How it works

The information is not hidden. The derivation is never run.

A dated event lands somewhere first. Somebody, somewhere, has already lived through the version of it that is coming for you, and what happened to them is on the public record. Almost nobody goes and reads it, and nobody at all works forward from it three steps to the thing it opens. That is the whole job.

3events in the corpus
5precedents behind them
31recorded outcomes, each with a link
10openings published from them

The five things that happen

  1. Find the date, from the authorityA gazette, a tax office, an observatory. News can point us at an event but never establishes one, and an event with no authoritative source is not admitted.
  2. Find where it already happenedThe same structural event, in another country or another decade. Not the same topic: the same shape. A cliff-edge deadline in Ireland in 2018 is a precedent for a cliff-edge deadline in Spain in 2027, whatever the two are about.
  3. Extract what actually followed, with a linkThis is the load-bearing step and the one everything else rests on. What did prices do, who entered, who consolidated, what failed. 31 outcomes are recorded and every one carries the URL it came from. No link, we bin it.
  4. Attack the analogy before using itThe step most people skip. What has changed since the precedent, and which of its outcomes does that change invalidate. An analogy nobody attacked is a coincidence with a footnote.
  5. Derive forward, then rank and labelConsequences three or more causal steps out, ranked by how few people have noticed. Every one is published, strongest first, and each carries a label saying how far its evidence goes: backed by a recorded outcome, supported by a pattern only, or reasoning with nothing in the corpus behind it yet. 10 openings are published across the corpus and 2 of them are labelled as evidenced.

The rules we cannot bend

These are not principles on a page. They are constraints in the database, so breaking one is not a decision anybody here can quietly make.

No basis, no number

There is no way to store a bare figure. Every value carries how it was obtained, which direction its error runs in, and the exact sentence the page will print. A number with no basis cannot be written down, so it cannot reach a page.

Three states, never two

“Nobody measures this”, “the source was down today” and “we measured it and it is zero” are three different facts. Collapsing them into one is how a broken source becomes a finding. A source that did not answer is written as could not check, never as absent.

Who pays decides which sources apply

Not the vertical. Public bodies buy cyber services and do not place packaging on the market, so the buyer decides which of the 5 measurement sources are even admissible for an event.

Sources fail silently, so every source has a canary

Observed in the wild: a 200 response containing an HTML error page, an API returning “no results” for a missing token, and an API returning four values in a different order than they were requested. The last one gives a wrong answer rather than no answer, which is why nothing here matches a request list to a response list by position.

Seen once is labelled seen once

A pattern from one country is an instance with the names rubbed off. The page says how many places it was actually seen in, next to the pattern, every time. 10 patterns are held and each one names the precedents it came from.

Versions are pinned

A derivation is keyed on the registry and corpus version as well as the event. Without that the graph is cyclic across time and reproducibility is a claim rather than a fact.

What we get wrong

Being wrong is not the risk. Being wrong invisibly is. So the parts of this that can fail are the parts we make easiest to check.

Every opening carries the thing that would kill it, written on the day it was derived. Those sit in the ledger and resolve in public, the misses too.

Nothing is hidden, and nothing is flattered. Every opening we derive is published, ranked, and labelled with how far its evidence actually goes. A hunch is printed as a hunch rather than dropped so the page looks stronger.

Absences are printed, not omitted. If nobody measures something we say so on the page, because an unmeasured market is usually also an unserved one.

The corpus is young. 3 events, mostly regulatory. That is where we started, not where this stops, and the library shows the empty shelves rather than hiding them.

What you get, and what it costs

Everything above a country is free to read in full. Paying adds one thing: the same event worked out again inside a single country. That is the only gate and there is no second one.

Read this week’s analysisSee what costs money