BUILD RECORD / 07
TIER 2 · QUANT / RESEARCH / FINTECH
FOREX PLATFORM
Foreign-exchange structure is fractal — one causal state ladder from monthly to 3-minute can carry bias, setup and entry without look-ahead.
An institutional-grade quantitative foreign exchange platform for systematic research, session microstructure analysis, risk governance and algorithmic execution: a single causally consistent 7-timeframe structural ladder (1M → 3M) built from Dukascopy M1 bars, five overlapping observation sets arbitrated by net structural R:R, and a predeclared four-gate elimination funnel every candidate must clear on unseen data.
DIAGRAM ABOVE IS ILLUSTRATIVE — IT DEPICTS THE SYSTEM SHAPE, NOT A MEASUREMENT.
OVERVIEW
An institutional-grade quantitative foreign exchange platform for systematic research, session microstructure analysis, risk governance and algorithmic execution: a single causally consistent 7-timeframe structural ladder (1M → 3M) built from Dukascopy M1 bars, five overlapping observation sets arbitrated by net structural R:R, and a predeclared four-gate elimination funnel every candidate must clear on unseen data.
QUANTITATIVE FX RESEARCH FRAMEWORK (PYTHON)
PROBLEM
Single-timeframe strategies ignore macro structure — a lower-timeframe trigger against the higher-timeframe trend is counter-trend noise with a high false-breakout rate.
Parameters chosen against the same history that reports the result produce edges that evaporate on unseen data once partitions and costs are declared after the fact.
APPROACH
- Construct one causally consistent state ladder from monthly to 3-minute over raw M1 vendor bars: each closed bar updates exactly one state, and no forward-looking information enters the model.
- Arbitrate signals from five overlapping three-timeframe sets by net structural R:R, so a single causal portfolio competes with itself instead of five independent strategies.
- Declare chronological 60/20/20 partitions and a four-gate elimination funnel before any parameter is touched — a candidate only advances by clearing every gate on unseen data.
ARCHITECTURE
ARCHITECTURE — DATA PATH
ARCHITECTURE DIAGRAM — THE DOCUMENTED LAYERS, NOT A MEASUREMENT
- 01
M1 VENDOR INTAKE
Dukascopy BID/ASK bars with provenance gates
- OHLC and anomaly checks
- No synthetic fabrication or forward-fill
- 02
CAUSAL LADDER
Seven closed-bar timeframes, no look-ahead
- 1M → 1W → 1D → 4H → 1H → 15M → 3M
- One state update per close
- 03
MARKET MODEL
Strategy-agnostic structure and zones
- Swings · BOS · CHoCH
- Premium / discount dealing ranges
- 04
SET ARBITRATION
Five overlapping views compete by R:R
- SET 1 macro → SET 5 session
- Single winner per movement
- 05
STRATEGY HARNESS
Frozen institutional rider parameters
- Two-stage OTE entry
- ATR-buffered structural stops
- 06
GATE FUNNEL
Predeclared elimination on unseen data
- 60/20/20 chronological split
- Four gates — all must pass
RESULTS
1.03M
M1 BARS / PAIR
APPROX. DUKASCOPY COVERAGE — 2024-01 → 2026-10
7
TIMEFRAME LADDER
CLOSED-BAR DESIGN — 1M DOWN TO 3M
$0.00
LIVE CAPITAL
RESEARCH ONLY — NO STRATEGY AUTHORIZED
All institutional upgrades are implemented and verified: two-stage OTE entry, ATR stop buffering, +2.0R break-even, +3.5R structural trailing, unified session liquidity filter and higher-timeframe correlation filter.
The full 3-year, 4-pair, 5-set batch validation campaign is processing; no gate results are published for it yet.
Live capital is $0.00 — no strategy is authorized for live execution until it clears every predeclared gate on unseen data.
LIMITATIONS
- No published performance results yet: the current validation campaign is still processing, so any expectancy figure would be unsupported.
- Historical coverage is Dukascopy M1 bars for four pairs from 2024-01 onward — other brokers, pairs and periods are not represented.
- Costs (spread, commission, slippage) are modeled assumptions rather than executed fills; live liquidity can exceed modeled tolerances.
TESTING
Unit and integration suites run under pytest from the repository's tests/ directory.
Intake quality gates verify OHLC and quote anomalies, report midweek gaps, and reject unexplained gaps over 5 minutes.
ARTIFACTS
- MIT-licensed repository: causal engine, market model, strategy harness, cost, risk and portfolio engines.
- Batch campaign runner, parameter sweeper and terminal report generation for the 5-set × 4-pair validation.
- Documentation set: architecture, strategy catalog, product roadmap and research outcomes.
CURRENT STATE
ACTIVE VALIDATION
Actively developed — repository public under MIT, last push 13 October 2026.
Fractal structural rider campaign in active validation; results publish through the repository's report and results files.