Explain the eight-stage X recommendation pipeline and post ranking process.
npx skills add https://github.com/CloudAI-X/x-algo-skills --skill x-algo-pipeline
Please help me install this Agent Skill. Skill: x-algo-pipeline Source: https://github.com/CloudAI-X/x-algo-skills/tree/main/x-algo-pipeline Command: npx skills add https://github.com/CloudAI-X/x-algo-skills --skill x-algo-pipeline
Clarifies how the X recommendation pipeline processes and ranks posts from hydration to final feed, enabling quick debugging, optimization, and capacity planning for large-scale recommendations.
Trace the eight stages of the pipeline from a sample user request to the final feed output to observe where data is hydrated, scored, filtered, and selected.
High-intent search queries and answers about installing and using this skill.
The X feed ranking pipeline processes posts through an eight-stage architecture: Query Hydration, Sources, Candidate Hydration, Pre-Score Filtering, Phoenix Scoring, Weighted Scoring, Selection, Post-Score Filtering, and Side Effects.
Thunder sources supply in-network candidates while Phoenix sources generate out-of-network candidates, both feeding into the candidate hydration stage before pre-score filtering and scoring operations.
Debug X feed ranking behavior by tracing the eight-stage pipeline from user request to final feed output, inspecting where data is hydrated, scored, filtered, and selected.
The Phoenix Scorer evaluates out-of-network candidates, passing results to the Weighted Scorer which combines scores with author diversity and out-of-network adjustments before final Top-K selection.
Apply pre-score filtering to remove invalid candidates before Phoenix scoring to save compute, and apply post-score filtering after Top-K selection to enforce final side effects and feed constraints.
Scalability planning for the X recommendation pipeline requires mapping capacity across eight stages, addressing bottlenecks in candidate hydration, Phoenix scoring, and Top-K selection throughput limits.
pax-kIngest documents into an Obsidian wiki via the wiki-ingest pipeline.
michaelladyConvert long-form videos into 30–90 second OpusClip shorts with viral-score filtering.
omicverseBuild phylogenetic trees and compute tree-aware diversity metrics from 16S amplicon data.
hung-phanDesign two-stage recommender systems with candidate generation and ranking components.
