cyb50-quant/research/dragon/v2/dragon_short_holding_master_review.py

from __future__ import annotations

import json
from dataclasses import asdict
from pathlib import Path

import pandas as pd

from dragon_branch_configs import alpha_first_selective_veto_config


def _load_csv(base_dir: Path, name: str) -> pd.DataFrame:
    return pd.read_csv(base_dir / name, encoding="utf-8-sig")


def _format_pct(value: float) -> str:
    if pd.isna(value):
        return "NA"
    if value == float("inf"):
        return "inf"
    return f"{value:.2%}"


def _format_num(value: float) -> str:
    if pd.isna(value):
        return "NA"
    if value == float("inf"):
        return "inf"
    return f"{value:.2f}"


def main() -> None:
    base_dir = Path(__file__).resolve().parent
    audit = _load_csv(base_dir, "dragon_short_holding_audit.csv")
    pressure = _load_csv(base_dir, "dragon_short_holding_family_pressure.csv")
    experiments = _load_csv(base_dir, "dragon_short_holding_experiments.csv")

    best = experiments[experiments["experiment"] != "baseline_alpha_first"].sort_values(
        ["avg_return", "profit_factor"], ascending=[False, False]
    ).iloc[0]

    winner_config = alpha_first_selective_veto_config().with_updates(
        glued_selective_hot_c1_min=40.0,
        glued_selective_hot_b1_min=0.10,
        glued_selective_low_c1_min=23.0,
        glued_selective_low_c1_max=28.0,
        glued_selective_low_b1_max=0.02,
    )
    snapshot = asdict(winner_config)
    snapshot["disabled_rules"] = sorted(winner_config.disabled_rules)
    (base_dir / "dragon_short_holding_candidate_config.json").write_text(
        json.dumps(snapshot, indent=2, ensure_ascii=False) + "\n",
        encoding="utf-8",
    )

    root_summary = (
        audit.groupby("failure_root")
        .agg(trades=("buy_date", "count"), avg_return=("return_pct", "mean"))
        .reset_index()
        .sort_values("trades", ascending=False)
    )
    top_entry_drag = pressure[pressure["group_type"] == "entry_family"].sort_values("drag_score", ascending=False).head(5)
    top_path_drag = pressure[pressure["group_type"] == "path_combo"].sort_values("drag_score", ascending=False).head(5)

    lines = [
        "# Dragon Short Holding Master Review",
        "",
        "- Branch under audit: `alpha_first_selective_veto`.",
        "- Goal: identify the dominant short-holding drag and the next narrow optimization target.",
        "",
        "## Audit Conclusions",
        f"- audited short trades: `{int(len(audit))}`",
    ]
    for _, row in root_summary.iterrows():
        lines.append(
            f"- failure_root `{row['failure_root']}`: trades `{int(row['trades'])}`, avg_return `{_format_pct(float(row['avg_return']))}`"
        )

    lines.extend(["", "## Lead Drag Families"])
    for _, row in top_entry_drag.iterrows():
        lines.append(
            f"- `{row['holding_bucket']} / {row['entry_family']}`: trades `{int(row['trades'])}`, "
            f"avg_return `{_format_pct(float(row['avg_return']))}`, drag_score `{row['drag_score']:.4f}`"
        )

    lines.extend(["", "## Lead Drag Paths"])
    for _, row in top_path_drag.iterrows():
        lines.append(
            f"- `{row['holding_bucket']} / {row['entry_family']} -> {row['sell_reason']}`: trades `{int(row['trades'])}`, "
            f"avg_return `{_format_pct(float(row['avg_return']))}`, drag_score `{row['drag_score']:.4f}`"
        )

    lines.extend(
        [
            "",
            "## Experiment Winner",
            f"- best branch: `{best['experiment']}`",
            f"- trades: `{int(best['trades'])}`",
            f"- avg_return: `{_format_pct(float(best['avg_return']))}`",
            f"- profit_factor: `{_format_num(float(best['profit_factor']))}`",
            f"- short_avg_return: `{_format_pct(float(best['short_avg_return']))}`",
            f"- `00-05d`: `{_format_pct(float(best['short_00_05d_avg_return']))}`",
            f"- `06-10d`: `{_format_pct(float(best['short_06_10d_avg_return']))}`",
            f"- real BUY / SELL overlap: `{int(best['real_buy_overlap'])}/{int(best['real_sell_overlap'])}`",
            "",
            "## Interpretation",
            "- `post_sell_rebound_buy` is not the main short-holding problem in this pack; disabling it hurts or adds little value.",
            "- The dominant short-holding drag is `glued_buy`, especially in mid-regime short exits.",
            "- The winning branch confirms that narrow glued-entry veto is more valuable than attacking `post_sell_rebound_buy` first.",
            "- The most useful next alpha-first direction is now a glued-focused selective-veto branch, not a post-sell-rebound branch.",
            "",
            "## Candidate Config",
            "- Snapshot file: `dragon_short_holding_candidate_config.json`.",
            "- This candidate should remain alpha-first research only until a branch-level governance decision upgrades it.",
        ]
    )

    (base_dir / "dragon_short_holding_master_review.md").write_text("\n".join(lines) + "\n", encoding="utf-8")


if __name__ == "__main__":
    main()