#!/usr/bin/env python3
"""The round's registered integers, in ONE place, and the refusal every scorer makes.

PREREG §0: *"Every integer in §4–§7 is a denominator a scorer refuses to disagree with."* This
module is that sentence as code. A scorer that recomputes a denominator from whatever rows it
happens to find on disk cannot tell a partial run from a complete one, and it publishes the
smaller number without a word — which is the whole failure class the collection-state vocabulary
exists to prevent. So every scorer calls :func:`refuse_unless` on its own shape before it scores,
names the prereg section it is checking, and dies loudly on a mismatch.

Nothing here is derived from a results file. These are the numbers an operator co-signed at
G-PREREG; a run that produces different ones is a re-registration, not a rounding difference.
"""

from __future__ import annotations

import sys

sys.dont_write_bytecode = True

ROUND_ID = "two-frontiers-r1"

# ── Leg A · PREREG §4 ─────────────────────────────────────────────────────────
LEG_A_CASES = 60
LEG_A_CLASSES = {"answered": 36, "abstained-correct": 12, "injection": 6, "corrupt-corpus": 6}
LEG_A_REPS = 3
LEG_A_WARMUPS_PER_ARM = 1
LEG_A_ARMS = 3
LEG_A_SCORED_CALLS = LEG_A_CASES * LEG_A_REPS * LEG_A_ARMS          # 540
#: PREREG §4 (the users' words, D-20260905-09 (b)): 26 canonical one-tap asks run verbatim,
#: 34 free-typed queries are substituted. The four canonical strings and their counts are the
#: bank's own `enum_canonical_question` texts.
LEG_A_CANONICAL_QUERIES = {
    "How do I play?": 18,
    "How do I setup the game?": 4,
    "How do I take my turn?": 2,
    "When does the game end?": 2,
}
LEG_A_CANONICAL_CASES = 26
LEG_A_SUBSTITUTED_CASES = 34

# ── Leg C · PREREG §6 ─────────────────────────────────────────────────────────
LEG_C_ITEMS = 36
LEG_C_TIERS = ("8k", "16k", "32k")
LEG_C_DEPTHS = ("d10", "d50", "d90")
LEG_C_KINDS = ("recall", "absent")
LEG_C_ITEMS_PER_CELL = 2
LEG_C_NEEDLES = 12
LEG_C_ABSENTS = 6
LEG_C_RECALL_ITEMS = 18
LEG_C_ABSENT_ITEMS = 18
LEG_C_REPS = 1
LEG_C_ARMS = 3
LEG_C_SCORED_CALLS = LEG_C_ITEMS * LEG_C_REPS * LEG_C_ARMS          # 108
#: PREREG A3 FOLDED (2): the local seat runs Leg C a second time at think:true.
LEG_C_THINK_TRUE_CALLS = LEG_C_ITEMS * LEG_C_REPS                   # 36
#: PREREG §6 / measurement §4.7: reported prompt tokens / our estimate, per item, EVERY arm (A3.3).
LEG_C_CONTEXT_RATIO_FLOOR = 0.80
#: PREREG §6: the local arm refuses above this fraction of its own num_ctx rather than truncating.
LEG_C_LOCAL_NUM_CTX_REFUSAL = 0.95
#: PLAN §2 / C7's own prereg: differences under 2 items read TIED.
LEG_C_TIE_BAND = 2

# ── Leg H + the panel · PREREG §5, §7 ─────────────────────────────────────────
LEG_H_COMPARISONS = 36
LEG_H_ORDERS = 2
PANEL_SEATS = 7
PANEL_FLOOR_FAMILIES = 4
JUDGE_REPS = 1
LEG_H_JUDGE_CALLS = LEG_H_COMPARISONS * LEG_H_ORDERS * PANEL_SEATS  # 504
G4_JUDGE_CALLS_MAX = LEG_A_CLASSES["answered"] * LEG_A_ARMS * PANEL_SEATS  # 756
#: PREREG §4 / §5: a call whose sources + answer exceed this fraction of a judge's context is
#: REFUSED with the count published, never truncated into a verdict.
JUDGE_CTX_REFUSAL_RATIO = 0.7
#: PREREG §7: one full-size audition per seat per call shape (the G4 shape and the Leg H shape).
AUDITIONS = PANEL_SEATS * 2                                          # 14
#: PREREG §5: 2,000 percentile resamples over the 36 cases; t(35) noted beside it.
BOOTSTRAP_RESAMPLES = 2000
BOOTSTRAP_CLUSTERS = LEG_H_COMPARISONS
#: PLAN §2: intervals only where the independent-unit N is at least this.
INTERVAL_MIN_N = 30

# ── G-CALIBRATE · PREREG §4 (v) ───────────────────────────────────────────────
CALIBRATION_FLOOR = 0.85


class PreregMismatch(AssertionError):
    """A shape that disagrees with a registered integer. Always fatal, never a warning."""


def refuse_unless(condition: bool, *, got: object, want: object, section: str, what: str) -> None:
    """The one refusal. ``section`` names the prereg clause the caller is honouring.

    The message carries got, want AND the section, because the reader of a refusal is a person
    at 03:00 deciding whether the run is broken or the roster is: "36 != 60" alone does not say.
    """
    if not condition:
        raise PreregMismatch(
            f"{what}: got {got!r}, the prereg registers {want!r} ({section}). A denominator this "
            "harness recomputed differently from the co-signed document is a re-registration, "
            "not a rounding difference — fix the shape or amend the prereg in §12, dated."
        )


def refuse_equal(got: object, want: object, *, section: str, what: str) -> None:
    refuse_unless(got == want, got=got, want=want, section=section, what=what)


def assert_bank_composition(composition: dict[str, int], n_cases: int) -> None:
    """Leg A's four class counts and its N, together — PREREG §4."""
    refuse_equal(n_cases, LEG_A_CASES, section="PREREG §4", what="Leg A bank case count")
    refuse_equal(dict(composition), LEG_A_CLASSES, section="PREREG §4",
                 what="Leg A bank class composition")


def assert_legC_grid(items: list[dict]) -> None:
    """Leg C's grid: 36 items, 18 recall, 18 absent, and the tier x depth x kind shape."""
    refuse_equal(len(items), LEG_C_ITEMS, section="PREREG §6", what="Leg C item count")
    kinds = {k: sum(1 for it in items if it.get("kind") == k) for k in LEG_C_KINDS}
    refuse_equal(kinds["recall"], LEG_C_RECALL_ITEMS, section="PREREG §6",
                 what="Leg C recall items")
    refuse_equal(kinds["absent"], LEG_C_ABSENT_ITEMS, section="PREREG §6",
                 what="Leg C absent items")
    for tier in LEG_C_TIERS:
        for depth in LEG_C_DEPTHS:
            for kind in LEG_C_KINDS:
                cell = [it for it in items
                        if (it.get("tier"), it.get("depth"), it.get("kind")) == (tier, depth, kind)]
                refuse_equal(len(cell), LEG_C_ITEMS_PER_CELL, section="PREREG §6",
                             what=f"Leg C cell {tier}/{depth}/{kind}")


def assert_panel(seats: list[dict]) -> None:
    """Seven seats, seven families, zero house — PREREG §7 (+ amendment A1)."""
    refuse_equal(len(seats), PANEL_SEATS, section="PREREG §7", what="panel seat count")
    families = [s.get("family") for s in seats]
    refuse_equal(len(set(families)), PANEL_SEATS, section="PREREG §7",
                 what="panel family count (never two seated models from one vendor)")
    for seat in seats:
        model = str(seat.get("model", "")).lower()
        if "gpt-oss" in model or "claude" in model:
            raise PreregMismatch(
                f"seat {seat.get('seat')!r} names model {seat.get('model')!r}: PREREG §7 seats no "
                "gpt-oss tag (the openai family, strict reading) and no Claude, because both arms' "
                "vendors are the two the panel exists to be independent of."
            )