"""星港值班实验。所有费率、轨迹与识别结果都是本地练习夹具。"""
import argparse


def cache_ledger(repeats):
    # 每千 token 的练习刻度。输出在两条路线中都要付费。
    shared, variable, output = 4000, 200, 50
    unit = {"ordinary": 100, "write": 125, "read": 10, "output": 300}
    ordinary = repeats * ((shared + variable) * unit["ordinary"] + output * unit["output"]) / 1000
    with_cache = (shared * unit["write"] + (repeats - 1) * shared * unit["read"]
                  + repeats * (variable * unit["ordinary"] + output * unit["output"])) / 1000
    return ordinary, with_cache


def show_cache(n):
    assert cache_ledger(1) == (435, 535)
    assert cache_ledger(4) == (1740, 760)
    normal, cached = cache_ledger(n)
    print(f"同前缀调用 {n} 次 | 普通 {normal:g} 刻度 | 缓存 {cached:g} 刻度 | 差额 {normal-cached:g}")
    print(f"假设全部通过验收 | 每成功件：普通 {normal/n:g}，缓存 {cached/n:g} 刻度")


# 同一单调时钟上的毫秒。依次为请求到达、检索完成并发起生成、首个可见文字、完成。
TRACES = {"T1": (0, 80, 290, 710), "T2": (0, 85, 300, 730), "T3": (0, 480, 690, 1120)}


def analyze_trace(marks):
    arrived, model_start, first_text, end = marks
    if not arrived <= model_start <= first_text <= end:
        raise ValueError("检查跨度嵌套与时间顺序")
    return model_start-arrived, first_text-model_start, first_text-arrived, end-arrived


def show_traces(retrieval_ms):
    assert analyze_trace(TRACES["T3"]) == (480, 210, 690, 1120)
    print("轨迹 | 检索段 | 模型请求至首字 | 用户首字 | 端到端 (ms)")
    for name, marks in TRACES.items():
        if name == "T3":
            shift = retrieval_ms - marks[1]
            marks = (marks[0], marks[1]+shift, marks[2]+shift, marks[3]+shift)
        print(name, *analyze_trace(marks), sep=" | ")


MANIFEST = ("J1", "J2", "J3")
RECEIPTS = (("J3", "expired"), ("J2", "succeeded"), ("J2", "succeeded"), ("J1", "succeeded"))


def settle(manifest, records):
    ledger = {}
    for key, status in records:
        if key not in manifest or status not in {"succeeded", "expired"}:
            raise ValueError("未识别的回执，请人工核对")
        if key in ledger and ledger[key] != status:
            raise ValueError("同一批次出现相互矛盾的终态")
        ledger[key] = status
    return tuple(k for k in manifest if ledger.get(k) == "succeeded"), tuple(
        k for k in manifest if ledger.get(k) == "expired"), tuple(k for k in manifest if k not in ledger)


def show_batch(omit_j3):
    delivered, retry, unknown = settle(MANIFEST, RECEIPTS)
    assert (delivered, retry, unknown) == (("J1", "J2"), ("J3",), ())
    assert settle(MANIFEST, reversed(RECEIPTS)) == (delivered, retry, unknown)
    assert settle(MANIFEST, RECEIPTS[1:])[2] == ("J3",)
    if omit_j3:
        delivered, retry, unknown = settle(MANIFEST, RECEIPTS[1:])
    print("已交付", ",".join(delivered), "| 可重投", ",".join(retry) or "无", "| 待对账", ",".join(unknown) or "无")


# 两行货单的归一化纵坐标和两个模型提取候选。页面号从 1 开始。
ROWS = {"P1-A": (1, .20, .35, "12"), "P1-B": (1, .52, .67, "21")}
EXTRACTS = (("P1-A", 1, .27, "12"), ("P1-B", 1, .27, "21"))


def check_claim(candidate):
    key, page, y, quantity = candidate
    ref_page, top, bottom, expected = ROWS[key]
    return page == ref_page and top <= y <= bottom and quantity == expected


def show_documents(row_y):
    flags = [check_claim(c) for c in EXTRACTS]
    assert flags == [True, False]
    corrected = ("P1-B", 1, .59, "21")
    assert check_claim(corrected)
    rows = [EXTRACTS[0], ("P1-B", 1, row_y, "21")]
    flagged = [c[0] for c in rows if not check_claim(c)]
    print(f"P1-B y={row_y:.2f} | 通过 {len(rows)-len(flagged)}/{len(rows)} | 复核", ",".join(flagged) or "无")


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("station", choices=("o1","o2","o3","o4"))
    parser.add_argument("--repeats", type=int, default=4, help="o1 同前缀的调用次数")
    parser.add_argument("--retrieval-ms", type=int, default=480, help="o2 的 T3 检索耗时")
    parser.add_argument("--omit-j3", action="store_true", help="o3 模拟尚未拿到 J3 的回执")
    parser.add_argument("--row-y", type=float, default=.27, help="o4 改变第二行证据的归一化中心纵坐标")
    args = parser.parse_args()
    if args.retrieval_ms < 0 or not 0 <= args.row_y <= 1:
        parser.error("耗时须非负，归一化坐标须在 0 至 1 之间")
    if args.repeats < 1:
        parser.error("--repeats 至少是 1")
    if args.station == "o1":
        show_cache(args.repeats)
    elif args.station == "o2":
        show_traces(args.retrieval_ms)
    elif args.station == "o3":
        show_batch(args.omit_j3)
    else:
        show_documents(args.row_y)
