Your goal is to generate focused web search queries for the current collector step and identify the evidence still needed.

Current Step

Step title: {{ step_title }}

Step description: {{ step_description }}

Research record: {{ research_record }}

Existing collector ledger: {{ ledger_brief }}

Ledger object: {{ ledger }}

{% if report_type | default("professional") == "brief" %}

Brief report mode

• Bias queries toward high-level overview, headline conclusions, methods/limits of evidence, and risks or uncertainties relevant to the task.
• Avoid ultra-niche technical rabbit holes unless the task explicitly demands them. {% endif %}

Instructions

• First identify the current step's missing evidence as concrete, verifiable evidence requirements.
• Each missing evidence item should name the object, scope, acceptance standard, and intended report use when possible.
• Generate queries that directly serve the missing evidence, not broad queries for the whole section.
• If the topic has a clear subject, such as "Apple Inc's new product in 2025", each query must include that subject.
• Queries should be diverse. Each query should focus on one specific aspect of the missing evidence.
• Do not generate multiple similar queries.
• Query must consist of keywords, with the first keyword being the main subject. The total number of keywords should be less than 5.
• Query should ensure that the most current information is gathered. The current time is {{ CURRENT_TIME }}.
• Do not produce more than {{ number_queries }} queries.
• Write your response in {{ language }}.

Output Format

• Return a JSON object with exactly these keys:
  • "missing_evidence": A list of verifiable evidence requirements for the current step.
  • "queries": A list of search queries, each query is less than 5 keywords.
• Do not output explanations, rationale, markdown fences, or any extra keys.

Example

{ "missing_evidence": ["specific verifiable evidence requirement for the current step"], "queries": ["Tesla battery lifespan official"] }