RESEARCH / FORWARD TESTING

Trading systems.
Nothing hidden.

Every active system keeps its own forward PAPER record. Winning and losing versions remain visible so a new model cannot quietly replace inconvenient history.

ACTIVE SYSTEMS3CLOSED TRADES22EVIDENCEPAPER AUTO UPDATED
LIVE DATA

See what is running now

Start with the current V6 PAPER record and sanitized closed-trade metrics.

Open V6 →
BUILD HISTORY

Understand how we got here

Follow the project from early failed branches through the current architecture.

Read V1 → V6 →
METHODOLOGY

Interpret the evidence

Understand what BACKTEST, PAPER and LIVE results can and cannot establish.

Research method →
PAPER
Simulated executionNo live capital is represented. Small samples and unequal observation windows are not treated as comparable proof.
SYSTEM STATUS

Current forward experiments

Updated 2026-09-24 23:18 UTC
REAL FORWARD DATA

PnL paths, not headline numbers.

CUMULATIVE REALIZED PNL+$2.98
17 closed trades · PAPER
CUMULATIVE REALIZED PNL+$10.42
5 closed trades · PAPER

Latest closed PAPER trades

MARKETSIDEEXITNET PNL
SOLSHORTexit signal+$0.35
ETHLONGroi+$2.00
ETHSHORTexit signal+$2.28
SOLLONGtrailing stop loss+$2.07
ETHLONGexit signal+$3.72

The chart is cumulative realized PAPER PnL from recorded closed trades. It is not a price chart and does not include unrealized open-position PnL.

READ THE DATA CAREFULLY

Same dashboard. Different evidence.

V4 has a larger closed-trade sample than V6, while V4.4 has almost no closed observations yet. A visually higher PnL or win rate therefore does not establish that one system is better.

Open the full V4 / V4.4 / V6 comparison →
01BACKTESTHistorical candidate discovery
02PAPERForward behavior without live capital
03LIVEReal execution and capital risk
EXPERIMENT HISTORY

The record stays when the system changes.

V4Control branch

Forward PAPER reference with the longest current closed-trade record.

V4.4Adaptive revision

Deployed separately so the revised logic cannot overwrite V4 history.

V5.1Shadow ML research

Earlier learning architecture retained as prior evidence after mixed results.

V6FreqAI experiment

Current ML branch. Rolling training and forward PAPER execution on BTC, ETH and SOL.

See version comparison →
BUILD LOG / LAB JOURNAL

The failures and decisions behind the current system.

Results make more sense when the development history is visible. These notes document why branches were created, what failed and why the architecture changed.

RESEARCH NOTES

Understand what can go wrong.