"""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))
os.makedirs(OUT, exist_ok=True)

raw = pd.read_csv(DATA, dtype={"iso3": str, "country_name": str})
n_raw = len(raw)
for c in ["iso3", "country_name"]:
    raw[c] = raw[c].str.strip()
excl = []
issues = []

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

# Step 1: restrict to study window
d0 = raw
d1 = d0[d0["year"].between(2000, 2022)].copy()
step("1_year_window", "keep 2000 <= year <= 2022", d0, d1, "study window defined by the task")

# Step 2: duplicated (iso3, year) rows
dup_mask = d1.duplicated(["iso3", "year"], keep=False)
dups = d1[dup_mask]
conflicting = dups.groupby(["iso3", "year"])["life_expectancy"].nunique(dropna=False)
conflicting = conflicting[conflicting > 1]
if len(conflicting):
    raise SystemExit(f"Conflicting duplicate values found: {conflicting}")
for iso, g in dups.groupby("iso3"):
    names = sorted(g["country_name"].unique())
    issues.append({
        "issue": f"{iso}: {g[['iso3','year']].drop_duplicates().shape[0]} country-years appear "
                 f"{len(g)} times in total under country names {names} (two extractions; API renamed the "
                 f"country); duplicate values are identical",
        "handling": "country identified by iso3; kept one row per (iso3, year) (first occurrence); "
                    "values identical so choice does not affect results"})
d2 = d1.drop_duplicates(["iso3", "year"], keep="first").copy()
step("2_dedup_iso3_year", "keep one row per (iso3, year); duplicates have identical values", d1, d2,
     "each country contributes at most one observation per year (assembly duplicates)")
# harmonise display name to the most recent spelling (last occurrence in file) per iso3
latest_name = d1.groupby("iso3")["country_name"].last()
d2["country_name"] = d2["iso3"].map(latest_name)

# Completeness per country
miss_rows = d2[d2["life_expectancy"].isna()]
for _, r in miss_rows.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"})
obs = d2.dropna(subset=["life_expectancy"])
years_by = obs.groupby("iso3")["year"].apply(set)
all_iso = sorted(d2["iso3"].unique())
balanced = sorted(i for i in all_iso if set(YEARS) <= years_by.get(i, set()))
incomplete = [i for i in all_iso if i not in balanced]
for i in incomplete:
    present = years_by.get(i, set())
    missing_years = sorted(set(YEARS) - present)
    if len(missing_years) > 1 and missing_years == list(range(missing_years[0], missing_years[-1] + 1)):
        desc = f"{missing_years[0]}-{missing_years[-1]}"
    else:
        desc = ",".join(map(str, missing_years))
    if not (len(missing_years) == 1 and i in set(miss_rows.iso3)):
        issues.append({"issue": f"{i}: no rows/values for years {desc} ({len(missing_years)} of 23 years missing)",
                       "handling": "treated as missing (no imputation); country excluded from balanced set"})

# Step 3: drop rows with empty value
d3 = d2.dropna(subset=["life_expectancy"])
step("3_drop_missing_value", "drop rows with empty life_expectancy", d2, d3,
     "a country-year without a recorded value is missing; no imputation")
# Step 4: drop countries not observed in all 23 years
d4 = d3[d3["iso3"].isin(balanced)]
step("4_balanced_panel", "keep only iso3 with a recorded value in every year 2000-2022", d3, d4,
     f"balanced panel; excluded countries: {', '.join(incomplete)}")

analytic = d4[["iso3", "country_name", "year", "life_expectancy"]].sort_values(["iso3", "year"]).reset_index(drop=True)
assert len(analytic) == len(balanced) * len(YEARS)
assert not analytic.duplicated(["iso3", "year"]).any()
analytic.to_csv(f"{OUT}/analytic.csv", index=False)
pd.DataFrame(excl).to_csv(f"{OUT}/exclusions.csv", index=False)

m = analytic.groupby("year")["life_expectancy"].mean()
m2000, m2022 = float(m[2000]), float(m[2022])
avail22 = d3[d3["year"] == 2022]  # already one row per iso3
results = {
    "task_id": "F1-conf-02",
    "n_countries_total": len(all_iso),
    "n_countries_balanced": len(balanced),
    "n_countries_incomplete": len(incomplete),
    "balanced_countries": balanced,
    "n_analysed": int(len(analytic)),
    "mean_2000_balanced": round(m2000, 6),
    "mean_2022_balanced": round(m2022, 6),
    "change_2000_2022_balanced": round(m2022 - m2000, 6),
    "mean_2022_all_available": round(float(avail22["life_expectancy"].mean()), 6),
    "n_countries_2022_available": int(avail22["iso3"].nunique()),
    "incomplete_countries": incomplete,
    "balanced_means_by_year": {str(y): round(float(v), 6) for y, v in m.items()},
    "feasibility": "complete",
    "feasibility_reason": None,
    "data_issues": issues,
    "n_input_rows": {"life_expectancy_panel.csv": int(n_raw)},
    "random_seed": None,
}
with open(f"{OUT}/results.json", "w") as f:
    json.dump(results, f, indent=2)
print(json.dumps({k: results[k] for k in ["n_countries_total", "n_countries_balanced", "n_analysed",
      "mean_2000_balanced", "mean_2022_balanced", "change_2000_2022_balanced", "mean_2022_all_available"]}, indent=1))