Research library

Teardown finds NavBoost click signals in the ranking repository

Moderate evidence

A credible test with a real sample and a stated method, but unreplicated or narrow in scope.

Researcher
Mike King
Published by
iPullRank
Date
27 May 2024
Method
Read the leaked Content API Warehouse modules line by line, mapped the named systems and attributes, and wrote a module-by-module technical analysis of what each signal appears to do.
Sample
2,596 modules and 14,014 attributes; NavBoost referenced 84 times across five modules, plus the Twiddler re-ranking framework and named boosts (QualityBoost, RealTimeBoost, WebImageBoost)
Original
ipullrank.com →

What they did

King parsed the leaked documentation as a technical practitioner, cataloguing named systems and attributes. He focused on NavBoost and the Twiddler re-ranking framework, counting references and grouping the click-related fields.

What they found

NavBoost modules define goodClicks, badClicks, lastLongestClicks and unsquashedClicks, with commentary that click data is normalised to prevent runaway manipulation. NavBoost appears 84 times across five modules and is described as long-standing.

How much weight to give it

The documents confirm these click fields are defined and named in the repository. They do not give the scoring weight, nor prove the fields are active in live Search today. King inferred an active click-based signal from the volume and detail of the references.

What it doesn't prove

Defined fields are not proof of live ranking use or importance. King states plainly that the weighting is unknown and that not everything documented is necessarily used. Google acknowledged the documents were authentic but rejected the ranking conclusions.

Google leaks

Spotted an error, or did we grade this unfairly? Tell us and we'll amend the entry: use the contact form.