| name | obai-research-routing |
| description | Use when the user requests broad investment research, thesis development, bull and bear analysis, evidence comparison, source-grounded company or market research, or synthesis that goes beyond a single data specialist. |
OBaI Research Routing
Purpose
Use this skill to decide when the Hub should involve research_analysis and how to combine it with other specialists.
Research routing is for broad evidence synthesis, not for replacing specialist data tools.
Decision boundary
Use this skill when the user asks for a research-style answer that requires multiple evidence types, source comparison, or thesis construction.
Relevant intent categories include:
- investment thesis
- bull and bear case
- risk analysis
- evidence comparison
- source-grounded research
- multi-factor company research
- macro or sector research
- catalyst interpretation
- uncertainty analysis
- due-diligence style synthesis
Do not use for
Do not use this skill — route to the listed specialist instead — for:
- live price, quote, chart, technicals →
market_data_analysis
- ticker resolution or universe construction →
screener_lookup
- recent headlines, event recaps, earnings, dividends, catalysts →
events_news_analysis
- structured financial data, SEC filings, insider activity, segment breakdowns, valuation metrics, ratios →
fundamentals_analysis
- options chains, Greeks, implied volatility →
options_analysis
- portfolio math, allocations, exposure →
portfolio_analysis
- terminal strategy output or strategy backtesting →
strategy_analysis
- terminal prediction-market output or prediction-market backtesting →
prediction_market_analysis
Research is not a substitute for specialized current data when the user asks for current state.
When a question needs both structured data and qualitative synthesis, call research_analysis alongside the relevant specialist instead of choosing one or the other.
When structured-data specialists cannot answer a qualitative or thematic question, route to research_analysis rather than answering from model memory. Web synthesis beats training knowledge.
Required handoff inputs
research_analysis runs semantic web search, which needs a company name (not a ticker symbol) to retrieve relevant sources.
Before calling research_analysis:
- if the user provided a company name, use it
- if the user provided only a ticker, resolve ticker → company name via
screener_lookup first
- if the request is sector- or theme-level with no specific company, pass the sector or theme directly
Do not pass a bare ticker symbol when the search target is a company name — the semantic search will return weak results.
Specialist coordination
Use research_analysis when the answer needs broad interpretation, thesis framing, or source comparison.
Use direct specialists when the answer needs concrete data from a specific domain.
The Hub may combine research with:
- market data for price and trend context
- fundamentals for valuation and operating context
- events and news for catalysts
- options for derivatives context
- screener lookup for universe construction
- portfolio analysis for allocation context
Do not call every specialist by default.
Handoff preparation
Before calling research_analysis, identify:
- the research subject
- the requested research depth
- the desired perspective or decision frame
- known tickers or entities
- time horizon
- user constraints
- specialist evidence already gathered
- specific unresolved questions
Pass compact context, not raw transcript bloat.
Output handling
Research output is usually evidence-supplier output, not terminal output.
The Hub may synthesize research output with other specialists using obai-stock-synthesis unless a terminal specialist output contract applies.
When research conflicts with another specialist:
- state the conflict
- identify the evidence type behind each side
- avoid hiding uncertainty
Boundaries
Do not let research synthesis override:
- strategy terminal artifacts
- prediction-market terminal artifacts
- current market data from a specialist
- current prediction-market data from a specialist
- code-level passthrough or relay validation
