The pipeline
Five stations, one machine.
Read
Agents read the open field continuously: threads, posts, comments, videos.
Posts, threads, comments, videos
Classify
The model reads every candidate mention and decides what it is: which brand, which niche, real talk or noise.
Brand, niche, context
Merge
Mentions dedupe and merge into one record per brand. One name, one record, however loud the field gets.
One brand, one record
Score
Each brand gets a momentum score from 0 to 100, normalized within its niche and smoothed over time.
0 to 100, niche normalized, smoothed
Deliver
Scores become a daily brief, dossiers on demand, and spike alerts, on the desk and in Telegram.
Brief, dossier, alert
The number
What the score means.
Range
0 to 100, recomputed continuously. The scale never moves. The reads move across it.
Chanel
72.8
Pixi
57.3
Laneige
45.9
Niche normalized
A niche of quiet forums and a niche of loud video are graded on their own curves.
Cosmetics
Chanel
72.8
Confectionery
Crafters Zone
67.8
Smoothed
Momentum is a trajectory, not a single spike. One viral day cannot whipsaw the number.
Chanel
Scale recomputed continuously. Snapshot 2026-07-17.
Four reads
The four states of a name.
rising
momentum building from a low base
Too Facedhot
high and still climbing
Goodrplateauing
high but no longer climbing
Alani Nufading
momentum leaving the name
Poppi
Discipline
Refusals, by design.
Thin data is labeled an early signal, never dressed as a verdict.
Every score traces back to mentions you can open and read.
Fads, giveaways and bought noise are filtered out.
No private data. The agents read the open field only.
The read
See the method working.
The same engine runs behind every Riveln desk. Open Prodrom to watch it read a board live, or ask us for a desk of your own.
Jurisdiction
Built and operated in the European Union.
Residency
Signal data is processed and stored on EU infrastructure.