Methodology

How a record gets in

Each source has an adapter that knows its shape, its cadence and its licence. Adapters run on a schedule — most every five minutes, some hourly, a few daily where the provider publishes no faster. Every run is logged with what it fetched and whether it succeeded, and the Terminal shows which feeds are currently reporting.

A record is admitted with, at minimum: a publisher, a timestamp for when the source published, a separate timestamp for when we first saw it, and wherever possible a location. Those two timestamps are kept apart on purpose — several sources publish hours to months after the fact, and conflating them would make a stale feed look live.

Evidence tiers

Not all records deserve equal weight, so they are not given it. The tier is assigned from the kind of publisher, not from the content, and it is visible on every record and filterable everywhere.

Tier	Meaning	Example
Measured	Direct observation by an instrument	Satellite radar, vessel telemetry, network measurement
Official	Government or intergovernmental publication	Advisories, official statistics
Research	Peer-reviewed or academically coded	Coded conflict datasets
Modelled	Model output, labelled as such	Derived indices
Reported	Single machine-coded press report	Wire and news extraction

The bottom tier is discounted heavily rather than excluded. A wire report is evidence — and pretending otherwise would leave the index blind on the first day of a crisis, when press is all there is. But one feed can emit tens of thousands a day and a corroborated source cannot, so a machine-coded press report counts for roughly a fifth of what a coded dataset counts for.

The same event, reported five times

One earthquake reaches us from a sensor network, two wire services and a national agency. Counting it four times would make the world look busier than it is, so records are collapsed by three independent rules:

1. Identity. The same source reporting the same event twice.
2. Near-duplicate text. Different publishers, same story, matched on content.
3. Proximity. Different sources describing one physical event at nearly the same place and time — the rule that catches a quake reported by four bodies with four IDs.

Collapsing keeps the strongest-tier record as the representative and retains the others as corroboration, which is why an event can show "+3 similar reports". Nothing is deleted; the duplicates remain reachable.

How the index is computed

Global pressure is a 0–100 reading of how loud the world is today. Three deliberate choices shape it:

Evidence weighting

Each record contributes in proportion to its tier, not as a tally mark. This is the half that makes the number worth anything — without it, the index measures how much the press wrote.

Diminishing returns per domain

Each domain contributes on a square-root curve, so the index responds to the breadth of stress across domains rather than to the raw volume in any one of them. Ten domains mildly stressed reads higher than one domain screaming.

Saturating, not clamped

The raw score is mapped through raw / (raw + K) rather than clipped at 100. A linear index pinned at maximum every day carries no information; the hyperbola stays responsive in the middle band and reserves the 90s for a genuinely extreme day.

Finding stress before it is news

Every tracked series is compared against its own trailing baseline. A reading that diverges far enough, for long enough, and on enough underlying observations becomes a signal.

Two gates keep the count honest. A significance gate requires the divergence to be large relative to that series' own normal variation. A triviality floor discards moves that are statistically striking but practically meaningless — a route going from two transits to four is a doubling and is not news.

Signals are kept separate from confirmed events everywhere in the interface, because most of them amount to nothing and mixing the two would quietly convert speculation into record.

How any of this is verified

The backtest asks one question: did signals actually precede confirmed events more often than chance? It is answered by replaying history point-in-time — reconstructing what would have been visible on each past date, using only what was known then — and scoring the result against independently sourced ground truth.

Two null baselines are published alongside every result: a random ordering, and a "big things move most" model. Beating random says the ordering carries information; beating size says it carries something beyond which entities are simply volatile.