OXARUS :: OXARUS QUANT

Market engineering and research · Oxarus Quant

Oxarus Quant builds L.I.T., a market terminal, alongside custom trading robots, published research and institutional market infrastructure.

page

/en/quant

unit

Quant

path

Oxarus home › Oxarus Quant

markdown

/en/quant.md

Interface data marked as representative is illustrative; it is not a real client, result or measurement.

Not investment advice. Trading in financial markets carries a risk of losing capital.

ENGINEERING FOR MARKETS.

OXARUS QUANT

We build systems that run on our own capital. We also build the tools, research and infrastructure behind them for clients.

  • Measure first, then decide
  • Publish the ideas that fail, too
  • Build the halt controls from day one
  1. Terminal
  2. Robots
  3. Research
  4. Engineering

Representative screen. Not investment advice.

FOUR AREAS, ONE DISCIPLINE.

AREAS

  • L.I.T. (Early access)

    A terminal that brings news, the economic calendar and market data onto one screen.

  • Custom trading robots (Service)

    Describe your rules, and we'll write a MetaTrader 5 robot built for you alone.

  • Research (Published)

    Measurements on levels, gaps and price behavior.

  • Engineering (Institutional)

    Data pipelines, testing infrastructure and risk controls.

Representative screens. Not investment advice.

EACH AREA FEEDS THE OTHERS.

LOOP

The four areas may look like separate products, but they run on shared infrastructure. What research measures becomes a robot rule; a control built in engineering carries over unchanged to the terminal and the robots.

  • Research
  • Robots
  • Engineering
  • L.I.T.

REPRESENTATIVE INTERFACE DATA

  • level_study.ipynb
  • Robot.ex5
  • pipeline · breakers
  • terminal

Handoff

  1. A finding becomes a rule

    If a level's name doesn't change the odds of a break, no robot gets a rule that says 'open a trade at this level.' Only behavior that holds up against a control group makes it into a rule.

    Proof

    Level families ~41%, control ~40%: no difference.

    REPRESENTATIVE INTERFACE DATA

    artifact

    level_study.ipynb → rules.md

  2. A measurement becomes context

    We measured how long gaps take to close. The terminal shows this next to the news as context, never as a trade recommendation.

    Proof

    65% of gaps closed in the opening session, 95% within five trading days and 98% within 60 days.

    REPRESENTATIVE INTERFACE DATA

    artifact

    gap_study → L.I.T. context panel

  3. Controls go inside the robot

    Daily loss limit, stale data and spread checks are added to every robot from the same module. Before a robot starts, the conditions under which it stops are already written down.

    Proof

    The stop rule is written before the start rule.

    REPRESENTATIVE INTERFACE DATA

    artifact

    breakers.yaml → Robot.mq5

  4. Clean data, honest measurement

    Minute prices from two sources are compared, and gaps and synthetic ticks are filtered out. Research runs on this cleaned data.

    Proof

    Minute data, 2020–2026, two pairs.

    REPRESENTATIVE INTERFACE DATA

    artifact

    feed_diff → M1 dataset

  5. Live logs feed back into tests

    The robot's live logs are compared with a replay of the same day in the test environment. If the two diverge, the test infrastructure gets fixed first.

    Proof

    Live and test must produce the same sequence of events.

    REPRESENTATIVE INTERFACE DATA

    artifact

    Experts journal → test suite

  6. The pipeline feeds the terminal

    News, calendar and price feeds pass through the same normalization pipeline. In the terminal, time, source and impact tag read consistently on every card.

    Proof

    One pipeline, one time reference.

    REPRESENTATIVE INTERFACE DATA

    artifact

    news_pipeline → Intelligence Wire

WE HAVE SKIN IN THE GAME.

STANCE

We run the systems we build on our own capital first. Before we use an idea, we measure it and compare the result with a control group under the same conditions. We don't publish result figures; this page describes the method.

  • No control group, no finding (CONTROL)

    Every measurement is compared against matched samples drawn at random under the same conditions. No difference means no finding.

    01

  • An idea that fails on two pairs is shelved (REPLICATION)

    A result on a single pair can be chance. It has to replicate on at least two independent pairs, across different years.

    02

  • Tail-event days get their own inventory (INVENTORY)

    Sudden gaps, liquidity pullbacks and surprise decisions are logged separately, so they don't disappear into the average.

    03

  • The stop rule is written before the start rule (HALT)

    Before a system goes live, the conditions under which it stops are defined in code and in documentation.

    04

Gates an idea has to clear

LET'S TALK MARKETS.

QUANT

Terminal, custom robot, research or infrastructure: tell us what you're working on, and we'll route you to the right team.

ACTIONS

DOCUMENTS

markdown

/en/quant.md

llms.txt

/llms.txt

agents.md

/en/agents.md

Every page is also served as Markdown: add .md to the address or request it with Accept: text/markdown.