okx-codex-trader/reports/eth-exploration/live-test-strategy-inventory.md

ETH strategy inventory and live-test order

This is a decision memo for what to shadow/live-test first. It does not place orders.

Data boundary

The ETH/BTC 15m aligned local history currently covers about 2019-12-30 to 2026-04-29, roughly 6.3 years. Reports that say 10y are using the full available local history when 10 years is not present.

Current strategy set

Strategy	Execution dependency	Main result	Recent result	Status
nextgen + micro recent_regime_switch	Taker/market-style signal portfolio, synthetic accounting	Full +167.17%, ann. 16.78%, DD 8.03%, Calmar 2.09	1y +29.15%, 6m +5.33%, 3m +6.06%	Best research candidate; shadow first, then smallest real test after live signal/accounting implementation is explicit.
eth-btc-nextgen equal-2-c0003	Taker/market-style synthetic 2-leg portfolio	Full +152.19%, ann. 15.72%, DD 10.29%, Calmar 1.53	1y +25.86%, 6m +2.65%, 3m +6.06%	Validated internally by independent replay; use as baseline.
eth-btc-low-turnover	Direct lower-frequency long-only signal	Taker/taker full +29.64%, ann. 4.19%, DD 3.52%, Calmar 1.19	Taker/taker 1y +3.07%, 6m +1.09%, 3m +0.05%	Best real-money pilot candidate because it is sparse and cost-stressed.
Conservative ETH-focused portfolio	Mixed portfolio; primary version includes maker-dependent TWAP	Lowest-DD variant return +57.47%, ann. 7.43%, DD 7.28%, worst month -4.44%	Previously qualified across horizons	Shadow only until maker fill/miss and slippage are observed.
No-maker conservative portfolio	No maker-dependent TWAP	Return +72.69%, ann. 9.01%, DD 10.46%, worst month -4.51%	Previously qualified across horizons	Secondary baseline; less attractive than nextgen after latest search.
ETH microstructure ATR compression/US	Taker/market-style, ETH-only	Best full candidate +59.4%, ann. 7.62%, DD 39.2%; robust US candidate full weak	Robust candidate 3y/1y/6m/3m all positive but low return	Use only as regime-switch component, not standalone live strategy.
ETH robust price TWAP	Maker/post-only price ladder	Optimistic maker assumptions strong	Conservative fill/slippage stress failed	Do not trade standalone; only shadow maker fill quality.
BB squeeze risk candidate	Taker/market-style	High return, high DD, low robust trade count	Insufficient confidence	Watchlist only.
Regime router	Taker/market-style	Gross captures moves but net after cost fails	Full/3y/1y/6m/3m negative	Reject for live.
Taker-entry TWAP	Taker entry	No qualified candidate	Failed positive-horizon requirement	Reject.

Live-test priority

1. Read-only shadow: nextgen + micro recent_regime_switch

This is the best current research candidate by risk-adjusted metrics. The blocker is not backtest math; it is execution definition. It combines independently compounded sub-strategies and a recent-regime switch, so the next implementation must produce a single explicit order intent stream before any real order.

1. Tiny real pilot candidate: eth-btc-low-turnover

This is the cleanest candidate for real money first because it is direct, sparse, and survives taker_taker=0.0030. With a 100 USDT account, this is the only one I would consider for first real order testing after a read-only signal appears. Expect few trades: about 0.96 trades/month in the best stressed candidate.

1. Read-only baseline: eth-btc-nextgen equal-2-c0003

Keep it running as the benchmark. It is validated and stronger than the conservative portfolio, but it is a synthetic 2-leg daily-equal portfolio, so direct live execution needs a deterministic portfolio-accounting translation.

1. Read-only maker-quality monitor: conservative portfolio with TWAP leg

The strategy is not the first live allocation because maker fill assumptions decide whether it is real. The useful live test is to log post-only ladder fill/miss/slippage without relying on it for PnL.

What to test with real orders first

First real-order test should be eth-btc-low-turnover, not the highest-return strategy. Reason: it has the smallest operational surface, low trade frequency, positive results under taker/taker cost, and no maker-fill dependency.

Minimum real test constraints:

• Only one ETH-USDT-SWAP position at a time.
• Use isolated margin.
• Use the smallest OKX-valid contract size from available balance.
• Do not pyramid.
• No TWAP ladder for the first live test.
• Log the signal snapshot, order response, fill price, fee, realized PnL, and whether it matched the backtest entry/exit candle assumption.

The current eth-btc-nextgen read-only signal report shows no_signal on the latest aligned candle available in local data. No order should be sent until the signal intent script produces a concrete entry intent and the live execution wrapper is reviewed separately.