Deep Research
Deep Research is the multi-agent investigation workflow in NataPulse.
It is designed for questions that require more than summarization: competing interpretations, cross-domain evidence, explicit risk review, time-horizon selection, and a final synthesis with sources and limitations.
Research modes
Section titled “Research modes”NataPulse mode is grounded in curated NataPulse data. It uses the product’s published events, clusters, reports, source evidence, market, social, SEC, on-chain, macro, and memory context as available.
General or Normal mode is not restricted to the curated NataPulse dataset. It can use model knowledge and permitted live web research. This mode must be read as broader web-grounded research, not as a statement produced only from NataPulse data.
A run can focus on a ticker, crypto asset, event, narrative, cluster, watchlist, market, or comparison. The builder also supports depth and horizon choices.
Team and phases
Section titled “Team and phases”The native workflow uses eleven specialist roles across six phases:
- Plan: Research Supervisor.
- Analysis: Fundamental, Technical, News & Macro, Sentiment & Social, and On-chain Analysts.
- Debate: Bull and Bear Researchers.
- Strategy: Research Strategist.
- Risk: Risk Reviewer.
- Synthesis: Portfolio Synthesizer.
Roles can be skipped when the requested scope or available evidence makes them irrelevant.
What the result contains
Section titled “What the result contains”Depending on the run, the detail view can include:
- current agent and phase;
- live findings;
- elapsed time and source count;
- confidence and stance;
- analyst views;
- bull and bear debate;
- risk review;
- final research synthesis;
- invalidation conditions;
- what to monitor;
- evidence groups;
- memory lessons;
- limitations.
Interpretation
Section titled “Interpretation”The final stance is a research synthesis, not an execution order. The most useful parts are often the disagreement, evidence quality, invalidation conditions, and monitoring plan.
Completed research can create a safe lesson for later runs on the same subject. This improves continuity without exposing internal traces, costs, or private control-plane data.