A psyop is an X account paired with a function that scores any tweet against the question "is this what I want in my feed?". Run the psyop on a schedule and X's recommendation algorithm converges that account's For You feed toward the function's high-scoring distribution. The loop has four steps: Capture — A Chromium extension running on x.com sends every tweet ID it sees in the For You feed to a local queue (for_you_queue). The extension only ever sees IDs; nothing leaves the machine until the next run. Hydrate & filter — psyops run drains the queue, fetches tweet metadata via the X v2 API, and applies per-source filters: thresholds on likes/retweets/replies/impressions, age windows, engagement ratios, or arbitrary Starlark expressions. Score — Surviving tweets pass through an N-stage ObjectiveAI pipeline. Each stage is a function + profile + threshold; tweets falling below output_threshold or outside output_top are dropped before the next stage. One stage works fine for cheap functions; multiple stages let you cascade (e.g. cheap heuristic → mid-tier model → expensive swarm vote) so only promising candidates pay the full cost. Engage — The top survivors are liked or retweeted as the psyop's X account via the X API. Authentic engagement is the only signal X's algorithm trusts; consistent on-target likes/retweets train For You to surface more of the same. The next run captures a more on-target For You feed than the last. After a few cycles, the feed converges. The function defines the destination; the loop drives the algorithm there.

psychological-operations by ObjectiveAI

Agentic X Algorithm Tuning with Intent

Agentic X Algorithm Tuning with Intent