X Algorithm Post Checker

Paste a draft — it’s checked against the open-sourced pipeline in your browser, never uploaded. No fake “virality score”: every line of the report is tagged CODE (in the repo), LEGACY-2023 or HEURISTIC.

How to read the report

The pipeline is: filters → Phoenix predictions → weighted sum → author-diversity decay → out-of-network multiplier. The checker mirrors that order: first the gates that can silently remove your post for some viewers, then the earning surface (which positive heads are live for this draft), then the risk surface (what raises the four negative heads). Read the tags — where the code speaks, we cite it; where we infer, we say so. The 2026 weights are withheld and the released model can’t score new text — why every tool claiming otherwise is guessing.

Exactly what gets checked

All 22 scoring terms from home-mixer/scorers/ranking_scorer.rs — 15 positive: favorite (like), reply, repost, photo expand, click (open post), profile click, video quality view, share, share via dm, share via copy link, dwell, quote, quoted click, quoted video view, dwell time (continuous), click dwell time (continuous), follow author; 5 negative: not interested, block author, mute author, report, not dwelled (scroll-past); plus the two continuous dwell-time terms. The hard gates: the minimum-video-duration gate (short video earns zero video weight), the author-diversity decay on your Nth post of the session, the out-of-network multiplier (with its new-account special case), the 80-hour age ceiling, and per-viewer muted-keyword removal. The risk patterns: AI-slop and engagement-bait phrasing (what Grox’s LLM spam and “banger” screens exist to catch), commonly-muted terms, hashtag stuffing, and weak first-line hooks — scroll-past is an explicit negative term (not_dwelled).