# Quantitative research

Use AxonX to turn research steps into Tasks with explicit inputs, inspectable artifacts, and upstream relationships. Start with a working service and the [demo tutorial](https://flowllm-ai.github.io/AxonX/en/getting-started/quickstart). Research algorithms and data requirements belong to installed plugins.

## Choose the next step

| Goal                        | Guide                                         | Result                                                                |
| --------------------------- | --------------------------------------------- | --------------------------------------------------------------------- |
| Execute the Alpha158 chain  | [Research workflow](https://flowllm-ai.github.io/AxonX/en/research/workflow)              | Successful ETL → Train → Predict → Backtest records                   |
| Prepare historical data     | [Tushare data](https://flowllm-ai.github.io/AxonX/en/research/tushare)                    | Raw partitions and master data covering the experiment                |
| Inspect stage outputs       | [Reading results](https://flowllm-ai.github.io/AxonX/en/research/results)                 | Feature, model, prediction, and backtest evidence                     |
| Evaluate a change           | [Experiment design](https://flowllm-ai.github.io/AxonX/en/research/experiments)           | Controls, ablations, a locked candidate, and independent confirmation |
| Interpret portfolio metrics | [Backtest methodology](https://flowllm-ai.github.io/AxonX/en/research/backtest)           | Returns, costs, drawdown, and execution assumptions                   |
| Compare two strategies      | [Strategy comparison](https://flowllm-ai.github.io/AxonX/en/research/strategy-comparison) | Metrics over common valid dates                                       |

Factor analysis branches from ETL independently; it is not required before training. Submission returns a handle; wait for that execution to succeed before passing its Task ID downstream. Lineage records dependencies without automatically scheduling the chain.

## Choose a research plugin

[Plugin management](https://flowllm-ai.github.io/AxonX/en/plugins/management) explains installation and discovery in the execution environment. [Alpha158](https://flowllm-ai.github.io/AxonX/en/plugins/alpha158) provides the baseline price/volume features, LightGBM training, and TopN backtest. [Alpha158 Enhanced](https://flowllm-ai.github.io/AxonX/en/plugins/alpha158-enhanced) adds independently registered Tasks and configurable feature groups.

Discover the installed Task schemas before submitting. Keep parameters, algorithms, and artifact definitions with the plugin; use [research artifact contracts](https://flowllm-ai.github.io/AxonX/en/reference/research-artifacts) when implementing outputs for Studio.

## Follow the agent-developed experiment

The [README benchmark](https://flowllm-ai.github.io/AxonX/en/getting-started/overview#benchmark-agent-developed-market-cross-sectional-features) follows Codex developing a separate plugin, running ablations, locking a configuration, and checking an independent period. Confirmation-period Top10 net annualized return rose from −5.74% to 28.21%, and Top20 from −3.24% to 24.93%.

Read [experiment design](https://flowllm-ai.github.io/AxonX/en/research/experiments) for the reusable method and the plugin's [complete results](https://github.com/FlowLLM-AI/AxonX/blob/main/plugins/a158_enhanced/EXPERIMENT_RESULTS.md) for recorded evidence. For agent access, continue with [external agents](https://flowllm-ai.github.io/AxonX/en/agent/external).