We've only covered the Department of Justice vs Google documents here and there but here are some documents that may be of interest to the community here. Google said Navboost is not a machine learning system, instead it is more of a big table. And there are more documents including one on hand crafting signals and another on a debugging interface.
Brett Tabke from WebmasterWorld/Pubcon uncovered these yesterday and I wanted to highlight them for more attention.
Dr. Eric Lehman, Former Google Distinguished Software Engineer, said Navboost is not a machine learning system but rather a big table (database table, I assume): He said, "Navboost is not a machine-learning system. It’s just a big table." He added, "It says for this document – sorry, for this search query, this document got two clicks. For this query, this document got three clicks, this – and so on. And it’s aggregated, and there’s a little bit of extra data. But you can think of it as just a giant table."
Then this bit on Hand Crafting of Signals that says "Almost every signal, aside from RankBrain and DeepRank (which are LLM-based) are hand-crafted and thus able to be analyzed and adjusted by engineers. To develop and use these signals, engineers look at data and then take a sigmoid or other function and figure out the threshold to use. So, the "hand crafting" means that Google takes all those sigmoids and figures out the thresholds."
Of course, this is part of that document, "Clicks (C) - historically, how long a user stayed at a particular linked page before bouncing back to the SERP." Something Google has outright said over the years is not a thing - pogosticking.
I am not sure if this quality rater explanation is about using there scores directly in the ranking systems or doing regression analysis after - Google has said countless times quality raters do not directly impact rankings:
Google has been telling us we don't use raters directly for ranking signals meanwhile HJ Kim says in performing regression for majority of signals label data from human raters is used. https://t.co/332T21BRdN pic.twitter.com/XSzYAcp6VO
— Gagan Ghotra (@gaganghotra_) May 13, 2025
And then this about a debugging interface for "Pandu typed "james allan umass" into a Google search window and then called up the internal debugger window." "containing a table with list of 10 blue links and corresponding score for each top-level signal as well as corresponding total score "Final IR" across all." Super interesting.
Of course, the bits on Q*, RankEmbed, Twiddlers and so on...
Interesting!!!! https://t.co/332T21BRdN pic.twitter.com/FgPz1OVz98
— Gagan Ghotra (@gaganghotra_) May 13, 2025
New evidence released in the Google Antitrust case —with fresh details about how Google's Algo works, via @btabke
— Cyrus (@CyrusShepard) May 12, 2025
This one, "February 18, 2025 Call with Google Engineer H.J Kim" is a doozy...
The ABCs of ranking: (A)nchors (B)ody (C)licks
Every SEO should read the whole thing pic.twitter.com/DCj9D1r0XX
Enjoy chewing on these bits, and there is more here.
Forum discussion at X.
