how much is happening, but not whether the activity is dangerous.

How a story gets into a lens

Two different mechanisms, chosen per lens according to what the news categoriser can already do reliably. The regulation and stablecoins lenses select on the category tags the categoriser assigns when a story is ingested. The counterparty and protocol lenses work differently, matching against the words in a story's title and summary, since those two areas have no single clean category tag but do have a consistent vocabulary.

The counterparty lens screens for the language of exchange and custodian stress: reserves and proof-of-reserves, withdrawals, redemptions, solvency and insolvency, delistings, freezes, halts, suspensions and liquidations. The protocol lens screens for on-chain failure language: exploits, hacks, bridges, vulnerabilities, smart contracts, validators, drained funds, rug pulls, 51% attacks, forks, upgrades, outages and audits. A story mentioning a frozen withdrawal surfaces in the first; one about a drained contract surfaces in the second.

Why keyword screens rather than an AI classifier

Because you can check a keyword screen and you cannot check a model's judgement. The lists above set out the actual rules, and understanding them shows exactly why a story appeared, just as importantly why one did not. A screen like this has real limits: it will catch a story that merely mentions a hack in passing, and it will miss one that describes an exploit without using any of those words. That is a genuine trade-off, and it is the one Northtape chose knowingly, since a transparent rule that occasionally misfires is easier to reason about than an opaque one that quietly decides what you see.

Why there is no risk score

Because any single risk number requires choosing how much each input matters, and those weightings would be arbitrary. A score of 72 implies a precision that does not exist behind it. Counting stories in a 48-hour window makes a narrower claim that is actually true: this many matching stories were published. You draw the conclusion, and the headlines that produced the count sit right there for you to read.

FAQs

Does a high 48-hour count mean something bad is about to happen? No. It means more stories matching that lens were published recently, nothing more. It is a measure of news activity, not a forecast, and Northtape makes no prediction from it.

Why are there four lenses and not more? Each lens needs a selection rule that holds up: either a category the news desk already tags reliably, or a distinct and consistent vocabulary. Four areas currently meet that bar.

Can a story appear in more than one lens? Yes. A regulator ordering an exchange to halt withdrawals matches both regulation and counterparty, and it appears in each, since the lenses are independent views of the same feed rather than a single sorting of it.

Does Risk Radar tell me what to do about a story? No. It surfaces and counts headlines, with each linked to its source so you can read the original. It never recommends an action, and nothing in it is advice.

None of this is investment advice. Risk Radar reports what has been published, not what will happen next, and a quiet lens is not an assurance that nothing is wrong in that area, only that few matching stories have appeared recently.