MRTAN Winner

Independent engineering · Quantitative systems

From research
to execution.
One infrastructure.

AI-assisted quantitative research & trading infrastructure. Connecting point-in-time data, systematic research, real-time signals and execution engineering.

Explore the productView architecture

Built with Claude and Claude Code as engineering and research copilots.

Point-in-time dataReal-time Linux engineHistorical replay
SYSTEM ARCHITECTURE01 / OVERVIEW
01 · INPUTMarket data
02 · TRANSFORMNormalization
03 · RESEARCHFeature engine
04 · EVALUATESignal engine
05 · CONTROLPortfolio & risk
06 · EXECUTEOMS & bridge
DETERMINISTIC PRODUCTION LOGICILLUSTRATIVE
Research / Historical replayEngineering / Claude-assistedExecution / Account isolationReview / Read-only archive

01 / Platform

Built as infrastructure,
not a single strategy.

A connected engineering foundation for research, production signals and the full execution lifecycle.

01 / DATA

Point-in-Time Feature Lake

Historical feature snapshots with explicit as-of dates, supporting reproducible research and live inference workflows.

02 / RESEARCH

Quantitative Research

Historical replay, signal profiling, feature validation and systematic experiment workflows.

03 / SIGNALS

Real-Time Signal Engine

Streaming market processing, scoring, ranking and decision gating in a Linux-based live environment.

04 / CONTROL

Portfolio & Risk Logic

Signal prioritization, capital allocation, slot management and execution-aware portfolio controls.

05 / OPERATIONS

Execution Infrastructure

Order management, routing, QMT integration, callbacks, reconciliation and operational monitoring.

06 / ISOLATION

Multi-Account Architecture

Account-level separation of capital, holdings, permissions, orders and execution lifecycle.

02 / System architecture

A structured path from
market data to execution.

Market dataIngestionNormalizationFeaturesSignal scoringDecision gatingPortfolioOMSExecution bridge

Research layer

  • Historical replay
  • Feature validation
  • Signal profiling
  • Ranking experiments

Live engine

  • Linux authority node
  • Real-time scoring
  • Portfolio controls
  • Operational monitoring

Execution layer

  • Order management
  • QMT integration
  • Callback reconciliation
  • Account-level isolation

03 / Built with Claude

An engineering partner.
Across the lifecycle.

Claude and Claude Code are integrated into the engineering workflow behind MRTAN Winner.

They support architecture, implementation, code review, debugging, research assistance, testing and documentation. Deterministic quantitative systems remain responsible for production signals and execution logic.

Claude does not control trading accounts or independently make investment decisions. No Anthropic endorsement is implied.

01Architecture & engineering
02Software development
03Code review & refactoring
04Debugging & failure analysis
05Quantitative research assistance
06Testing & documentation

04 / Inside MRTAN Winner

Real data.
A separate review surface.

Explore a dated export from the research data archive. Public historical market fields and dataset metadata, with production accounts kept private.

RESEARCH ARCHIVE / 2026-09-30READ ONLY · HISTORICAL
223Feature snapshot files foundFile inventory; not a completeness certification
5,589Rows in the latest feature snapshot2026-09-30 · one feature dataset
125Columns in this datasetIncludes identifiers and metadata

The interactive archive contains actual exported records. These are historical observations, not live system health, portfolio values or trading recommendations.

Open interactive archive

05 / Engineering journey

Built over time.
Documented along the way.

Retained engineering documents date back to November 2025. Repository activity and research records offer a closer look at the work behind the platform.

SELECTED RETAINED RECORDS

  1. Order-book data integration documentation

    Document content date · 3 matching retained copies
    SHA256 7328f16e395d…

  2. Research console architecture documentation

    Document content date · 2 matching retained copies
    SHA256 ef4d5c659616…

  3. Existing production code imported as the repository baseline

    Commit 40cb0212452a

  4. Read-only market observation panel

    Commit cd725ab71169

  5. Execution console evolves into an account center

    Commit d617e2079d7d

Document dates come from their contents. Matching backup copies preserve the same bytes; they are not independent timestamp certification.

RESEARCH AT SCALERECORDED
1,348Registered research records
1,343Registered run records

Across 1,172 distinct numeric research identifiers. Highest registered identifier: R1384.

The registry includes historical imports, engineering work, active runs and unsuccessful investigations. These counts do not imply completed, validated or profitable experiments. An R-number is an identifier, not a count.

CAPTURED PROCESS STATES

  • Web consoleACTIVE / RUNNING
  • Read-side APIACTIVE / RUNNING
  • Market state serviceACTIVE / RUNNING
  • Account coordination serviceACTIVE / RUNNING

Captured 08 Oct 2026 · 22:33 UTC+0800. A one-time process-state observation, not live monitoring, trading readiness or an uptime claim.

PRIVATE REPOSITORY / PUBLIC ACTIVITY SUMMARYGIT METADATA
1,471Reachable commits, including merges
78Dates with retained commit activity
4Calendar months represented
08 Jul 202608 Oct 2026

Each square represents one day in the retained Git period. The first commit imports an existing production baseline; it is not the start of all development.

Commits by month · UTC+08
MonthCommits
Jul 2026378
Aug 2026389
Sep 2026499
Oct 2026205
Evidence sources & counting method

Snapshot collected 08 Oct 2026 · 22:33 UTC+0800. Git statistics cover all commits reachable from the captured source HEAD 9d434ae7a0b1, including merges, using committer dates in UTC+08. Retained commit activity covers 08 Jul 2026–08 Oct 2026; earlier engineering is represented by dated internal documents, not by invented Git activity.

Research totals count distinct round keys and distinct round/run pairs in the local research registry. The highest R-number is reported separately. Market observation dates and feature as-of dates in the product archive are not used as proof of development age.

The downloadable summary includes document and registry hashes, daily commit counts and sanitized milestones. Hashes identify the examined bytes; this is a founder-published evidence summary, not a third-party audit. Repository contents, raw research records and account data remain private.

Download sanitized evidence · JSON

06 / Reliability by design

Clear boundaries.
Controlled execution.

Independent research and execution layers, deterministic production controls, account-level isolation, recovery and reconciliation workflows, and operational monitoring.

This showcase serves a static, read-only export. Browsing it cannot submit orders or modify the production system.

Founder

JunYong Tan

Founder, MRTAN Winner

Building AI-assisted quantitative research and trading infrastructure.

Contact founder

founder@mrtanwinner.xyz