okx-codex-trader/reports/eth-exploration

ETH Exploration Summary

This directory contains three completed ETH exploration worker outputs.

Worker outputs

1. Risk/exit variants

Script:

• scripts/search_eth_risk_exit_variants.py

Reports:

• reports/eth-exploration/eth-risk-exit-summary.md
• reports/eth-exploration/eth-risk-exit-total.csv
• reports/eth-exploration/eth-risk-exit-horizon.csv
• reports/eth-exploration/eth-risk-exit-top10.csv

Core conclusion: negative. The best maker_taker result is the unchanged market baseline (market-t50-r3.0-x45.0-stop-none-mh10000). Tested stop, max-hold, and risk-filter variants did not reduce drawdown while preserving return.

2. BTC regime variants

Script:

• scripts/search_eth_btc_regime_variants.py

Reports:

• reports/eth-exploration/eth-btc-regime-report.md
• reports/eth-exploration/eth-btc-regime-summary.csv
• reports/eth-exploration/eth-btc-regime-total.csv
• reports/eth-exploration/eth-btc-regime-horizon.csv
• reports/eth-exploration/eth-btc-regime-top10.csv

Core conclusion: effective. BTC trend plus momentum gating on the ETH RSI2 entry produced the strongest robust sampled-window result. The best maker_taker candidate was eth-btc-rsi-filter-et50-l3.0-x55.0-bt480-bm240-br0.0, with positive net_ci95_low and improved full-period Calmar versus baseline.

3. Price-TWAP variants

Script:

• scripts/search_eth_price_twap_variants.py

Reports:

• reports/eth-exploration/eth-price-twap-summary.md
• reports/eth-exploration/eth-price-twap-search.csv
• reports/eth-exploration/eth-price-twap-horizons.csv
• reports/eth-exploration/eth-price-twap-top10.csv

Core conclusion: search methodology needs review. The standalone price-TWAP search ranks maker_taker candidates with negative annualized return and Calmar, while the BTC regime report shows strong recent horizon returns for baseline price-TWAP rows. The price-TWAP comparison/search口径 should be reconciled before promoting any price-TWAP result.

Second round followups

Scripts:

• scripts/compare_eth_twap_evaluation.py
• scripts/search_eth_regime_price_twap_variants.py

Reports:

• reports/eth-exploration/eth-twap-evaluation-diff.csv
• reports/eth-exploration/eth-twap-evaluation-diff.md
• reports/eth-exploration/eth-regime-price-twap-summary.csv
• reports/eth-exploration/eth-regime-price-twap-total.csv
• reports/eth-exploration/eth-regime-price-twap-horizon.csv
• reports/eth-exploration/eth-regime-price-twap-top10.csv
• reports/eth-exploration/eth-regime-price-twap-report.md

Core conclusion: negative for promotion. The price-TWAP search versus main-report mismatch came from two direct causes: the earlier search grid did not include the main-report ETH deep parameter set, and it evaluated a 3-year search sample while the main report used a 10-year total evaluation with recent horizons cut from that continuous equity curve.

Reporting rule going forward: evaluate selected strategies with the shared runner over the 10-year continuous cached history, then cut 3y/1y/6m/3m horizons from the same run.

BTC regime plus ETH price-TWAP did not exceed the ETH/BTC RSI filter. Its best maker_taker window robustness and full-period Calmar both remained below the standalone baselines, so the ETH/BTC RSI filter remains the stronger candidate.

Third round followups

Scripts:

• scripts/search_eth_twap_10y_refine.py
• scripts/search_eth_strategy_portfolio_10y.py
• scripts/search_eth_non_rsi_10y.py

Reports:

• reports/eth-exploration/eth-twap-10y-refine-summary.md
• reports/eth-exploration/eth-twap-10y-refine-all.csv
• reports/eth-exploration/eth-twap-10y-refine-horizons.csv
• reports/eth-exploration/eth-twap-10y-refine-top15.csv
• reports/eth-exploration/eth-twap-10y-refine-run.log
• reports/eth-exploration/eth-strategy-portfolio-10y-report.md
• reports/eth-exploration/eth-strategy-portfolio-10y-strategies.csv
• reports/eth-exploration/eth-strategy-portfolio-10y-total.csv
• reports/eth-exploration/eth-strategy-portfolio-10y-top10.csv
• reports/eth-exploration/eth-strategy-portfolio-10y-horizon.csv
• reports/eth-exploration/eth-strategy-portfolio-10y-correlation.csv
• reports/eth-exploration/eth-strategy-portfolio-10y-worst-months.csv
• reports/eth-exploration/eth-strategy-portfolio-10y-equity.csv
• reports/eth-exploration/eth-non-rsi-10y-report.md
• reports/eth-exploration/eth-non-rsi-10y-summary.csv
• reports/eth-exploration/eth-non-rsi-10y-horizon.csv
• reports/eth-exploration/eth-non-rsi-10y-top10.csv

Core conclusion: ETH TWAP refined toward the t60/r5/x45 direction, with rsi2-long-guarded-price-twap-o0.0015-0.0040-0.0070-v3-t60-l5.0-x45.0-sl0.008-mh48 ranking first by eligible maker_taker 3y Calmar. The ETH-focused portfolio is the most stable live-allocation candidate in this round: the best maker_taker portfolio combines eth_btc_rsi_filter with btc_lead_eth_lag_5m, producing lower drawdown than ETH price-TWAP deep. Non-RSI exploration found BB squeeze breakout as the high-return but high-drawdown branch, led by bb-squeeze-l96-bw480-q0.15-sl0.008.

Fourth round followups

Scripts:

• scripts/search_eth_twap_robustness_10y.py
• scripts/search_eth_bb_squeeze_risk_10y.py

Reports:

• reports/eth-exploration/eth-twap-robustness-10y-summary.md
• reports/eth-exploration/eth-twap-robustness-10y-ranked.csv
• reports/eth-exploration/eth-twap-robustness-10y-totals.csv
• reports/eth-exploration/eth-twap-robustness-10y-horizons.csv
• reports/eth-exploration/eth-twap-robustness-10y-rolling.csv
• reports/eth-exploration/eth-bb-squeeze-risk-10y-report.md
• reports/eth-exploration/eth-bb-squeeze-risk-10y-summary.csv
• reports/eth-exploration/eth-bb-squeeze-risk-10y-horizon.csv
• reports/eth-exploration/eth-bb-squeeze-risk-10y-top10-calmar.csv
• reports/eth-exploration/eth-bb-squeeze-risk-10y-top10-worst-month.csv

Core conclusion: ETH robust TWAP recommends t60/r3/x50/sl0.012/offsets 0.003-0.006-0.009/valid4. BB squeeze has a low-drawdown version, but trade count is low and it should be treated cautiously.

Fifth round followups

Scripts:

• scripts/validate_eth_robust_twap_external.py
• scripts/search_eth_robust_twap_fill_slippage.py
• scripts/design_eth_robust_twap_live_plan.py

Reports:

• reports/eth-exploration/eth-robust-twap-validation-summary.md
• reports/eth-exploration/eth-robust-twap-validation-*.csv
• reports/eth-exploration/eth-robust-twap-fill-slippage-summary.md
• reports/eth-exploration/eth-robust-twap-fill-slippage-*.csv
• reports/eth-exploration/eth-robust-twap-live-plan.md
• reports/eth-exploration/eth-robust-twap-live-plan.json

Core conclusion: independent validation is consistent with the selected ETH robust TWAP result. Under more conservative fill assumptions, the strategy is not recommended for small-size live trading yet. The next step is read-only signal and order-intent observation.

Sixth round followups

Scripts:

• scripts/search_eth_twap_conservative_variants.py
• scripts/build_eth_signal_intent_readonly.py

Reports:

• reports/eth-exploration/eth-twap-conservative-summary.md
• reports/eth-exploration/eth-twap-conservative-*.csv
• reports/eth-exploration/eth-signal-intent-readonly.md
• reports/eth-exploration/eth-signal-intent-readonly.json

Core conclusion: under conservative fill assumptions, there are no qualified ETH TWAP candidates. The read-only signal/order-intent prototype is complete and currently has no signal.

Seventh round followups

Scripts:

• scripts/search_eth_twap_taker_entry.py
• scripts/search_eth_focused_portfolio_conservative.py

Reports:

• reports/eth-exploration/eth-twap-taker-entry-summary.md
• reports/eth-exploration/eth-twap-taker-entry-*.csv
• reports/eth-exploration/eth-focused-portfolio-conservative-report.md
• reports/eth-exploration/eth-focused-portfolio-conservative-*.csv

Core conclusion: taker-entry TWAP is negative. No ETH taker-entry TWAP candidate in this grid had positive taker_taker net Calmar across 3y/1y/6m/3m. ETH-focused conservative portfolio has qualified combinations: the lowest-DD qualified maker_taker portfolio has annualized 7.43% with DD 7.28%, and the highest-return qualified maker_taker portfolio has annualized 13.25% with DD 16.13%.

Final decision

Scripts:

• scripts/build_eth_conservative_portfolio_report.py
• scripts/build_eth_focused_portfolio_signal_intent.py

Reports:

• reports/eth-exploration/eth-conservative-portfolio-final.md
• reports/eth-exploration/eth-conservative-portfolio-final.json
• reports/eth-exploration/eth-focused-portfolio-signal-intent.md
• reports/eth-exploration/eth-focused-portfolio-signal-intent.json

Core conclusion: prioritize the ETH-focused conservative portfolio for quasi-live/read-only intent tracking. Standalone ETH TWAP is not approved for live trading under conservative maker-fill assumptions. The current portfolio signal intent has no active signal and produced no order intent.

Freqtrade comparison and live readiness

Scripts:

• scripts/compare_eth_focused_portfolio_freqtrade.py
• scripts/check_eth_focused_portfolio_live_readiness.py

Reports:

• reports/eth-exploration/eth-focused-portfolio-freqtrade-20260429T182815Z.md
• reports/eth-exploration/eth-focused-portfolio-freqtrade-20260429T182815Z.json
• reports/eth-exploration/eth-focused-portfolio-live-readiness.md
• reports/eth-exploration/eth-focused-portfolio-live-readiness.json

Core conclusion: freqtrade is usable as a no-maker-dependent comparison path for the ETH/BTC RSI filter and BTC lead-lag legs, but it does not save time for TWAP execution because maker-fill lifecycle fidelity still needs custom order/fill tracking. Live readiness is not satisfied yet: post_only entry payloads, batch order intent, cancel intent/client support, dedicated state, position protection, scheduler, and append-only logging are still missing.

Execution implementation plans

Scripts:

• scripts/implement_okx_readonly_order_support_plan.py
• scripts/build_freqtrade_eth_informative_skeleton.py

Reports:

• reports/eth-exploration/okx-order-support-plan.md
• reports/eth-exploration/okx-order-support-plan.json
• reports/eth-exploration/freqtrade-eth-skeleton-20260429T183230Z.md
• reports/eth-exploration/freqtrade-eth-skeleton-20260429T183230Z.json

Core conclusion: OKX 实盘支持缺口的最小实现计划已生成；Freqtrade no-maker-dependent 对照骨架已生成。

Execution readonly prototypes

Scripts:

• scripts/run_freqtrade_eth_skeleton_backtest.py
• scripts/prototype_okx_order_support_readonly.py

Reports:

• reports/eth-exploration/freqtrade-eth-skeleton-backtest-20260429T183627Z.md
• reports/eth-exploration/freqtrade-eth-skeleton-backtest-20260429T183627Z.json
• reports/eth-exploration/okx-order-support-readonly-prototype.md
• reports/eth-exploration/okx-order-support-readonly-prototype.json

Core conclusion: Freqtrade runner currently blocks because uv run cannot find freqtrade; fix by exposing Freqtrade through uvx or installing it into the uv environment. OKX payload prototype has 9 tests passed and can be migrated into okx_client.