okx-codex-trader/scripts/search_eth_nextgen_micro_portfolio.py

from __future__ import annotations

import argparse
import json
import sys
from dataclasses import dataclass
from pathlib import Path

import pandas as pd

sys.path.insert(0, str(Path(__file__).resolve().parents[1]))

from scripts import explore_ultrashort as explore
from scripts import search_eth_btc_nextgen_variants as nextgen
from scripts import search_eth_microstructure_variants as micro


OUTPUT_DIR = Path("reports/eth-exploration")
PREFIX = "eth-nextgen-micro-portfolio"
PRIMARY_COST = "maker_taker"
COST_MODELS = {
    "maker_taker": 0.0021,
    "taker_taker": 0.0030,
}
NEXTGEN_BASELINE = "equal-2-c0003"
NEXTGEN_LEGS = (
    "btc_trend_eth_rsi:15m:eth-btc-rsi-filter-et50-l3.0-x55.0-bt480-bm240-br0.0",
    "btc_shock_guard_eth_rsi:15m:eth-btc-shock-filter-et50-l3.0-x55.0-bt480-bm240-br0.01-sw96-sv0.01-sd0.05",
)
MICRO_NAMES = (
    "atr-compress-expand-r48-q0.15-sl0.008-tp0.016-mf0.25-us",
    "atr-compress-expand-r48-q0.15-sl0.008-tp0.016-mf0.4-us",
    "atr-compress-expand-r96-q0.15-sl0.008-tp0.016-mf0.25-us",
    "atr-compress-expand-r96-q0.15-sl0.008-tp0.016-mf0.4-us",
)
HORIZONS = (
    ("full", None),
    ("3y", pd.DateOffset(years=3)),
    ("1y", pd.DateOffset(years=1)),
    ("6m", pd.DateOffset(months=6)),
    ("3m", pd.DateOffset(months=3)),
)


@dataclass(frozen=True)
class LegReturn:
    source: str
    exit_date: pd.Timestamp
    value: float


def daily_equity_from_frame(frame: pd.DataFrame, index: pd.DatetimeIndex) -> pd.Series:
    series = frame.set_index("ts")["equity"].sort_index()
    daily = series.reindex(index.union(series.index)).sort_index().ffill().reindex(index).ffill()
    daily.iloc[0] = explore.INITIAL_EQUITY
    return daily


def metrics_from_daily(series: pd.Series) -> dict[str, float]:
    years = (series.index[-1] - series.index[0]).total_seconds() / 86_400 / 365
    total = float(series.iloc[-1] / series.iloc[0] - 1.0)
    annualized = (1.0 + total) ** (1.0 / years) - 1.0 if total > -1.0 and years > 0.0 else 0.0
    drawdown = explore.max_drawdown_from_equity([float(value) for value in series])
    returns = series.pct_change().dropna()
    daily_std = float(returns.std(ddof=1)) if len(returns) > 1 else 0.0
    risk_reward = float(returns.mean()) / daily_std * (365**0.5) if daily_std else 0.0
    return {
        "net_total_return": total,
        "net_annualized_return": annualized,
        "net_max_drawdown": drawdown,
        "net_calmar": annualized / drawdown if drawdown else 0.0,
        "risk_reward_ratio": risk_reward,
    }


def monthly_rows(name: str, cost_model: str, series: pd.Series) -> pd.DataFrame:
    monthly = series.resample("ME").last()
    frame = pd.DataFrame(
        {
            "name": name,
            "cost_model": cost_model,
            "month": monthly.index.strftime("%Y-%m"),
            "start_equity": monthly.shift(1).fillna(series.iloc[0]).to_numpy(),
            "end_equity": monthly.to_numpy(),
        }
    )
    frame["return"] = frame["end_equity"] / frame["start_equity"] - 1.0
    return frame


def horizon_rows(name: str, cost_model: str, series: pd.Series, trades: list[LegReturn], monthly: pd.DataFrame) -> list[dict[str, object]]:
    end = series.index[-1]
    rows: list[dict[str, object]] = []
    for label, offset in HORIZONS:
        horizon = series if offset is None else series[series.index >= end - offset]
        if len(horizon) < 2:
            horizon = series
        start = horizon.index[0]
        horizon_trades = [trade for trade in trades if trade.exit_date >= start]
        horizon_monthly = monthly[monthly["month"] >= start.strftime("%Y-%m")]
        worst = horizon_monthly.sort_values("return").iloc[0] if len(horizon_monthly) else None
        rows.append(
            {
                "name": name,
                "cost_model": cost_model,
                "horizon": label,
                "horizon_start": horizon.index[0].strftime("%Y-%m-%d"),
                "horizon_end": horizon.index[-1].strftime("%Y-%m-%d"),
                "worst_month": "" if worst is None else str(worst["month"]),
                "worst_month_return": 0.0 if worst is None else float(worst["return"]),
                **trade_return_stats(horizon_trades),
                **metrics_from_daily(horizon),
            }
        )
    return rows


def returns_to_equity(name: str, returns: pd.Series) -> pd.Series:
    equity = explore.INITIAL_EQUITY * (1.0 + returns.fillna(0.0)).cumprod()
    equity.iloc[0] = explore.INITIAL_EQUITY
    equity.name = name
    return equity


def trade_return_stats(returns: list[LegReturn]) -> dict[str, float]:
    values = [row.value for row in returns]
    wins = [value for value in values if value > 0.0]
    losses = [value for value in values if value < 0.0]
    avg_win = sum(wins) / len(wins) if wins else 0.0
    avg_loss_abs = abs(sum(losses) / len(losses)) if losses else 0.0
    gross_profit = sum(wins)
    gross_loss_abs = abs(sum(losses))
    return {
        "trades": len(values),
        "win_rate": len(wins) / len(values) if values else 0.0,
        "payoff_ratio": avg_win / avg_loss_abs if avg_loss_abs else 0.0,
        "profit_factor": gross_profit / gross_loss_abs if gross_loss_abs else 0.0,
    }


def nextgen_trade_returns(result: object, weight: float, roundtrip_cost: float) -> list[LegReturn]:
    rows: list[LegReturn] = []
    for trade in result.trades:
        value = (float(trade["return_pct"]) / 100.0 - roundtrip_cost * float(trade.get("cost_weight", 1.0))) * weight
        rows.append(LegReturn("nextgen", pd.to_datetime(str(trade["exit_time"]), utc=True).normalize(), value))
    return rows


def micro_trade_returns(result: micro.SegmentResult, weight: float, roundtrip_cost: float) -> list[LegReturn]:
    rows: list[LegReturn] = []
    for trade in result.trades:
        value = (float(trade["return_pct"]) / 100.0 - roundtrip_cost * float(trade["cost_weight"])) * weight
        rows.append(LegReturn("micro", pd.to_datetime(str(trade["exit_time"]), utc=True).normalize(), value))
    return rows


def combine_trade_returns(
    *,
    nextgen_returns: list[LegReturn],
    micro_returns: list[LegReturn],
    nextgen_weight: float,
    micro_weight: float,
    micro_mask: pd.Series | None = None,
) -> list[LegReturn]:
    values = [LegReturn(row.source, row.exit_date, row.value * nextgen_weight) for row in nextgen_returns]
    for row in micro_returns:
        if micro_mask is None or bool(micro_mask.reindex([row.exit_date]).fillna(False).iloc[0]):
            values.append(LegReturn(row.source, row.exit_date, row.value * micro_weight))
    return values


def load_nextgen(index: pd.DatetimeIndex, roundtrip_cost: float) -> tuple[pd.Series, list[LegReturn]]:
    strategies = {
        f"{strategy.family}:{strategy.bar}:{strategy.candidate.name}": strategy
        for strategy in nextgen.build_strategies()
    }
    missing = [key for key in NEXTGEN_LEGS if key not in strategies]
    if missing:
        raise KeyError(f"missing nextgen legs: {missing}")
    data = {
        (symbol, "15m"): nextgen.load_candles(symbol, "15m", 20.0)
        for symbol in ("ETH-USDT-SWAP", "BTC-USDT-SWAP")
    }
    leg_series: list[pd.Series] = []
    trade_returns: list[LegReturn] = []
    for key in NEXTGEN_LEGS:
        result = nextgen.run_strategy(strategies[key], data)
        frame = explore.cost_adjusted_trade_equity_frame(result, roundtrip_cost)
        leg_series.append(daily_equity_from_frame(frame, index))
        trade_returns.extend(nextgen_trade_returns(result, 0.5, roundtrip_cost))
    returns = pd.DataFrame({key: series.pct_change().fillna(0.0) for key, series in zip(NEXTGEN_LEGS, leg_series)}).mean(axis=1)
    return returns_to_equity(NEXTGEN_BASELINE, returns), trade_returns


def load_micro_candidates(index: pd.DatetimeIndex, roundtrip_cost: float) -> dict[str, tuple[pd.Series, list[LegReturn]]]:
    candles = micro._load_candles(micro.SYMBOL, micro.BAR)
    requested = int(20.0 * 365 * 24 * 60 / 15)
    candles = candles[-requested:]
    variants = {variant.name: variant for variant in micro.build_variants()}
    missing = [name for name in MICRO_NAMES if name not in variants]
    if missing:
        raise KeyError(f"missing micro variants: {missing}")
    output: dict[str, tuple[pd.Series, list[LegReturn]]] = {}
    for name in MICRO_NAMES:
        result = variants[name].run(candles)
        frame = micro.cost_equity_frame(result, roundtrip_cost)
        output[name] = (daily_equity_from_frame(frame, index), micro_trade_returns(result, 1.0, roundtrip_cost))
    return output


def evaluate_portfolio(
    *,
    name: str,
    cost_model: str,
    roundtrip_cost: float,
    kind: str,
    series: pd.Series,
    nextgen_weight: float,
    micro_weight: float,
    micro_name: str,
    trade_returns: list[LegReturn],
) -> tuple[dict[str, object], list[dict[str, object]], pd.DataFrame]:
    monthly = monthly_rows(name, cost_model, series)
    worst = monthly.sort_values("return").iloc[0]
    horizons = horizon_rows(name, cost_model, series, trade_returns, monthly)
    min_recent = min(float(row["net_total_return"]) for row in horizons if row["horizon"] != "full")
    row = {
        "name": name,
        "cost_model": cost_model,
        "roundtrip_cost_on_margin": roundtrip_cost,
        "kind": kind,
        "micro_name": micro_name,
        "nextgen_weight": nextgen_weight,
        "micro_weight": micro_weight,
        "worst_month": str(worst["month"]),
        "worst_month_return": float(worst["return"]),
        "min_recent_total_return": min_recent,
        **trade_return_stats(trade_returns),
        **metrics_from_daily(series),
    }
    return row, horizons, monthly


def build_portfolios(
    *,
    cost_model: str,
    roundtrip_cost: float,
    nextgen_series: pd.Series,
    nextgen_trade_returns: list[LegReturn],
    micro_candidates: dict[str, tuple[pd.Series, list[LegReturn]]],
) -> tuple[pd.DataFrame, pd.DataFrame, pd.DataFrame, pd.DataFrame]:
    rows: list[dict[str, object]] = []
    horizon_output: list[dict[str, object]] = []
    monthly_frames: list[pd.DataFrame] = []
    equity_frames: list[pd.DataFrame] = []
    nextgen_returns = nextgen_series.pct_change().fillna(0.0)
    baseline_row, baseline_horizons, baseline_monthly = evaluate_portfolio(
        name=NEXTGEN_BASELINE,
        cost_model=cost_model,
        roundtrip_cost=roundtrip_cost,
        kind="baseline",
        series=nextgen_series,
        nextgen_weight=1.0,
        micro_weight=0.0,
        micro_name="",
        trade_returns=nextgen_trade_returns,
    )
    rows.append(baseline_row)
    horizon_output.extend(baseline_horizons)
    monthly_frames.append(baseline_monthly)
    equity_frames.append(equity_frame(NEXTGEN_BASELINE, cost_model, nextgen_series))

    for micro_name, (micro_series, micro_trade_rows) in micro_candidates.items():
        micro_returns = micro_series.pct_change().fillna(0.0)
        for micro_weight in (0.10, 0.15, 0.20, 0.25, 0.30):
            nextgen_weight = 1.0 - micro_weight
            returns = nextgen_returns * nextgen_weight + micro_returns * micro_weight
            name = f"blend-ng{nextgen_weight:.2f}-{short_micro_name(micro_name)}"
            series = returns_to_equity(name, returns)
            trades = combine_trade_returns(
                nextgen_returns=nextgen_trade_returns,
                micro_returns=micro_trade_rows,
                nextgen_weight=nextgen_weight,
                micro_weight=micro_weight,
            )
            add_result(rows, horizon_output, monthly_frames, equity_frames, name, cost_model, roundtrip_cost, "equity_blend", micro_name, nextgen_weight, micro_weight, series, trades)

        flat_mask = nextgen_returns.abs() < 1e-12
        for micro_weight in (0.25, 0.40):
            returns = nextgen_returns + micro_returns.where(flat_mask, 0.0) * micro_weight
            name = f"nonoverlap-m{micro_weight:.2f}-{short_micro_name(micro_name)}"
            series = returns_to_equity(name, returns)
            trades = combine_trade_returns(
                nextgen_returns=nextgen_trade_returns,
                micro_returns=micro_trade_rows,
                nextgen_weight=1.0,
                micro_weight=micro_weight,
                micro_mask=flat_mask,
            )
            add_result(rows, horizon_output, monthly_frames, equity_frames, name, cost_model, roundtrip_cost, "signal_non_overlap", micro_name, 1.0, micro_weight, series, trades)

        for lookback in (30, 60, 90, 120):
            nextgen_regime = nextgen_series / nextgen_series.shift(lookback) - 1.0
            micro_regime = micro_series / micro_series.shift(lookback) - 1.0
            active = ((nextgen_regime < 0.0) & (micro_regime > 0.0)).shift(1).fillna(False).astype(bool)
            switch_returns = nextgen_returns.where(~active, micro_returns)
            switch_name = f"switch-l{lookback}-{short_micro_name(micro_name)}"
            switch_series = returns_to_equity(switch_name, switch_returns)
            switch_trades = combine_trade_returns(
                nextgen_returns=[row for row in nextgen_trade_returns if not bool(active.reindex([row.exit_date]).fillna(False).iloc[0])],
                micro_returns=micro_trade_rows,
                nextgen_weight=1.0,
                micro_weight=1.0,
                micro_mask=active,
            )
            add_result(rows, horizon_output, monthly_frames, equity_frames, switch_name, cost_model, roundtrip_cost, "recent_regime_switch", micro_name, 1.0, 1.0, switch_series, switch_trades)

            for micro_weight in (0.25, 0.40):
                overlay_returns = nextgen_returns + micro_returns.where(active, 0.0) * micro_weight
                overlay_name = f"riskoff-overlay-l{lookback}-m{micro_weight:.2f}-{short_micro_name(micro_name)}"
                overlay_series = returns_to_equity(overlay_name, overlay_returns)
                overlay_trades = combine_trade_returns(
                    nextgen_returns=nextgen_trade_returns,
                    micro_returns=micro_trade_rows,
                    nextgen_weight=1.0,
                    micro_weight=micro_weight,
                    micro_mask=active,
                )
                add_result(
                    rows,
                    horizon_output,
                    monthly_frames,
                    equity_frames,
                    overlay_name,
                    cost_model,
                    roundtrip_cost,
                    "riskoff_overlay",
                    micro_name,
                    1.0,
                    micro_weight,
                    overlay_series,
                    overlay_trades,
                )

    summary = pd.DataFrame(rows).sort_values(
        ["cost_model", "net_calmar", "net_annualized_return", "min_recent_total_return", "worst_month_return"],
        ascending=[True, False, False, False, False],
    )
    horizons = pd.DataFrame(horizon_output)
    horizons["horizon"] = pd.Categorical(horizons["horizon"], categories=[label for label, _ in HORIZONS], ordered=True)
    horizons = horizons.sort_values(["cost_model", "name", "horizon"])
    monthly = pd.concat(monthly_frames, ignore_index=True)
    equity = pd.concat(equity_frames, ignore_index=True)
    return summary, horizons, monthly, equity


def add_result(
    rows: list[dict[str, object]],
    horizon_output: list[dict[str, object]],
    monthly_frames: list[pd.DataFrame],
    equity_frames: list[pd.DataFrame],
    name: str,
    cost_model: str,
    roundtrip_cost: float,
    kind: str,
    micro_name: str,
    nextgen_weight: float,
    micro_weight: float,
    series: pd.Series,
    trades: list[LegReturn],
) -> None:
    row, horizons, monthly = evaluate_portfolio(
        name=name,
        cost_model=cost_model,
        roundtrip_cost=roundtrip_cost,
        kind=kind,
        series=series,
        nextgen_weight=nextgen_weight,
        micro_weight=micro_weight,
        micro_name=micro_name,
        trade_returns=trades,
    )
    rows.append(row)
    horizon_output.extend(horizons)
    monthly_frames.append(monthly)
    equity_frames.append(equity_frame(name, cost_model, series))


def equity_frame(name: str, cost_model: str, series: pd.Series) -> pd.DataFrame:
    return pd.DataFrame({"name": name, "cost_model": cost_model, "date": series.index.strftime("%Y-%m-%d"), "equity": series.to_numpy()})


def short_micro_name(name: str) -> str:
    return name.replace("atr-compress-expand-", "").replace("sl0.008-tp0.016-", "").replace("-", "_")


def markdown_table(frame: pd.DataFrame) -> str:
    rows = [list(frame.columns), ["---" for _ in frame.columns]]
    rows.extend(frame.astype(object).where(pd.notna(frame), "").values.tolist())
    return "\n".join("| " + " | ".join(format_cell(value) for value in row) + " |" for row in rows)


def format_cell(value: object) -> str:
    if isinstance(value, float):
        return f"{value:.6g}"
    return str(value).replace("|", "\\|")


def monthly_stability(monthly: pd.DataFrame) -> pd.DataFrame:
    return (
        monthly.groupby(["cost_model", "name"], as_index=False)
        .agg(
            months=("return", "count"),
            positive_month_rate=("return", lambda values: float((values > 0.0).mean())),
            avg_month_return=("return", "mean"),
            median_month_return=("return", "median"),
            worst_month_return=("return", "min"),
            best_month_return=("return", "max"),
        )
        .sort_values(["cost_model", "positive_month_rate", "worst_month_return"], ascending=[True, False, False])
    )


def robust_survivors(horizons: pd.DataFrame) -> pd.DataFrame:
    full = horizons[horizons["horizon"] == "full"].copy()
    recent = full
    for horizon in ("3y", "1y", "6m", "3m"):
        part = horizons[horizons["horizon"] == horizon][["cost_model", "name", "net_total_return", "net_calmar"]].rename(
            columns={"net_total_return": f"ret_{horizon}", "net_calmar": f"calmar_{horizon}"}
        )
        recent = recent.merge(part, on=["cost_model", "name"], how="inner")
    return recent[
        (recent["net_total_return"] > 0.0)
        & (recent["net_calmar"] > 0.0)
        & (recent["ret_3y"] > 0.0)
        & (recent["ret_1y"] > 0.0)
        & (recent["ret_6m"] > 0.0)
        & (recent["ret_3m"] > 0.0)
        & (recent["calmar_3y"] > 0.0)
        & (recent["calmar_1y"] > 0.0)
        & (recent["calmar_6m"] > 0.0)
        & (recent["calmar_3m"] > 0.0)
    ].copy()


def write_report(
    *,
    command: str,
    output_files: list[Path],
    summary: pd.DataFrame,
    horizons: pd.DataFrame,
    stability: pd.DataFrame,
    worst_months: pd.DataFrame,
    survivors: pd.DataFrame,
) -> str:
    primary_summary = summary[summary["cost_model"] == PRIMARY_COST]
    stress_summary = summary[summary["cost_model"] == "taker_taker"]
    baseline = primary_summary[primary_summary["name"] == NEXTGEN_BASELINE].iloc[0]
    best = primary_summary.iloc[0]
    best_horizons = horizons[(horizons["cost_model"] == PRIMARY_COST) & (horizons["name"] == best["name"])]
    baseline_horizons = horizons[(horizons["cost_model"] == PRIMARY_COST) & (horizons["name"] == NEXTGEN_BASELINE)]
    dilution = (
        "The best combination improves risk-adjusted return versus nextgen equal-2-c0003."
        if float(best["net_calmar"]) > float(baseline["net_calmar"]) and float(best["net_total_return"]) >= float(baseline["net_total_return"]) * 0.98
        else "The best combination is mainly return dilution unless its lower drawdown or better worst month is preferred."
    )
    lines = [
        "# ETH nextgen + microstructure portfolio exploration",
        "",
        f"Run command: `{command}`",
        "",
        "No order placement or exchange API path is used; this script reads local candle/report data only.",
        "",
        "Output files:",
        *[f"- `{path}`" for path in output_files],
        "",
        "Base nextgen: `equal-2-c0003` with the two documented maker_taker legs.",
        "Micro candidates: ATR compression/expansion US-session robust candidates only.",
        "Costs: maker_taker=0.0021 and taker_taker=0.0030 roundtrip on margin. Funding and slippage remain excluded.",
        "",
        "## Conclusion",
        "",
        dilution,
        f"Strict robust survivors with positive full/3y/1y/6m/3m net return and Calmar: {len(survivors)}.",
        "",
        "## Top maker_taker combinations",
        "",
        markdown_table(
            primary_summary.head(12)[
                [
                    "name",
                    "cost_model",
                    "kind",
                    "micro_name",
                    "net_total_return",
                    "net_annualized_return",
                    "net_max_drawdown",
                    "net_calmar",
                    "risk_reward_ratio",
                    "worst_month",
                    "worst_month_return",
                    "min_recent_total_return",
                    "trades",
                    "win_rate",
                    "payoff_ratio",
                    "profit_factor",
                ]
            ]
        ),
        "",
        "## Taker/taker stress combinations",
        "",
        markdown_table(
            stress_summary.head(12)[
                [
                    "name",
                    "cost_model",
                    "kind",
                    "micro_name",
                    "net_total_return",
                    "net_annualized_return",
                    "net_max_drawdown",
                    "net_calmar",
                    "risk_reward_ratio",
                    "worst_month",
                    "worst_month_return",
                    "min_recent_total_return",
                    "trades",
                ]
            ]
        ),
        "",
        "## Strict robust survivors",
        "",
        markdown_table(
            survivors.head(20)[
                [
                    "name",
                    "cost_model",
                    "net_total_return",
                    "net_calmar",
                    "ret_3y",
                    "ret_1y",
                    "ret_6m",
                    "ret_3m",
                    "calmar_3y",
                    "calmar_1y",
                    "calmar_6m",
                    "calmar_3m",
                ]
            ]
        ),
        "",
        "## Best horizon metrics",
        "",
        markdown_table(
            best_horizons[
                [
                    "horizon",
                    "horizon_start",
                    "horizon_end",
                    "net_total_return",
                    "net_annualized_return",
                    "net_max_drawdown",
                    "net_calmar",
                    "trades",
                    "win_rate",
                    "payoff_ratio",
                    "profit_factor",
                    "risk_reward_ratio",
                    "worst_month",
                    "worst_month_return",
                ]
            ]
        ),
        "",
        "## Baseline horizon metrics",
        "",
        markdown_table(
            baseline_horizons[
                [
                    "horizon",
                    "horizon_start",
                    "horizon_end",
                    "net_total_return",
                    "net_annualized_return",
                    "net_max_drawdown",
                    "net_calmar",
                    "trades",
                    "win_rate",
                    "payoff_ratio",
                    "profit_factor",
                    "risk_reward_ratio",
                    "worst_month",
                    "worst_month_return",
                ]
            ]
        ),
        "",
        "## Monthly stability",
        "",
        markdown_table(stability[stability["cost_model"] == PRIMARY_COST].head(20)),
        "",
        "## Worst months",
        "",
        markdown_table(worst_months.head(20)[["name", "cost_model", "month", "return"]]),
    ]
    return "\n".join(lines) + "\n"


def main() -> int:
    parser = argparse.ArgumentParser()
    parser.add_argument("--output-dir", type=Path, default=OUTPUT_DIR)
    args = parser.parse_args()

    existing_equity = pd.read_csv(args.output_dir / "eth-btc-nextgen-equity.csv")
    base = existing_equity[(existing_equity["cost_model"] == PRIMARY_COST) & (existing_equity["name"] == NEXTGEN_BASELINE)].copy()
    if base.empty:
        raise KeyError(f"missing existing nextgen equity for {NEXTGEN_BASELINE}")
    index = pd.DatetimeIndex(pd.to_datetime(base["date"], utc=True))

    summary_frames: list[pd.DataFrame] = []
    horizon_frames: list[pd.DataFrame] = []
    monthly_frames: list[pd.DataFrame] = []
    equity_frames: list[pd.DataFrame] = []
    for cost_model, roundtrip_cost in COST_MODELS.items():
        nextgen_series, nextgen_returns = load_nextgen(index, roundtrip_cost)
        micro_candidates = load_micro_candidates(index, roundtrip_cost)
        summary, horizons, monthly, equity = build_portfolios(
            cost_model=cost_model,
            roundtrip_cost=roundtrip_cost,
            nextgen_series=nextgen_series,
            nextgen_trade_returns=nextgen_returns,
            micro_candidates=micro_candidates,
        )
        summary_frames.append(summary)
        horizon_frames.append(horizons)
        monthly_frames.append(monthly)
        equity_frames.append(equity)

    summary = pd.concat(summary_frames, ignore_index=True)
    primary = summary[summary["cost_model"] == PRIMARY_COST]
    stress = summary[summary["cost_model"] != PRIMARY_COST]
    summary = pd.concat([primary, stress], ignore_index=True)
    top_pairs = set(zip(primary.head(25)["cost_model"], primary.head(25)["name"]))
    top_pairs.update(zip(stress.head(25)["cost_model"], stress.head(25)["name"]))
    top_pairs.add((PRIMARY_COST, NEXTGEN_BASELINE))
    for cost_model in COST_MODELS:
        top_pairs.add((cost_model, NEXTGEN_BASELINE))

    horizons = pd.concat(horizon_frames, ignore_index=True)
    monthly = pd.concat(monthly_frames, ignore_index=True)
    equity = pd.concat(equity_frames, ignore_index=True)
    survivors = robust_survivors(horizons).merge(
        summary[
            [
                "cost_model",
                "name",
                "kind",
                "micro_name",
                "nextgen_weight",
                "micro_weight",
                "roundtrip_cost_on_margin",
            ]
        ],
        on=["cost_model", "name"],
        how="left",
    ).sort_values(["cost_model", "net_calmar", "net_total_return"], ascending=[True, False, False])
    stability = monthly_stability(monthly)
    worst_months = monthly.sort_values("return").head(100)
    horizons = horizons[horizons.apply(lambda row: (row["cost_model"], row["name"]) in top_pairs, axis=1)]
    monthly = monthly[monthly.apply(lambda row: (row["cost_model"], row["name"]) in top_pairs, axis=1)]
    equity = equity[equity.apply(lambda row: (row["cost_model"], row["name"]) in top_pairs, axis=1)]

    args.output_dir.mkdir(parents=True, exist_ok=True)
    summary_path = args.output_dir / f"{PREFIX}-summary.csv"
    horizon_path = args.output_dir / f"{PREFIX}-horizons.csv"
    monthly_path = args.output_dir / f"{PREFIX}-monthly.csv"
    stability_path = args.output_dir / f"{PREFIX}-monthly-stability.csv"
    worst_path = args.output_dir / f"{PREFIX}-worst-months.csv"
    robust_path = args.output_dir / f"{PREFIX}-robust-survivors.csv"
    equity_path = args.output_dir / f"{PREFIX}-equity.csv"
    json_path = args.output_dir / f"{PREFIX}-top.json"
    report_path = args.output_dir / f"{PREFIX}-report.md"
    output_files = [summary_path, horizon_path, monthly_path, stability_path, worst_path, robust_path, equity_path, json_path, report_path]

    summary.to_csv(summary_path, index=False)
    horizons.to_csv(horizon_path, index=False)
    monthly.to_csv(monthly_path, index=False)
    stability.to_csv(stability_path, index=False)
    worst_months.to_csv(worst_path, index=False)
    survivors.to_csv(robust_path, index=False)
    equity.to_csv(equity_path, index=False)
    json_path.write_text(json.dumps(primary.head(20).to_dict("records"), indent=2), encoding="utf-8")
    command = f"rtk .venv/bin/python {Path(__file__).as_posix()}"
    report_path.write_text(
        write_report(
            command=command,
            output_files=output_files,
            summary=summary,
            horizons=horizons,
            stability=stability,
            worst_months=worst_months,
            survivors=survivors,
        ),
        encoding="utf-8",
    )
    print(primary.head(12).to_string(index=False))
    return 0


if __name__ == "__main__":
    raise SystemExit(main())