"""按财年和新鲜度定位预测分歧；原因仅由另供合成摘录确认。"""
import json
from datetime import date
from decimal import Decimal as D
from statistics import median

cutoff = date(2026, 10, 26)
reports = [("A", 2027, "2026-10-01", "0.90"), ("A", 2027, "2026-10-20", "1.00"),
           ("B", 2027, "2026-10-21", "1.10"), ("C", 2027, "2026-10-22", "1.20"),
           ("D", 2027, "2026-10-23", "2.00"), ("E", 2027, "2026-07-01", "3.00"),
           ("F", 2028, "2026-10-24", "4.00")]
def select_latest(records):
    latest, same_day_values = {}, {}
    for broker, year, published, eps in records:
        day = date.fromisoformat(published)
        if year != 2027 or day > cutoff:
            continue
        parsed = D(eps)
        if not parsed.is_finite():
            raise ValueError("nonfinite_forecast_eps")
        identity = (broker, year, day)
        if identity in same_day_values and same_day_values[identity] != parsed:
            raise ValueError("same_date_forecast_conflict")
        same_day_values[identity] = parsed
        if broker not in latest or day > latest[broker][0]:
            latest[broker] = (day, parsed)
    return latest

latest = select_latest(reports)
fresh = {b: value for b, (day, value) in latest.items() if (cutoff-day).days <= 60}
stale = {b: value for b, (day, value) in latest.items() if (cutoff-day).days > 60}
center = median(fresh.values())
flagged = sorted(b for b, value in fresh.items() if abs(value-center)/abs(center) >= D("0.50"))
cause_before = {b: "unknown_requires_original_report" for b in flagged}
supplied_excerpt = {"A": {"profit_millions": D("100"), "diluted_shares_millions": D("100"), "source": "synthetic-excerpt-A"},
                    "D": {"profit_millions": D("100"), "diluted_shares_millions": D("50"), "source": "synthetic-excerpt-D"}}
for broker, fields in supplied_excerpt.items():
    assert fields["profit_millions"]/fields["diluted_shares_millions"] == fresh[broker]
assert center == D("1.15") and flagged == ["D"] and set(stale) == {"E"}
assert set(fresh) == {"A", "B", "C", "D"}
fresh_mean = sum(fresh.values())/len(fresh)
assert fresh_mean == D("1.325")
assert cause_before == {"D": "unknown_requires_original_report"}
assert select_latest(list(reversed(reports))) == latest
for candidates in [reports + [("B", 2027, "2026-10-21", "9.00")],
                   list(reversed(reports + [("B", 2027, "2026-10-21", "9.00")]))]:
    try:
        select_latest(candidates)
    except ValueError as exc:
        assert str(exc) == "same_date_forecast_conflict"
    else:
        raise AssertionError("同日不同预测不能按行序挑选")
assert select_latest(reports + [("B", 2027, "2026-10-21", "1.10")]) == latest
for token in ["NaN", "Infinity", "-Infinity"]:
    try:
        select_latest([("B", 2027, "2026-10-21", token)])
    except ValueError as exc:
        assert str(exc) == "nonfinite_forecast_eps"
    else:
        raise AssertionError("非有限预测不能进入分歧计算")
print(json.dumps({"fresh": {b: str(v) for b, v in fresh.items()}, "stale": {b: str(v) for b, v in stale.items()},
                  "median": str(center), "fresh_mean": str(fresh_mean), "flagged": flagged,
                  "cause_before_excerpt": cause_before,
                  "confirmed_pair_A_D": "same supplied profit 100; diluted share assumptions 100 versus 50",
                  "outlier_deleted": False}, sort_keys=True))