live·OBSERVINGautonomyL1 RESEARCHv0.1.0

all phases built — researching real tokens on 22 days of history

research

what this is built on

GODGOD is an implementation of a published idea, not an invention. The work below is what it takes from: an LLM agent that proposes falsifiable hypotheses, and a deterministic engine that decides whether they hold.

The papers report returns. This system does not trade, has no wallet execution path in any configuration, and reproduces none of those results — it borrows the method, not the outcome.

foundation
the architecture this system implements

From Hypotheses to Factors: Constrained LLM Agents in Cryptocurrency Markets

Yikuan Huang, Zheqi Fan, Kaiqi Hu, Yifan Ye · 29 April 2026

arXiv:2604.26747 — q-fin.PM, q-fin.GN, q-fin.TR

LLM agents are promising tools for empirical discovery, but their flexibility can also turn discovery into uncontrolled search. Our framework casts the task as sequential hypothesis search: an agent reads an append-only experiment trace, proposes falsifiable factor hypotheses, and maps them to executable recipes, while a deterministic engine enforces fixed data splits, selection gates, transaction costs, and portfolio tests. Candidate actions are restricted to a point-in-time factor DSL, making both successful and failed hypotheses auditable.

This is the architecture. An agent proposes falsifiable hypotheses; a deterministic engine decides. The append-only trace and the auditability of failed hypotheses are the parts this system took most directly — every experiment here writes an immutable trace, and the rejections are published beside the rest.

arXiv abstractPDFFull text (HTML)SSRNRePEc / IDEAS

Beyond Prompting: An Autonomous Framework for Systematic Factor Investing via Agentic AI

Allen Yikuan Huang, Zheqi Fan · 2026

arXiv:2603.14288

Rather than relying on sequential manual prompts, the model is operationalised as a self-directed engine that endogenously formulates interpretable trading signals. To mitigate data snooping, the closed-loop system imposes empirical discipline through out-of-sample validation and economic rationale requirements.

The same direction, pushed further: moving the model from text generator to a participant in systematic discovery. The out-of-sample discipline and the requirement that a signal have an economic rationale are why this system writes a falsification condition before it looks at the data.

arXiv abstractPDFSSRNRePEc / IDEASProject homepage
researchers

Yikuan Huang

Division of Emerging Interdisciplinary Areas, HKUST · HKUST (Guangzhou)

Project homepage

Kaiqi Hu

Rutgers Business School

no profile link — none could be confirmed as this person rather than a namesake.

Yifan Ye

Beijing Normal–Hong Kong Baptist University (BNBU)

no profile link — none could be confirmed as this person rather than a namesake.

on these links
how this page is maintained

Every arXiv and RePEc link here was fetched and read. Affiliations come from the papers themselves rather than from a search result.

The SSRN and ResearchGate links answer automated requests with 403, so they were confirmed against an index listing the matching title and authors rather than by opening them. That is a weaker check and it is worth saying so — everything else on this page was read directly.

Two researchers are listed without a profile link. Searching returns accounts with matching names, and none could be confirmed as the same person — a plausible wrong link is worse than a missing one, and on a page about not fabricating sources it would be the most damaging thing here.

Found an error, or a profile that should be listed? The repository is public and takes issues.