Alpha Search · Frontier Notes¶
A living research workspace on AI-driven alpha factor search, autonomous research agents, and the broader agent technical landscape.
Maintained by Paul Weng · Last meaningful update: 2026-05-19
What's in this site¶
This site is the public-rendered version of a research workspace. It collects three layers of work that nest into each other:
-
Agent Landscape
A 6-layer concentric mapping from Financial Agents (innermost) outward to Algorithmic Foundations (outermost). ~70 papers standardized with author / year / venue / core idea / code.
Concentric (paper-level) Technical stack (vertical) 70-paper index
-
Alpha Search Research
Deep dive into automated alpha factor search — frontier papers, 8-tradition systematic survey, statistical methodology, and a concrete Crypto-Alpha-Bench proposal.
Research Plan Frontier baseline notes 8-tradition systematic survey
-
HKU Briefing
Talk preparation materials for a 2026-05-20 research discussion with Prof. Kai Han and Prof. Guodong Li at HKU.
-
Risk & Reality Checks
Critical examination of the proposals — what could go wrong, what's missing, where the cognitive blind spots are.
Crypto-Alpha-Bench risk analysis Human-expert-in-loop direction Financial SOTA agents survey
One-sentence thesis¶
Alpha auto search has the same generator-verifier structure as AlphaProof / AlphaEvolve / ASI-ARCH, but the field has no unified baseline. Building one is both a research contribution in its own right and the prerequisite for everything else.
30-second elevator pitch (nested 6 layers)¶
| Layer | What we claim |
|---|---|
| L5 | Walk-forward + PBO + DSR is our finance-grade analog of AlphaProof's Lean kernel. |
| L4 | We are LLM-substrate agnostic — DeepSeek-V4 or Claude both fit our generator-verifier interface. |
| L3 | We use ReAct + multi-agent (Researcher / Engineer / Analyst). |
| L2 | Our production system rivals Claude Code / Devin in engineering rigor, specialized for trading. |
| L1 | We are an autonomous research agent in finance, inheriting FunSearch's verifiable-novelty pattern. |
| L0 | We span alpha-mining + quant R&D + risk-gating under one verifier — Crypto-Alpha-Bench. |
How to navigate¶
- First-time visitor: read this index, then
Research Plan(30 min for the full agenda) - Looking for a specific paper or technique: use the 70-paper index — every paper has standardized metadata
- Want the visual landscape: open Agent Landscape Concentric — 6-layer mapping with cross-references
- Preparing to give a talk on related material: see HKU Briefing for a worked example
What's missing / will be added¶
- ETH validation experiment results (Crypto-Alpha-Bench risk Risk 3 mitigation)
- First version of Crypto-Alpha-Bench protocol spec
- Reference implementation of L1-L5 baseline suite (gplearn / FactorVAE / AlphaAgent / M8.6 tradability gate)
- Backlinks / graph visualization (planned upgrade to Quartz layer)
Source¶
This site is auto-generated from the agent-landscape branch of the paulweng/alpha-search-frontier-notes repository via MkDocs Material + GitHub Actions.
To suggest a correction or extension, edit any page directly on GitHub (use the pencil icon at the top-right of each page).