# Score E19 with grader v2: majority vote over normalized boxed answers (ties go to the lowest seed),
# and a paired exact McNemar test against the greedy E17 run (e17_math500_greedy.jsonl).
import json, math
from collections import Counter
import math_grade_v2 as G2
E17 = {}
for l in open("e17_math500_greedy.jsonl", encoding="utf-8"):
    r = json.loads(l); E17[r["item"]] = bool(G2.correct(r["text"], r["gold"]))
maj = orc = b = c = tok = 0
for l in open("e19_math500_sc4.jsonl", encoding="utf-8"):
    r = json.loads(l); S = r["samples"]
    for s in S:
        s["n"] = G2.norm(s["boxed"]) if s["boxed"] else None
        s["ok"] = bool(G2.correct(s["text"], r["gold"]))
    tok += sum(s["predicted_n"] or 0 for s in S)
    cnt = Counter(s["n"] for s in S if s["n"] is not None)
    ok = False
    if cnt:
        top = max(cnt.values()); win = next(s["n"] for s in S if s["n"] is not None and cnt[s["n"]] == top)
        ok = next(s["ok"] for s in S if s["n"] == win)
    maj += ok; orc += any(s["ok"] for s in S)
    g = E17[r["item"]]; b += g and not ok; c += ok and not g
n = b + c; p = min(1, 2 * sum(math.comb(n, i) for i in range(min(b, c) + 1)) / 2 ** n) if n else 1
print(f"greedy {sum(E17.values())}/500  sc@4 {maj}/500  oracle {orc}/500  greedy-only {b}  sc-only {c}  McNemar p={p:.3g}  tokens {tok}")