How it works / signal engine

From raw signal to a trend you can inspect.

What's Happening groups related source observations, scores the movement, and keeps the evidence trail attached. You can see what changed without taking a black box on trust.

Inspect the public feed

Signal path

Designed for a 10-minute ingestion cycle

source-linked

Hacker News

official feed

GitHub

official feed

Google Trends

official feed

01

Normalize

Common fields

02

Cluster

Related topics

03

Score

0–100

If the production data service is not connected, the feed says so. It does not swap in demo figures and call them live.

01

What counts as a signal?

One signal is one normalized observation from one source. A Hacker News story, a public repository, or a trending search query can each become a signal. It is not a unique person, impression, or verified fact on its own.

About 24,832

The prototype number was a fixture, not a live production total.

The early interface displayed 24,832 as a fixed sample count. The current product only shows records returned by its connected data service. When that service is unavailable, the interface shows an unavailable state instead of presenting the fixture as current activity.

02

Where does the feed look?

The default ingestion uses official, credential-free endpoints. Optional Reddit, X, Tavily, and Exa adapters stay off unless their provider credentials are configured.

Hacker News

Official Firebase API

Story title, link, score, comments, and time

GitHub

Official REST Search API

Repository details, link, stars, forks, and time

Google Trends

Official trending RSS

Query, approximate traffic, feed country, and time

03

How is country context assigned?

Country context comes from the earliest observation in a cluster that carries usable geographic metadata. Today, that usually means a feed-level country code such as the selected Google Trends market.

What it can say

"The earliest country-attributed evidence in this cluster appeared in the United States feed."

What it cannot say

It does not prove who invented a topic, where an event truly began, or which country caused attention to move.

04

What makes a breakout candidate?

Related signals receive a score from 0 to 100. Candidates below the Global Pulse threshold stay on the watchlist. A score above 80 moves a trend into Global Pulse.

This is ranking logic, not a forecast of future popularity. A strong candidate can still fade as quickly as it appeared.

Velocity

How quickly engagement is arriving, with recent observations weighted into the score.

45%

Reach

Total engagement, estimated audience where available, and the number of distinct sources.

35%

Novelty

How varied the topic language is and whether the evidence spans more than one source.

20%

The useful boundary

Use the score to decide where to look. Use the linked sources to decide what to believe.

Return to the source-linked feed