Skip to main content

​
Zonein Deep Research Engine

The Deep Research Engine is Zonein’s analyst, designed to answer complex, multi‑part crypto questions with sourced, explainable reports. It plans the work, gathers live data (on‑chain, market, social, docs), checks for gaps and conflicts, and synthesizes conclusions you can verify. The same engine powers quick briefs in the dashboard, page‑anchored answers in the browser extension, and full research notes. Advanced Research Workflow.png

​
How It Works

​
Input & intent

You ask in plain language (“Compare L2 X and Y: fees, decentralization, and treasury risk”). The engine classifies the request (comparison, event study, risk review, tokenomics, regulatory) and sets success criteria (what must be answered, what evidence is required, and how recency/coverage will be judged).

​
Planning & decomposition

A planner breaks the problem into sub‑tasks (e.g., fee telemetry, validator set analysis, treasury wallet mapping, incident history) and orders them to minimize re‑work. Each task specifies required sources, freshness targets, and validation steps.

​
Retrieval & extraction

Self-Reflective Research Architecture.png Orchestrated agents pull:
  • On‑chain (tx logs, holders, liquidity, governance votes).
  • Market (prices, volume, depth, derivatives).
  • Social/news (credibility‑weighted sentiment; release notes; governance forums; audits; GitHub).
  • Reference (whitepapers, docs, transparency reports).
    Extractors normalize entities (canonical token/contract map, chain aliases), deduplicate events, and attach provenance (source URL/id, timestamp, parse method).
A validation pass scores evidence (recency, source authority, cross‑source agreement, internal consistency). Conflicts are resolved by weighted consensus; unresolved items are flagged with an “Uncertain” badge and kept out of the executive summary. A synthesis agent assembles findings into: TL;DR; key evidence; risks/caveats; recommended next steps; appendices. It links each claim to citations and shows data freshness. Tables and charts render from the same features used in the dashboard (e.g., whale_netflow_1h, pool_depth@slippage, sentiment_delta_24h). A reflector checks for missing angles (e.g., unlocked supply, team wallets, oracle dependencies), weak evidence, or bias. If gaps are found, the plan is adjusted and missing data is fetched. The loop ends when success criteria are met and open questions are explicitly listed.

​
Delivery & updates

Reports stream in sections: a brief appears first; sections fill as evidence hardens. As new data lands (unlock posted, proposal passed), the report updates and a changelog notes what moved and why. Under the hood: an MCP‑compatible tool layer exposes search, chain queries, social/news APIs, and internal stores; a workflow orchestrator (e.g., LangGraph‑style) coordinates multi‑agent steps with state persistence; a feature store guarantees the same metric definitions across surfaces; and an uncertainty module attaches confidence scores to claims. Complete Research Process Flow.png

​
What You Use

​
Deliverables

  • Executive summary: 5–10 bullets: what matters, why, confidence.
  • Findings: concise sections (tokenomics, market structure, smart‑money flows, security, governance).
  • Evidence: linked tables/charts; raw transactions; forum/audit references.
  • Risks & watch‑items: concrete triggers to monitor.
  • Actions: decision rubrics (enter/avoid/size/hedge), or research next steps.
  • Appendices: methodology, data cuts, parameter settings.

​
Interactive report

  • Citations you can click to open source context; evidence tooltips show how a metric was computed.
  • Time sliders to compare now vs. last week/month; what‑changed diffs between report versions.
  • Ask‑inside‑report follow‑ups (“show peer TVL comps using same window”).

​
Controls

  • Scope: chains, tokens, date windows, peer set.
  • Depth: quick brief → standard → deep dive (adds extra checks like developer activity, MEV profile).
  • Risk lens: conservative / balanced / aggressive (alters thresholds and emphasis).
  • Export: PDF/CSV for tables, shareable link with filters encoded; alerts from findings (e.g., “alert if treasury outflow > X”).