A deterministic simulator built from the leaked X recommendation pipeline. Enter any @username, compose a post, and watch our engine run it through Phoenix action predictions, the WeightedScorer, author diversity decay, and visibility pre-filters — just like the real system.
Every component mirrors a real module from the X recommendation source code.
19 distinct action probability predictions per post. Real X weights: Reply 54×, Retweet 20×, VQV 40×, Quote 20×, Report −200×. From weighted_scorer.rs and phoenix_scorer.rs.
Simulates X's VF pipeline: link penalty (−20×), spam detection (−32×), brand safety (−64×), profanity scanning (50+ TR/EN words), and spam pattern matching. From vf_filter.rs and safety classifiers.
Out-of-Network penalty (×0.7 from oon_scorer.rs), Author Diversity decay (×0.6^n from author_diversity_scorer.rs), In-Network boost for followed accounts. Same formulas as the production X pipeline.
Real X rule from tweet_type_metrics: videos ≤10 seconds get ZERO video_quality_view weight. Three duration buckets with distinct dwell bonuses. Photo expand (2×) for image posts.
Engagement pool multipliers: weekday morning ×1.20 peak, night ×0.50 trough. Post age decay from engagement_counts_hydrator.rs cache TTL: fresh <30min ×1.15, stale >24hr ×0.25.
Live follower counts, following, account age, and verification status from X's public pages. No OAuth, no tokens. 7 follower tiers with power-law reach rates calibrated to real X benchmarks.
The exact 7-stage flow that mirrors X's recommendation architecture.
The architecture is taken directly from the leaked X recommendation pipeline. Phoenix ML predictions → WeightedScorer → Author Diversity → Pre-Filters → Selection is the exact flow described in the source. The numerical weights (Reply 54×, VQV 40×, etc.) are verified against the codebase.
That's a real X rule from tweet_type_metrics_hydrator.rs. The VQV (Video Quality View) weight — the single largest positive signal at 40× — only activates for videos above the minimum duration threshold. We mirror this rule exactly.
Out-of-Network. From oon_scorer.rs: posts from accounts you don't follow get a 0.7× score multiplier. This is why following count and follower ratio matter so much for reach.
Nothing is saved, logged, or sent anywhere. The X profile data is scraped from public pages in real-time. All algorithm computation runs in your browser. No cookies, no tracking, no database.
No sign-up. No API key. Enter a username and start simulating.
Look up an account, write a post, then simulate.
Live detection of profanity, spam patterns, links, sentiment, and X pipeline risks as you type.