M37R10N Research · Quantitative Systems

We build LLM-native research systems and derived factor datasets for China equity structure.

Translating continuous non-price corporate disclosure streams into point-in-time quantitative factors, network topology datasets, and model-native operating surfaces for institutional research.

Read Research Notes Institutional Diligence

Research Lines

Two ongoing lines of work.

Non-price market microstructure, and the agent runtime interfaces that make high-dimensional research reproducible.

Line 01 · Non-Price Structure & Factor Research

Non-price disclosure & topology factors

We extract structured corporate assertions, entity relationship graphs, and thematic community clusters from continuous exchange disclosure streams across China A-shares. The research translates unstructured textual disclosures into point-in-time correlation topology and structural filter factors under strict point-in-time discipline.

non-price disclosure network topology point-in-time china a-share

Line 02 · Model-Decoupled Agent Infrastructure

Model Context Protocol (MCP) systems

We build production Model Context Protocol (MCP) services that provide institutional quantitative research agents with an operational surface. Agents can query derived graph topology, inspect immutable evidence logs, and execute automated backtesting pipelines while downstream econometric rules neutralize model variance.

mcp services agent workflows reproducibility model-decoupled

Operating Principles

Four working principles.

We operate as a disciplined research partnership with an AI-native quantitative and operational stack.

Concentration

One core market, one focused partnership, two research lines. We do not spread across unrelated asset classes.

Reproducibility

Every number that leaves this desk traces back to versioned source data and timestamp audits. If a result cannot be re-derived, it does not ship.

Restraint

We publish empirical perspective, methodology, and postmortems. Factor weights, full ledger datasets, and specific execution parameters remain private to licensed deliverables.

Model-Decoupled

A senior research team augmented by AI agents. Downstream econometric discipline ensures our alpha outcomes remain robust and decoupled from single-model stochastic drift.

From the research desk

Notes and perspectives.

Selected public postmortems and cultural observations on quantitative pipeline plumbing, model behavior, and the wider history of technological systems.

MCP / Product Strategy · 18 October 2026

When First Principles Meet the Brutal Funnel

Why We Pivoted Our Revenue Back to Humans, Without Dropping a Single Line of MCP Code

6 min read Read article →

Quant & LLM Engineering · 4 October 2026

When Your Alpha Rests on a 0.087 Jaccard Index

Confessions of an LLM extraction pipeline that healed itself (or possibly just stopped noticing). A candid postmortem on evaluating competing models in production quant backtesting, stochastic model variance, and how deterministic downstream rules maintain institutional stability.

7 min read Read Technical Postmortem →

History & Risk · 4 October 2026

Filing the Apocalypse

Two centuries of machine panic from Asimov to pre-IPO S-1 filings. An analysis of eight cycles of technological anxiety, labor industrialization, and where frontier generative models sit in the historical coordinate system.

8 min read Read Essay →