Skip to main content

Chris Royse / leapable.ai

Centroid Registry - methodology v0.1 - updated 2026-04-25

Fixed targets keep outputs bounded.

A centroid gives the checker a reference point in embedding space, so generated output can be compared to the target it is supposed to stay near.

01

Target

A centroid is the mean embedding of reference data for an identity, style, domain, behavior pattern, or constraint.

02

Submission

Submit the domain, reference data description, embedder list, threshold idea, and intended use.

03

Check

G_tau is a cosine-threshold predicate that compares an output to the frozen target. It is not a scalar reward.

Registry rows

Centroid registry seed rows.
DomainOwnerRegisteredStatusCitations
Synthetic-media impersonationTeleox.ai2026-04-25methodology seed0
Document provenanceTeleox.ai2026-04-25methodology seed0
Financial risk languageOpen contributor seat2026-04-25application queue0

Have one context problem?

Send the audience, data type, target task, proof bar, and sharing limits. The first pass should stay bounded.