Skip to content
CN2.DEVBUKYA NARESH

PROJECT WORLD / 08

QUANT / FINTECH / ENGINEERING

CRYPTO PLATFORM

Market state should be modeled once, across scales — the authoritative engine owns state, risk and execution, and clients consume its contracts rather than their own calculations.

PAPER OPERATIONALTEST SUITE · 300/300 TESTS

Core engine, risk governor, web workstation and Android client implemented and tested in CI. Live exchange execution is intentionally disabled — real capital is pinned to $0.00.

Python 3.12 · aiohttpSQLite WAL persistenceKotlin · Jetpack ComposeGitHub Actions CI
EXPLORE ARCHITECTUREVIEW EVIDENCEREPOSITORY →

SIGNATURE / STATE PIPELINE

ILLUSTRATIVE — SYSTEM SHAPE, NOT A MEASUREMENT

OVERVIEW

An institutional-grade autonomous cryptocurrency trading platform: a single Python 3.12 backend models market structure across a continuous 7-timeframe ladder and five overlapping observation sets, freezes research under a SHA-256 contract, enforces fail-closed risk governance, and serves identical state contracts to an institutional web workstation and a native Android client over REST and WebSocket.

AUTONOMOUS DIGITAL-ASSET TRADING & RESEARCH PLATFORM (PYTHON)

Python 3.12 · aiohttpSQLite WAL persistenceKotlin · Jetpack ComposeGitHub Actions CI

ARCHITECTURE

SELECT A COMPONENT

COMPONENT

INGESTION

Binance public feeds, timestamp-verified

  • Tick validation against UTC
  • 7-timeframe candle engine

CONNECTED IN THIS RECORD

DATA FLOW — ARCHITECTURE DIAGRAM, NOT A MEASUREMENT

  1. 01

    INGESTION

    Binance public feeds, timestamp-verified

    • Tick validation against UTC
    • 7-timeframe candle engine
  2. 02

    STATE ENGINE

    Swings, breaks, zones and phase

    • CHOCH / BOS · FVG · order blocks
    • 5 overlapping observation sets
  3. 03

    DECISION ENGINE

    Frozen contract, confidence gating

    • Contract SHA-256 8fbc923a…
    • Confidence ≥ 0.50 to proceed
  4. 04

    RISK GOVERNOR

    Pre-trade capital controls

    • ≥ 4.0R destination floor
    • ≤ 1.0% trade · ≤ 3.0% heat
  5. 05

    CAPITAL GATE

    Fail-closed live authorization

    • Real capital $0.00
    • Live order submission disabled
  6. 06

    SERVER & CLIENTS

    One state contract, two clients

    • 21 REST endpoints · WebSocket
    • Web workstation · Android

CASE STUDY

THE PROBLEM

Strategies developed in notebooks are rewritten for execution in another language, introducing unmodeled latency and divergence between research behavior and live behavior.

Single-timeframe strategies ignore macro structural trend and higher-timeframe liquidity pools, which surfaces as false breakouts.

Client-computed state lets two views of the same market disagree about risk.

THE APPROACH

  • Keep 100% of market state generation, risk validation and execution logic in one authoritative Python engine; the web workstation and Android client are stateless projections over shared REST and WebSocket contracts.
  • Model structure once on a continuous closed-bar 7-timeframe ladder and observe it through five overlapping three-timeframe sets — state updates only on closed candles, entries execute at next bar open.
  • Freeze alpha models and risk boundaries under a SHA-256 research contract, validate every candidate against a ≥ 4.0R destination floor with ≤ 1.0% per-trade risk and ≤ 3.0% portfolio heat, and pin real capital to $0.00 behind a fail-closed gate.

RESULTS — WITH PROVENANCE

9,608

CANDIDATE RECORDS

HISTORICAL RESEARCH — PHASE Q.2 BASELINE

+0.8885R

EXPECTANCY · CONF ≥ 0.60

BACKTEST SIMULATION — NOT LIVE PERFORMANCE

4.918

PROFIT FACTOR · CONF ≥ 0.60

BACKTEST SIMULATION — NOT LIVE PERFORMANCE

CI passes 300/300 Python tests with 9,608/9,608 tick-invariant replay confirmed on every run.

Historical research baseline over 9,608 mapped candidate records across BTC, ETH, SOL and BNB (July 2020 onward): the confidence ≥ 0.60 cohort reports +0.8885R expectancy and 4.918 profit factor as backtested figures.

The published Phase Q.2 analysis reports the model pruning 80.18% of baseline losses (3,907 of 4,873).

Paper simulation is operational with 10 bps taker fees, 4 bps slippage and adverse-first intrabar stop collision.

LIMITATIONS

  • All expectancy, profit-factor and win-rate figures are historical backtest simulations; live exchange execution is intentionally disabled and real capital is pinned to $0.00.
  • Evaluations cover high-liquidity assets (BTC, ETH, SOL, BNB) — other pairs may behave differently, and modeled slippage can be exceeded in liquidity gaps.
  • Cloud staging is configured but not provisioned; the Android release APK builds with R8 but is debug-signed, so store distribution is incomplete.

EVIDENCE

TEST SUITE · 300/300 TESTSHOW THE CLAIMS ABOVE ARE CHECKED
  • The full pytest suite — unit, integration and invariant regression — runs in GitHub Actions alongside repository secret scanning.
  • python cli.py verify-contract re-checks the frozen research contract digest (8fbc923a…) on demand.
  • Restart recovery exercises idempotent order reconciliation against SQLite WAL and dual JSON checkpoints.
INSPECT THE REPOSITORY →

DOCUMENTATION

  • MIT-licensed repository: Python engine, aiohttp server, web workstation, native Android client, docs and research reports.
  • CI workflows for Python tests, Android debug/staging/release builds and secret scanning.
  • Append-only decision ledger with SHA-256 event lineage and frozen research contracts.
← ALL PROJECTS↑ WORLD TOP
RESEARCH LIBRARY →GITHUB →