# Trouve-KA — priorités et recrawl adaptatif
# Author: Simon-Pierre Boucher
# Contact: contact@spboucher.ai

"""Fonction de priorité du frontier et scheduling de recrawl.

P = w_q·Q + w_a·A + w_f·F + w_l·L + w_n·N − w_d·D − w_s·S  (CLAUDE.md §5.3)
Les poids sont des constantes nommées, destinées à être calibrées par mesures.
"""

from datetime import timedelta

# Poids de la fonction de priorité — à calibrer empiriquement, jamais figés
W_QUEBEC = 0.40
W_AUTHORITY = 0.15
W_FRESHNESS = 0.10
W_LINKS = 0.10
W_NOVELTY = 0.15
W_DEPTH = 0.05
W_SPAM = 0.30


def compute_priority(
    *,
    domain_quebec_score: float,
    authority_score: float = 0.0,
    freshness_hint: float = 0.0,
    link_signal: float = 0.0,
    is_new_domain: bool = False,
    depth: int = 0,
    spam_signal: float = 0.0,
    is_seed: bool = False,
) -> float:
    """Priorité ∈ [0, 1]. Les seeds démarrent au maximum."""
    if is_seed:
        return 1.0
    p = (
        W_QUEBEC * domain_quebec_score
        + W_AUTHORITY * min(authority_score, 1.0)
        + W_FRESHNESS * min(freshness_hint, 1.0)
        + W_LINKS * min(link_signal, 1.0)
        + W_NOVELTY * (1.0 if is_new_domain else 0.3)
        - W_DEPTH * min(depth, 10) / 10.0
        - W_SPAM * min(spam_signal, 1.0)
    )
    return max(0.0, min(1.0, round(p, 4)))


def next_recrawl_delay(
    *,
    changed: bool,
    previous_delay_hours: float | None,
    min_hours: float = 1.0,
    max_hours: float = 24 * 30.0,
    default_hours: float = 24.0,
) -> timedelta:
    """Recrawl adaptatif (§5.5) : page inchangée → intervalle ×2; page volatile → intervalle ÷2."""
    prev = previous_delay_hours or default_hours
    hours = max(min_hours, prev / 2.0) if changed else min(max_hours, prev * 2.0)
    return timedelta(hours=hours)


def retry_delay(retries: int) -> timedelta:
    """Backoff exponentiel plafonné pour les erreurs transitoires."""
    return timedelta(minutes=min(15 * (2**retries), 60 * 24))