EXP-0003
MarketGenome foundation
Frozen multi-timeframe BTCUSDT dataset and 468-feature observation schema. Technical validation completed before the next RL iteration.
COMPLETED475 cols
systematic trading research
ZeroAI Trading
A public research log for causal market data, reinforcement-learning environments, frozen execution economics, preregistered experiments and independent validation.
Current experiment
EXP-0005
BTCUSDT USDⓈ-M Futures
Market features
468
6 causal timeframes
Observation size
472
468 market + 4 state values
Genome rows
3,506,400
1-minute futures timeline
project status
Research before training.
The pipeline is intentionally gated. Training is not authorized until economics, observations, execution, risk and liquidation semantics are frozen and independently validated.
Research pipelineACTIVE
EXP-0005 contractDRAFT / FREEZING
Canonical PPO trainingNO
Live trading authorizationNO
Test-set performance accessLOCKED
01
Raw futures datacontract OHLCV, mark price, funding
PASS02
Causal MarketGenome6 timeframes, no future leakage
PASS03
Observation layer472 float32 values
PASS04
Execution / position enginefunding, risk, next-open close
PASS05
Liquidation guardmaintenance-margin research model
IN PROGRESS06
PPO trainingblocked until preregistration gate
NOexperiment history
Failures stay published.
ZeroAI does not erase unsuccessful experiments. Failed validation is part of the research record.
EXP-0004
Discrete PPO validation
LONG / FLAT / SHORT PPO experiment. All validation candidates became ineligible under risk constraints. Test remained locked.
FAILED VALIDATIONTEST LOCKED
EXP-0005
Futures-native RL stack
BTCUSDT perpetual futures with mark-to-market equity, funding, exact execution friction, deterministic risk semantics and a frozen research liquidation guard.
ACTIVE3x / 25%
validation ledger
PASS means reproducible.
Key artifacts are bound to SHA-256 hashes. Validators independently recompute important economics rather than trusting engine output.
| Layer | Status | Key result | Frozen SHA-256 / reference |
|---|---|---|---|
| Raw futures data | PASS | Contract TF continuity; mark gaps preserved; funding matched | CA2CF43F665A41D28A9757396CF51C80ECBA2257F9D2AE7B47CC89A5E4D7613D |
| Causal futures genome | PASS | 3,506,400 rows · 475 columns · 468 numeric features | 6DF3806EED6BB6F8D6722B6DCA629E1FF68991A363D204E446D18552DCB14100 |
| Dataset + normalization | PASS | TRAIN-only fit · z-score · clip [-5,+5] | 52549EE51695A8AA9C07C7D6FA3DE1A91D47F6B47342625C994D742A21354877 |
| Observation layer | PASS | 472 float32 · no runtime NULL imputation | A6AFFF80B27320FB091F50B4F8877C65EA7C2568B754D01AFBF74B35A6FDB64E |
| Leverage / margin sizing | PASS | 25% margin × 3x = 0.75× equity target notional | 11F5E86CB7FA5F63E11D8A0246B74A2DFB244F28AA596C945046F94529A91AFB |
| Position / execution engine | PASS | 15.0000 bps zero-move RT · funding-before-execution | D1C51953DC6C515CB51B8A1EA6772EBD7AC71F76A6DA2964DEAF008AF7EB8738 |
| BTCUSDT research brackets | PASS | 12 contiguous tiers · public undocumented frontend snapshot | DA6AA3FD2A408BDBBC2C95B8FDBBB1CC62B918724C6704A4C8CD8E91BEC11094 |
| Liquidation engine | PENDING | Conservative research liquidation guard being frozen | not yet validated |
Research limitation: the BTCUSDT maintenance-margin bracket reference comes from a public, undocumented Binance frontend response and is not account-specific. Exact live-account liquidation fidelity is not claimed.
EXP-0005
Current research contract.
The active experiment is futures-native. Contract execution prices, mark-to-market prices, funding and risk accounting have distinct roles.
Market & data
ExchangeBINANCE
InstrumentBTCUSDT PERP
Base timeframe1m
Higher TFs5m 15m 30m 1h 4h
Feature count468
Observation472 float32
Execution economics
ActionsFLAT LONG SHORT
Leverage3x
Margin allocation25%
Target notional0.75× equity
Taker fee5.0 bps / leg
Half-spread0.5 bps / leg
Slippage2.0 bps / leg
Zero-move RT15.0000 bps
Risk & accounting
Equitywallet + unrealized
MTM sourcemark price
Fundinghistorical
Max drawdown10%
Daily loss5%
Risk closenext contract OPEN
Liquidationresearch guard
Live tradingNO
methodology
Built to resist self-deception.
The objective is not to manufacture an attractive backtest. It is to create a research pipeline where economic assumptions and failures remain visible.
[01]
Causal data only
Higher-timeframe values are aligned only after their information is actually available.
[02]
Validation before training
Datasets, observations, execution economics and risk rules are independently checked first.
[03]
Real friction
Fees, spread, slippage and funding are modeled explicitly instead of hidden behind headline PnL.
[04]
Failed runs stay visible
EXP-0004 failed validation and remains part of project history rather than being quietly discarded.