"""F4-conf-02: verdicts on six assertions against their cited source extracts.
Reads only data/; writes out/results.json deterministically. No randomness."""
import json, math, os, re

ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
D = os.path.join(ROOT, "data")
DCR = "arxiv-2609.27732v1"
RES = "arxiv-2606.14621v1"

def lines(p):
    with open(p, encoding="utf-8") as f:
        return f.read().split("\n")

src = {DCR: lines(os.path.join(D, "sources", DCR + ".txt")),
       RES: lines(os.path.join(D, "sources", RES + ".txt"))}
with open(os.path.join(D, "assertions.json"), encoding="utf-8") as f:
    assertions = json.load(f)
A = {a["id"]: a for a in assertions}
for a in assertions:
    assert a["cites"] in src, a

def wcl(p):
    with open(p, "rb") as f:
        return f.read().count(b"\n")

n_input_rows = {
    "assertions.json": len(assertions),
    "README.md": wcl(os.path.join(D, "README.md")),
    DCR + ".txt": wcl(os.path.join(D, "sources", DCR + ".txt")),
    RES + ".txt": wcl(os.path.join(D, "sources", RES + ".txt")),
}

def find(sid, pattern, start=0):
    """Return (index, line) of first line at/after start matching regex; print it."""
    for i in range(start, len(src[sid])):
        if re.search(pattern, src[sid][i]):
            print(f"  [{sid} L{i+1}] {src[sid][i].strip()}")
            return i, src[sid][i]
    raise ValueError(f"pattern not found: {pattern}")

def cells(line):
    return [c.strip() for c in line.split("|")]

def num(s):
    return float(re.sub(r"[^0-9.\-]", "", s.replace("−", "-")))

def pct(b):
    return 100.0 * (math.exp(b) - 1.0)

def ci(b, se, z):
    return (b - z * se, b + z * se)

def stars(s):
    return len(re.findall(r"\*", s))

verdicts, locators, reasons, data_issues = {}, {}, {}, []
cols = {"1 km": 1, "3 km": 2, "5 km": 3}

# ---------------- Table 6 (DCR) ----------------
print("Table 6 lines relied on:")
i6, _ = find(DCR, r"^Table 6:")
iatt, latt = find(DCR, r"^ATT \|", i6)
_, lse = find(DCR, r"^\| \(", iatt)
_, lpct = find(DCR, r"^% change \|", iatt)
t6 = {k: {"att_s": cells(latt)[v], "att": num(cells(latt)[v]),
          "se": num(cells(lse)[v]), "pct": num(cells(lpct)[v])} for k, v in cols.items()}
find(DCR, r"statistically significant at the 5% level", i6)

# A1
print("A1:", A["A1"]["text"])
r = t6["5 km"]
z = r["att"] / r["se"]
lo, hi = ci(r["att"], r["se"], 1.96)
ok = (r["att"] == -0.431 and r["se"] == 0.210 and stars(r["att_s"]) >= 2
      and abs(z) > 1.96 and r["pct"] == -35.0 and round(pct(r["att"]), 1) == -35.0)
print(f"  check: ATT={r['att']} SE={r['se']} stars={stars(r['att_s'])} |z|={abs(z):.3f} "
      f"CI95=[{lo:.3f},{hi:.3f}] printed%={r['pct']} 100(exp(b)-1)={pct(r['att']):.2f}")
verdicts["A1"] = "supported" if ok else "not_supported"
locators["A1"] = f'{DCR} | Table 6 | row "ATT" | column "5 km"'
reasons["A1"] = ("Table 6 prints the 5 km ATT as -0.431** (SE 0.210, ** = p<0.05) with a % change of -35.0% "
                 "(100*(exp(-0.431)-1) = -35.0%), matching every element of the assertion.")

# A3
print("A3:", A["A3"]["text"])
r = t6["3 km"]
print(f"  check: ATT={r['att']} printed%={r['pct']} 100(exp(b)-1)={pct(r['att']):.2f} vs asserted -28.3%")
ok = r["att"] == -0.283 and abs(pct(r["att"]) - (-28.3)) < 0.05
verdicts["A3"] = "supported" if ok else "not_supported"
locators["A3"] = f'{DCR} | Table 6 | row "% change" | column "3 km"'
reasons["A3"] = ("The ATT of -0.283 is in log points and Table 6 converts it to a -24.7% change "
                 f"(100*(exp(-0.283)-1) = {pct(r['att']):.2f}%), not a 28.3% reduction; the estimate is also not "
                 "significant (SE 0.279).")

# ---------------- Table 7 (DCR) ----------------
print("Table 7 lines relied on:")
i7, _ = find(DCR, r"^Table 7:")
def t7row(label):
    i, l = find(DCR, r"^" + re.escape(label) + r" \|", i7)
    _, s = find(DCR, r"^\| \(", i)
    return {k: {"att_s": cells(l)[v], "att": num(cells(l)[v]), "se": num(cells(s)[v])} for k, v in cols.items()}
hiv = t7row("HIV")
hep = t7row("Hepatitis")
find(DCR, r"HIV estimates are insignificant and close to zero", i7)

# A2
print("A2:", A["A2"]["text"])
r = hiv["1 km"]
print(f"  check: HIV 1 km ATT={r['att_s']} (SE {r['se']}); asserted -0.063")
verdicts["A2"] = "supported" if r["att"] < 0 else "not_supported"
locators["A2"] = f'{DCR} | Table 7 | row "HIV" | column "1 km"'
reasons["A2"] = ("Table 7 prints the 1 km HIV ATT as 0.063, i.e. positive (SE 0.801), so the asserted negative sign "
                 "and the claimed fall in HIV hospitalisations are contradicted.")

# A4
print("A4:", A["A4"]["text"])
r = hep["5 km"]
c196 = ci(r["att"], r["se"], 1.96); c2 = ci(r["att"], r["se"], 2.0)
c1 = ci(r["att"], r["se"], 1.0); c1645 = ci(r["att"], r["se"], 1.645)
print(f"  check: Hepatitis 5 km ATT={r['att_s']} SE={r['se']} CI95(1.96)=[{c196[0]:.3f},{c196[1]:.3f}] "
      f"CI95(2.0)=[{c2[0]:.3f},{c2[1]:.3f}] CI90=[{c1645[0]:.3f},{c1645[1]:.3f}] b±1SE=[{c1[0]:.3f},{c1[1]:.3f}]")
excl = c196[1] < 0 or c196[0] > 0
verdicts["A4"] = "supported" if excl else "not_supported"
locators["A4"] = f'{DCR} | Table 7 | row "Hepatitis" | column "5 km"'
reasons["A4"] = (f"With ATT -0.406 and SE 0.250 (no stars), the 95% CI is about [{c196[0]:.2f}, {c196[1]:.2f}], "
                 "which includes zero; the asserted [-0.66, -0.16] is only the estimate ± 1 SE.")

# A5
print("A5:", A["A5"]["text"])
hits = [i + 1 for i, l in enumerate(src[DCR]) if re.search(r"\b(men|women|male|female|sex|gender)\b", l, re.I)]
print(f"  check: lines mentioning men/women/male/female/sex/gender in {DCR}: {hits}")
verdicts["A5"] = "not_assessable" if not hits else "MANUAL_REVIEW"
locators["A5"] = f"not reported in {DCR}"
reasons["A5"] = ("The extract reports no estimates by sex (Tables 6, 7 and A19 report only buffer, outcome-category and estimator breakdowns, none by sex), "
                 "so a men-versus-women comparison can be neither confirmed nor refuted.")

# ---------------- RESEA Table 2 ----------------
print("A6:", A["A6"]["text"])
_, l = find(RES, r"^Weeks on UI \|")
c = cells(l)
ctrl = num(re.match(r"([\d.]+) weeks", c[2]).group(1))
m = re.match(r"(-?[\d.]+) weeks; p<0\.001; 95% CI: \[(-?[\d.]+), (-?[\d.]+)\]", c[4])
ate, lo6, hi6 = (float(m.group(k)) for k in (1, 2, 3))
find(RES, r"11\.5% decrease")
rel = 100 * ate / ctrl
print(f"  check: ATE={ate} CI=[{lo6},{hi6}] control mean={ctrl} ATE/control={rel:.2f}%")
ok = ate == -1.99 and (lo6, hi6) == (-2.24, -1.75) and ctrl == 17.3 and round(rel, 1) == -11.5
verdicts["A6"] = "supported" if ok else "not_supported"
locators["A6"] = f'{RES} | Table 2 | row "Weeks on UI" | column "Average treatment effect estimate"'
reasons["A6"] = ("Table 2 prints the ATE for weeks on UI as -1.99 weeks (95% CI [-2.24, -1.75]) with a control mean "
                 f"of 17.3 weeks, and -1.99/17.3 = {rel:.1f}%, matching the text's 11.5% decrease.")

# ---------------- internal-consistency scan (data issues) ----------------
print("Consistency checks on DCR extract:")
iA19, _ = find(DCR, r"^Table A19:")
ia, la = find(DCR, r"^ATT \|", iA19)
_, lsa = find(DCR, r"^\| \(", ia)
_, lpa = find(DCR, r"^% change \|", ia)
find(DCR, r"never-treated comparison regions \(640 regions\)", iA19)
se_a19 = num(cells(lsa)[1])
data_issues.append({
    "issue": (f"{DCR}: the CS-DR 5 km overall ATT -0.431** has SE 0.210 in Table 6 but SE {se_a19:.3f} in Appendix "
              "Table A19 (same estimator, buffer, window and N=17,250)."),
    "handling": ("A1 is judged on Table 6 (the cited main result); the 5% significance claim holds under both SEs "
                 f"(|z| = {0.431/0.210:.2f} and {0.431/se_a19:.2f} > 1.96), so the verdict is unaffected.")})
data_issues.append({
    "issue": f"{DCR}: Table A19 lists 608 control regions but its note says 640 never-treated comparison regions.",
    "handling": "Not decisive for any assertion; recorded only."})
fatal5 = num(cells(find(DCR, r"^Drug-related \(fatal\) \|", i7)[1])[3])
data_issues.append({
    "issue": (f"{DCR}: the abstract says fatal outcomes fall by 35%, whereas Table 7 gives a 5 km fatal ATT of "
              f"{fatal5} (100*(exp(b)-1) = {pct(fatal5):.1f}%) and Section 6.1 says approximately 30%."),
    "handling": "Not decisive for any assertion (none concerns fatal outcomes); recorded only."})
for k, lab in [(3, "Synthetic DiD"), (4, "Stacked DiD")]:
    b = num(cells(la)[k]); p = num(cells(lpa)[k])
    print(f"  A19 {lab}: b={b} printed%={p} computed={pct(b):.2f}")
data_issues.append({
    "issue": (f"{DCR}: Table A19 printed % changes for Synthetic DiD ({num(cells(lpa)[3])}%) and Stacked DiD "
              f"({num(cells(lpa)[4])}%) differ by 0.1 pp from 100*(exp(b)-1) "
              f"({pct(num(cells(la)[3])):.1f}%, {pct(num(cells(la)[4])):.1f}%), likely rounding from unrounded b."),
    "handling": "Not decisive for any assertion; recorded only."})
data_issues.append({
    "issue": (f"{RES}: the results text reports the weeks-on-UI CI as magnitudes ([1.75 weeks, 2.24 weeks]) while "
              "Table 2 prints signed bounds [-2.24, -1.75]; the text also contains a rendering defect 'p ¡ 0.001'."),
    "handling": "Treated as equivalent sign conventions; A6 judged on the signed Table 2 values."})

# ---------------- naive baseline: share of asserted numbers found verbatim in the cited extract ----------------
baseline = {}
for a in assertions:
    toks = re.findall(r"\d+(?:\.\d+)?", a["text"].replace("−", "-"))
    text = "\n".join(src[a["cites"]])
    found = sum(1 for t in toks if t in text)
    baseline[a["id"]] = round(found / len(toks), 4) if toks else None
print("Baseline (share of asserted numeric tokens found verbatim in cited extract):", baseline)

verdicts = {k: verdicts[k] for k in sorted(verdicts)}
locators = {k: locators[k] for k in sorted(locators)}
reasons = {k: reasons[k] for k in sorted(reasons)}
counts = {v: sum(1 for x in verdicts.values() if x == v) for v in ("supported", "not_supported", "not_assessable")}
results = {
    "task_id": "F4-conf-02",
    "verdicts": verdicts,
    "locators": locators,
    "reasons": reasons,
    "n_supported": counts["supported"],
    "n_not_supported": counts["not_supported"],
    "n_not_assessable": counts["not_assessable"],
    "feasibility": "complete",
    "feasibility_reason": None,
    "data_issues": data_issues,
    "n_input_rows": n_input_rows,
    "n_analysed": len(verdicts),
    "random_seed": None,
    "baseline_share_numbers_found_verbatim": baseline,
}
assert sum(counts.values()) == 6
os.makedirs(os.path.join(ROOT, "out"), exist_ok=True)
with open(os.path.join(ROOT, "out", "results.json"), "w", encoding="utf-8") as f:
    json.dump(results, f, indent=2, ensure_ascii=False)
    f.write("\n")
print("Verdicts:", verdicts)
print("Counts:", counts)