Skip to main content
Identity clustering is the first step of the Fusion program. Burst groups related launches so it can later resolve which token becomes canonical.

​
How clustering works

Clustering happens off-chain. Burst groups launches only when they appear to represent the same underlying idea. Signals can include:
  • normalized ticker similarity,
  • normalized name similarity,
  • image similarity through perceptual hashing,
  • and optional source references such as a tweet or media link.
This is what prevents unrelated tokens from being merged.

​
What defines an identity

An identity is not a single field. It is a composite signal derived from the token’s name, ticker, image, and any external references. Burst uses this composite to group tokens into an identity_id. All tokens within the same identity_id are considered variants of the same underlying idea.

​
Clustering signals

​
Ticker similarity

Burst normalizes tickers and compares them. Tokens with the same or near-identical tickers are strong candidates for the same cluster.

​
Name similarity

Names are normalized and compared. Minor variations in capitalization, spacing, or formatting do not prevent clustering.

​
Image similarity

Burst uses perceptual hashing to compare token images. This catches edited, resized, or slightly modified versions of the same image. Learn more about visual clustering

​
Source references

If a token references a specific tweet, media link, or external event, that signal can also contribute to clustering.

​
What clustering does not do

  • It does not block token creation.
  • It does not prevent variants from trading on the bonding curve.
  • It does not affect pre-graduation behavior.
Clustering only determines which tokens are related. The actual resolution happens at graduation.