Pipeline
How trend intelligence is produced
Six stages run continuously. Collection repeats every 2 hours. Nothing skips the public eligibility gate, and nothing is published to fill empty space.
Inputs
Three signal channels
Search, social, and marketplace — each contributing a different kind of attention.
01
Search signals
→ Google Trends
Public Trending Now interest by country. Traffic values are ordinal buckets, not exact search volume.
02
Social signals
→ YouTube
Most popular videos by country, with view and engagement counts as a video-attention signal — not proof of sales.
03
Marketplace signals
→ eBay Browse API
Listing snapshots including rank and persistence. Current collection is Singapore search results, not every marketplace.
Then
Entity Resolution → History → Score → Quality → Eligibility
01
Collect
Automated collectors capture public observations from Google Trends, YouTube, and eBay on a recurring cadence.
02
Resolve
Observations are linked to entities by exact or alias matching. Similar-looking names are not merged automatically.
03
Score
A deterministic Trend Score weighs recency, frequency, coverage, and movement from the sources that actually exist for an entity.
04
Quality-check
History depth, freshness, and evidence strength decide whether an entity is fit to publish.
05
Explain
An optional AI interpretation is generated from the same evidence. Missing explanations do not block publication.
06
Publish
Only entities that pass the public eligibility gate reach the website, API, and sitemap.
Why the public catalog can be empty
Collectors may hold thousands of observations while zero entities are public-eligible. History needs time to accumulate. The website is designed for that state: it explains what is being monitored, not fabricated trends.