Pulsar relevance feedback: a personalised model reshaping video results
A video twiddler carries the name Pulsar, which RESONEO describes as a personalised generative retrieval model also used for suggestions in the search box. Its image counterpart is named Starburst. None of this is confirmed by Google.
01What the schema and RESONEO show
The schema object is VideoPulsarRelevanceFeedbackTwiddlerParams in the quality_video namespace: settings sent with the query. Relevance feedback is an old idea in information retrieval: what a user does with the first results is used to adjust the next ones.
RESONEO presents this as the relevance feedback twiddler for video; the image variant is named Starburst. It describes Pulsar as a personalised generative retrieval model, meaning one that produces the results suited to a particular user directly rather than looking them up in an index, and says Google also uses it for personalised suggestions in the search box. Like every twiddler it only recommends; the twiddler framework in Superroot, the layer that merges what Google's indexes return, makes the decision.
02What the leak shows about Starburst
The name Starburst does appear in the 2024 Content Warehouse leak, as frame-level embeddings: numerical representations computed for individual frames of a video or image, in several versions. That confirms Starburst exists as an embedding family. It does not confirm RESONEO's pairing of Starburst and Pulsar as two relevance feedback twiddlers.
For Pulsar itself we found no leak entry and no Google text. The description of a generative, personalised model rests on RESONEO alone.
03E-E-A-T and the system map
Relevance feedback and personalisation decide what fits a person, not whether a source is trustworthy. The twiddler has no E-E-A-T bearing. The parent article places the image and video twiddlers outside this site's E-E-A-T system map. On our map it belongs to the twiddler layer, nothing more.
Personalised results cannot be optimised for in the usual sense: two people may see different videos for the same query. What a creator controls is the video itself, not the history of the viewer.