"Engagement pods / early-boost groups beat the algorithm"

Verdict: MECHANICALLY WEAKENED by the 2026 design

Phoenix predicts per-viewer engagement probabilities from each viewer's own history — candidates never see each other's engagement counts in the scoring pass (candidate-isolation attention mask), so a pod's likes don't raise other viewers' predicted P(action) directly. Meanwhile Grox runs LLM spam screens including low-follower reply-spam detection. Pod likes are also the cheapest head (2023: 0.5).

The code

phoenix/README.md (candidate isolation); grox/classifiers/content/spam.py; grox/plans/plan_spam_comment.py.

Verified against xai-org/x-algorithm @ 0bfc279 on 2026-07-31. Withheld layers (params, visibility filtering, Grox prompts) mean “absent from open code” is the strongest claim honestly available — see methodology.

Test a real draft instead →