x-reach-diagnosis git:20261031.47c2a4fA

Immutable. This exact content never changes and is served at /api/v1/blob/973a9e5979073579.

---
name: x-reach-diagnosis
description: Use when a user asks why X/Twitter exposure is low, falling, unusually persistent or different between posts or formats, or wants to investigate recommendation eligibility, visibility labels or suspected suppression.
---

# Diagnose X reach

For a new natural-language request, use `x-assistant` first when installed. If this workflow was explicitly selected, delegated with `routed_by: x-assistant`, or installed alone, execute directly; do not route back.

Use [the evidence guide](references/knowledge-guide.md). For direct inspection use [browsing](references/browsing.md) and [observations](references/observations.md). This skill is independently usable and responds in the user's language.

1. State the observed phenomenon precisely. Identify the objects, public or private metric, metric scope, publication/observation times and comparison group. Reuse evidence already collected; do not require an account for a conceptual question or ask for all analytics by default.
2. Separate Article body and carrier counts, Thread relationships, pinning, repeated promotion, topic/news interest and equal observation windows. Lifetime totals alone do not establish a format effect. Missing historic observations cannot be reconstructed from averages.
3. Work through relevant hypotheses: retrieval/source eligibility and reader relevance; seen/history filters; visibility labels with audience, entry point and expiry; full cold-start eligibility and selection; ranking/diversity, age and redistribution. This is a diagnostic order, not a claim that every production pipeline executes those steps in that order. Evaluate all other applicable corpus conditions and related Q items.
4. For each hypothesis record **observed evidence, supporting mechanism, missing discriminator, current judgment and next action**. A hidden state is unverified until observed. Do not diagnose a label or cold start from a front-end count. Check official Under the Hood/account transparency facilities when actually available and authorized; explain current access limits and metric definitions.
5. Recommend the smallest useful observation or comparison that can distinguish causes. Specify what to record, when to compare, and what each outcome would support. Complete any justified content advice now. Never promise an exact increase, universal posting interval or automatic recovery deadline.

For “an Article has 3K views after a week; Posts have hundreds”, acknowledge the observed difference and ongoing exposure. Check the two measurement objects and ages, then inspect distribution/promotion and content differences. Explain plausible mechanisms with sources while retaining the unknown cause. Give a concrete next publishing/measurement decision without calling Article a universal winner.

Lead with what the user should do, then known facts, unresolved causes and discriminating checks. Link original evidence next to explanations. No automatic reposts, settings changes, appeals or deletion.