"""How the Treasury derived the curve, and when it changed how.

The par yield curve is not a measurement, it is the output of a procedure, and
the procedure changed on 6 December 2021: the monotone convex method replaced
the quasi-cubic hermite spline. The Treasury says so on its methodology page
and publishes a comparison of what the two produce.

That matters because a rate from 2015 and a rate from 2022 are not the same
kind of number, and neither is recomputed under the other's method. The
published values stand as published; `methodology_id` on every curve row says
which procedure produced it, and this registry says what each one was.
"""

from __future__ import annotations

from datetime import date
from pathlib import Path

import polars as pl

METHODOLOGY_URL = (
    "https://home.treasury.gov/policy-issues/financing-the-government/interest-rate-statistics/"
    "treasury-yield-curve-methodology"
)

SCHEMA = {
    "methodology_id": pl.String,
    "effective_from": pl.Date,
    "effective_to": pl.Date,
    "name": pl.String,
    "description": pl.String,
    "source_url": pl.String,
}

# `effective_to` is the last day produced under that method. The Treasury's
# wording is "replaced ... as of December 6, 2021", so 6 December is the first
# monotone convex curve and 3 December, the trading day before it, the last
# spline one.
METHODOLOGIES = [
    {
        "methodology_id": "quasi_cubic_hermite_spline",
        "effective_from": date(1990, 1, 2),
        "effective_to": date(2021, 12, 3),
        "name": "Quasi-cubic hermite spline",
        "description": (
            "The par yield curve was derived with a quasi-cubic hermite spline fitted to "
            "indicative bid-side quotations for the most recently auctioned securities, "
            "obtained by the Federal Reserve Bank of New York at or near 3:30 PM Eastern. "
            "The Treasury has archived the full description of this method. Its start date "
            "here is the first day of the published daily series, not the date the method "
            "was adopted, which the Treasury does not state."
        ),
        "source_url": METHODOLOGY_URL,
    },
    {
        "methodology_id": "monotone_convex",
        "effective_from": date(2021, 12, 6),
        "effective_to": None,
        "name": "Monotone convex",
        "description": (
            "Inputs are indicative, bid-side market price quotations -- not transactions -- "
            "for the most recently auctioned securities, obtained by the Federal Reserve "
            "Bank of New York at or near 3:30 PM Eastern. The prices are converted to yields "
            "and used to bootstrap the instantaneous forward curve by a monotone convex "
            "method. The Treasury states that this replaced the quasi-cubic hermite spline "
            "as of December 6, 2021."
        ),
        "source_url": METHODOLOGY_URL,
    },
]


def registry() -> pl.DataFrame:
    return pl.DataFrame(METHODOLOGIES, schema=SCHEMA).sort("effective_from")


def methodology_for(day: date) -> str | None:
    """Which procedure produced the curve published for this day."""

    for entry in METHODOLOGIES:
        if day < entry["effective_from"]:
            continue
        if entry["effective_to"] is None or day <= entry["effective_to"]:
            return entry["methodology_id"]
    return None


def write_registry(metadata_dir: Path) -> int:
    metadata_dir.mkdir(parents=True, exist_ok=True)
    table = registry()
    table.write_parquet(metadata_dir / "yield_curve_methodology.parquet", compression="zstd")
    return table.height


__all__ = [
    "METHODOLOGIES",
    "METHODOLOGY_URL",
    "SCHEMA",
    "methodology_for",
    "registry",
    "write_registry",
]