"""Replication entry point for F6-conf-02. Run from workspace root:
    ~/wp29/venv/bin/python out/analysis.py
Reads only data/gc5y_daily.csv; writes out/results.json, out/figure1.png,
out/table1.csv, out/exclusions.csv. Deterministic (no randomness used)."""
import json, os
import numpy as np
import pandas as pd
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
from scipy import stats

OUT = "out"
FN = "gc5y_daily.csv"
PRE_WINDOW = 120
os.makedirs(OUT, exist_ok=True)

raw = pd.read_csv(os.path.join("data", FN), dtype=str)
n_raw = len(raw)
df = raw.copy()
df["date"] = pd.to_datetime(df["date"], format="%Y-%m-%d")
df["yield_5y"] = pd.to_numeric(df["yield_5y"], errors="coerce")
df["post_cut"] = pd.to_numeric(df["post_cut"], errors="coerce").astype("Int64")

issues = []
# --- audit ---
n_missing = int(df[["date", "yield_5y", "post_cut"]].isna().any(axis=1).sum())
n_dup = int(df["date"].duplicated().sum())
monotonic = bool(df["date"].is_monotonic_increasing)
n_treated = int((df["post_cut"] == 1).sum())
gaps = df["date"].diff().dt.days
long_gaps = df.loc[gaps > 4, "date"].dt.strftime("%Y-%m-%d").tolist()
if n_missing:
    issues.append({"issue": f"{n_missing} rows with missing/non-numeric values",
                   "handling": "reported; differences across them left NaN"})
if n_dup:
    issues.append({"issue": f"{n_dup} duplicated dates", "handling": "reported, not removed"})
if not monotonic:
    issues.append({"issue": "dates not sorted ascending", "handling": "file order kept per availability rule"})
sample_start = df["date"].min().strftime("%Y-%m-%d")
sample_end = df["date"].max().strftime("%Y-%m-%d")
if n_treated == 0:
    issues.append({
        "issue": (f"Coverage ends {sample_end} (README: Valet query end_date=2019-12-31), "
                  "before the 2020-03-04 cut; post_cut is 0 for all "
                  f"{n_raw} rows, so there are zero treated days"),
        "handling": ("Estimand not computable; effect, CI and model RMSE set to null, "
                     "conclusion premise_invalid. post_cut not redefined and no proxy "
                     "event or extrapolation substituted.")})
if long_gaps:
    issues.append({
        "issue": (f"{len(long_gaps)} calendar gaps of 5 days (year-end holidays, e.g. "
                  f"{long_gaps[0]}); one-day changes span non-trading days"),
        "handling": "kept; changes defined between consecutive rows per the task rule"})

# --- one-day changes (bp) between consecutive rows, dated t ---
df["chg_bp"] = 100.0 * df["yield_5y"].diff()
n_chg = int(df["chg_bp"].notna().sum())
baseline_metric = float(df["chg_bp"].mean())

excl = [{"step": 1, "rule": "first row has no previous row for a one-day change",
         "n_before": n_raw, "n_after": n_chg, "n_removed": n_raw - n_chg,
         "rationale": "change defined as 100*(y_t - y_{t-1}) between consecutive rows"}]

# --- effect: post-cut mean change minus mean over 120 days before first treated day ---
effect = lo = hi = rmse = None
n_analysed = None
treated_idx = np.flatnonzero((df["post_cut"] == 1).to_numpy())
if len(treated_idx) > 0:
    first = treated_idx[0]
    post = df.loc[df["post_cut"] == 1, "chg_bp"].dropna()
    pre = df["chg_bp"].iloc[max(0, first - PRE_WINDOW):first].dropna()
    if len(post) >= 2 and len(pre) >= 2:
        effect = float(post.mean() - pre.mean())
        v1, v0 = post.var(ddof=1) / len(post), pre.var(ddof=1) / len(pre)
        se = np.sqrt(v1 + v0)
        dof = (v1 + v0) ** 2 / (v1 ** 2 / (len(post) - 1) + v0 ** 2 / (len(pre) - 1))
        q = stats.t.ppf(0.975, dof)
        lo, hi = float(effect - q * se), float(effect + q * se)
        resid = np.concatenate([post - post.mean(), pre - pre.mean()])
        rmse = float(np.sqrt(np.mean(resid ** 2)))
        n_analysed = int(len(post) + len(pre))
    n_win = 0 if n_analysed is None else n_analysed
else:
    n_win = 0
excl.append({"step": 2, "rule": "keep post_cut=1 days and the 120 rows preceding the first post_cut=1 day",
             "n_before": n_chg, "n_after": n_win, "n_removed": n_chg - n_win,
             "rationale": ("estimand windows; no post_cut=1 rows exist so no window can be formed"
                           if n_win == 0 else "estimand windows")})
pd.DataFrame(excl).to_csv(os.path.join(OUT, "exclusions.csv"), index=False)

if effect is None:
    conclusion, feas = "premise_invalid", "infeasible"
    reason = (f"The file covers {sample_start} to {sample_end} and contains no rows with post_cut=1 "
              "(the cut was announced 2020-03-04), so the post-cut mean change and the "
              "120-day pre-window anchored on the first treated day do not exist.")
else:
    feas, reason = "complete", None
    conclusion = ("positive_effect" if lo > 0 else "negative_effect" if hi < 0
                  else "no_detectable_effect")

# --- table1: per calendar year ---
df["year"] = df["date"].dt.year
t1 = (df.groupby("year")
        .agg(n_days=("yield_5y", "size"), mean_yield=("yield_5y", "mean"),
             mean_change_bp=("chg_bp", "mean")).reset_index())
t1["mean_yield"] = t1["mean_yield"].round(6)
t1["mean_change_bp"] = t1["mean_change_bp"].round(6)
t1.to_csv(os.path.join(OUT, "table1.csv"), index=False)

# --- figure1 ---
fig, ax = plt.subplots(figsize=(9, 4))
ax.plot(df["date"], df["yield_5y"], lw=0.8, color="C0")
ax.axvline(pd.Timestamp("2020-03-04"), color="red", ls="--", lw=1,
           label="2020-03-04 cut (outside file coverage)")
ax.set_xlabel("date"); ax.set_ylabel("5-year GoC benchmark yield (%)")
ax.set_title(f"BD.CDN.5YR.DQ.YLD, {sample_start} to {sample_end} (post_cut=1 rows: {n_treated})")
ax.legend(loc="upper right"); fig.tight_layout()
fig.savefig(os.path.join(OUT, "figure1.png"), dpi=120)
plt.close(fig)

res = {
    "task_id": "F6-conf-02",
    "sample_start": sample_start, "sample_end": sample_end,
    "n_treated": n_treated,
    "baseline_metric": baseline_metric,
    "model_metric": rmse,
    "effect_estimate": effect, "effect_ci_low": lo, "effect_ci_high": hi,
    "conclusion": conclusion,
    "feasibility": feas, "feasibility_reason": reason,
    "data_issues": issues,
    "n_input_rows": {FN: n_raw},
    "n_analysed": n_analysed,
    "random_seed": None,
    # auxiliary audit values
    "n_changes": n_chg,
    "n_missing_rows": n_missing, "n_duplicate_dates": n_dup,
    "baseline_change_sd_bp": float(df["chg_bp"].std(ddof=1)),
}
with open(os.path.join(OUT, "results.json"), "w") as f:
    json.dump(res, f, indent=2)
print(json.dumps({k: res[k] for k in ["sample_start", "sample_end", "n_treated",
      "baseline_metric", "effect_estimate", "conclusion", "feasibility"]}, indent=2))