Quickstart

From zero to your first evidence-backed factor report in about 15 minutes.

The goal: connect Quandora to your AI agent, run one research task, and analyze one evidence-backed factor result.

Who This Is For

Solo systematic traders, AI power users, and market-curious builders who use Codex, Claude, Cursor, CodeBuddy, WorkBuddy China, or Kimi Code and want their agent to work from market evidence instead of guesses.

Before You Start

• A supported agent host
• A Quandora account
• 10–15 minutes
• No exchange API key — this research workflow does not place trades

1. Install And Connect

Follow the Installation Guide for your host. Complete the browser OAuth flow, then start a new chat or task so the host can discover all five Quandora skills.

2. Run Your First Factor Task

List the public research tasks:

/quandora:factor-mining show public tasks

Pick one, or let the agent choose:

Use Quandora Factor Mining to pick a public research task, generate a factor,
run the backtest, and give me a short result summary.

Your agent reads the task card, checks for materially similar work, writes plugin.py, validates the complete source, and submits it. Quandora binds market data and runs the evaluation server-side.

3. Analyze The Exact Result

After the run reaches a terminal state, ask for read-only analysis:

/quandora:factor-analysis analyze my latest factor result

Factor Analysis resolves one exact run and reads its owner-scoped, server-persisted In-Sample evidence. It does not need a local ZIP, Python runtime, notebook, or extracted chart folder.

Read the Factor Card verdict, Health evidence, grade, metrics, risks, and proposed controlled experiments. See How Factors Are Judged for the current Success and grade semantics.

4. Find The Optional Export

When your host can write local files and the completed Result Bundle is ready, Factor Mining saves one verified ZIP:

Quandora result/factor/<factor_slug>.zip

The archive is not automatically extracted. Its manifest is authoritative for included, pending, and omitted items. A partial bundle can still be readable; a materializing bundle can be requested again later.

You Are Done When

Your agent ran one task and explained one exact result using server evidence. A failed factor still counts: negative evidence helps identify the next useful experiment.

Factor Mining and Factor Analysis do not place trades. Strategy Paper Trading uses simulated orders. Live trading is not part of the public product. A backtest is evidence about tested conditions, not a promise about the future.

Next Steps

• Browse the research task families
• Learn how to improve and rerun with explicit confirmation
• Build a cross-sectional strategy with the Strategy Tutorial
• Keep the Glossary open for unfamiliar terms