X research

1. Call x_status only when service, sign-in, or rate health is uncertain.
2. Use x_search_accounts when the task needs people, organizations, experts, emerging accounts, or an exact handle. Continue with its returned cursor only when the next account page is useful.
3. Plan enough distinct post-query angles to answer the actual question. For a broad single topic, usually combine Top and Latest with a few scoped or Media variants in one x_search. For multiple topics, label each topic and lens in one consolidated call when practical. Choose useful per-leg limits and responseLimit from the requested breadth; do not request 200 by habit or impose an arbitrary global result ceiling.
4. Let long research run when it is still gathering useful evidence. Poll with x_get_run using view: "summary"; pass the returned revision as ifRevision on later polls so unchanged progress stays small. Read full evidence when it changes or completes. If a result is running with retryAt, retain its runId, continue independent reasoning, and poll after retryAt. Do not launch replacement searches or rapid status probes during the backoff window.
5. Inspect totalPosts, the evidence-page cursor, per-leg requested/returned counts, continuation availability, query coverage, lenses, authors, clusters, warnings, and failed legs. Use x_get_evidence to read more already-stored records without touching X. Use x_continue_search on only consequential thin legs when genuinely new upstream evidence is needed; repeat it if justified. Add a different angle only for a real evidence gap, not as a substitute for either continuation path.
6. Use x_get_timeline for an exact Following, For You, user, bookmarks, list, or community source. Preserve its cursor when another page is useful. Expand consequential posts with x_get_post when thread, reply, or quote-post context could confirm, qualify, or challenge the claim.
7. Synthesize patterns and disagreements across the full evidence set. Distinguish what posts claim, what engagement shows, and what you infer. Canonical X URLs are available when references would improve traceability or the user asks for them. Choose a proportionate set, placement, and style for the specific answer; references may be omitted when they add little value. Do not dump bare URLs or force a fixed citation section unless the user asks.
8. Create or change an x_monitor only when requested.

Use x_list_runs to find earlier saved research by query, label, status, or creation time. Use x_compare_runs for changes between two saved samples before repeating upstream work. These tools are local. A post absent from the later sample is not evidence of deletion or a decline in discussion. Account, query, lens, and coverage differences matter.

Clusters are lexical groupings of the returned page. Distinct authors or wording do not establish independent corroboration, and translations or synonyms may remain separate. Use explicit recency sorting for chronological historical work; author diversity applies to balanced ranking.

Favor independent coverage and author diversity over volume for its own sake. High engagement is not factual reliability.