"""F1-conf-02: balanced-panel mean life expectancy 2000-2022.
Run from workspace root: ~/wp29/venv/bin/python out/analysis.py
Reads only data/; writes out/results.json, out/analytic.csv, out/exclusions.csv.
No randomness is used (random_seed = null)."""
import json, os
import pandas as pd

DATA = "data/life_expectancy_panel.csv"
OUT = "out"
YEARS = list(range(2000, 2023))
CANON_NAME = {"TUR": "Turkiye"}  # documented choice for the duplicated Turkey/Turkiye rows

raw = pd.read_csv(DATA, dtype={"country_name": str, "iso3": str})
n_raw = len(raw)
ledger, issues = [], []

def step(name, rule, before, after, why):
    ledger.append(dict(step=name, rule=rule, n_before=before, n_after=after,
                       n_removed=before - after, rationale=why))

# --- detect duplicates on (iso3, year)
dup = raw[raw.duplicated(["iso3", "year"], keep=False)]
for iso, g in dup.groupby("iso3"):
    conflicting = (g.groupby("year")["life_expectancy"].nunique(dropna=False) > 1).sum()
    if conflicting:
        raise SystemExit(f"Conflicting duplicate values for {iso}: decision required")
    names = sorted(g.country_name.unique())
    issues.append({"issue": f"{iso} appears twice per year under names {names} "
                   f"({len(g)} rows for {g.year.nunique()} years; values identical in every year) - "
                   "assembly error from the two extractions",
                   "handling": f"deduplicated on (iso3, year), keeping one row per year; "
                   f"country_name standardised to '{CANON_NAME.get(iso, names[-1])}'; values unaffected"})
d = raw.copy()
d["country_name"] = d.apply(lambda r: CANON_NAME.get(r.iso3, r.country_name), axis=1)
d = d.sort_values(["iso3", "year"]).drop_duplicates(["iso3", "year"], keep="first")
step("1_dedup", "keep one row per (iso3, year); duplicates had identical values",
     n_raw, len(d), "country identified by iso3; at most one observation per country-year")

# --- missing values
miss = d[d.life_expectancy.isna()]
for _, r in miss.iterrows():
    issues.append({"issue": f"{r.iso3} {int(r.year)}: empty life_expectancy cell",
                   "handling": "treated as missing (no imputation); country excluded from balanced set"})
d_obs = d.dropna(subset=["life_expectancy"])
step("2_drop_missing", "drop country-years with empty life_expectancy", len(d), len(d_obs),
     "empty cell = no recorded value; no imputation allowed")

# --- countries with absent years
cov = d_obs.groupby("iso3")["year"].apply(lambda s: sorted(set(s)))
for iso, ys in cov.items():
    absent_rows = sorted(set(YEARS) - set(d.loc[d.iso3 == iso, "year"]))
    if absent_rows:
        issues.append({"issue": f"{iso}: no rows at all for years {absent_rows[0]}-{absent_rows[-1]} "
                       f"({len(absent_rows)} years; coverage only {min(ys)}-{max(ys)})",
                       "handling": "treated as missing (no imputation); country excluded from balanced set, "
                       "kept in the 2022 all-available baseline"})

balanced = sorted(iso for iso, ys in cov.items() if set(YEARS) <= set(ys))
an = d_obs[d_obs.iso3.isin(balanced) & d_obs.year.isin(YEARS)]
step("3_balanced", "keep countries with a recorded value in every year 2000-2022", len(d_obs), len(an),
     "balanced-panel estimand: fixed composition across years")
an = an[["iso3", "country_name", "year", "life_expectancy"]].sort_values(["iso3", "year"]).reset_index(drop=True)
assert len(an) == len(balanced) * len(YEARS)
assert not an.duplicated(["iso3", "year"]).any()

# --- estimates
ymean = an.groupby("year")["life_expectancy"].mean()
m00, m22 = float(ymean[2000]), float(ymean[2022])
base = d_obs[d_obs.year == 2022]
assert base.iso3.is_unique
m22_all = float(base.life_expectancy.mean())

n_tot = int(d.iso3.nunique())
res = {
    "task_id": "F1-conf-02",
    "n_countries_total": n_tot,
    "n_countries_balanced": len(balanced),
    "n_countries_incomplete": n_tot - len(balanced),
    "balanced_countries": balanced,
    "n_analysed": int(len(an)),
    "mean_2000_balanced": round(m00, 6),
    "mean_2022_balanced": round(m22, 6),
    "change_2000_2022_balanced": round(m22 - m00, 6),
    "mean_2022_all_available": round(m22_all, 6),
    "n_countries_2022_all_available": int(base.iso3.nunique()),
    "incomplete_countries": sorted(set(d.iso3) - set(balanced)),
    "balanced_mean_by_year": {str(y): round(float(v), 6) for y, v in ymean.items()},
    "feasibility": "complete",
    "feasibility_reason": None,
    "data_issues": issues,
    "n_input_rows": {"life_expectancy_panel.csv": int(n_raw)},
    "random_seed": None,
}
os.makedirs(OUT, exist_ok=True)
an.to_csv(f"{OUT}/analytic.csv", index=False, float_format="%.3f")
pd.DataFrame(ledger).to_csv(f"{OUT}/exclusions.csv", index=False)
with open(f"{OUT}/results.json", "w") as f:
    json.dump(res, f, indent=2, ensure_ascii=False)
print(f"balanced countries: {len(balanced)} {balanced}")
print(f"mean 2000 = {res['mean_2000_balanced']}, mean 2022 = {res['mean_2022_balanced']}, "
      f"change = {res['change_2000_2022_balanced']}; 2022 all-available = {res['mean_2022_all_available']}")