"""Neutral filler text for the context depth axis.

Deterministic for a given seed. Nothing here mentions invoices, payments,
customers or instructions, so the only thing it adds is distance.
Token counts are estimated at 4 characters per token; the runner records
the real prompt_tokens from the API response.
"""
import random

CHARS_PER_TOKEN = 4

_SUBJECTS = [
    "The Leeds depot", "The night shift", "The facilities team", "The new forklift",
    "The timber rack", "The yard gate", "The staff canteen", "The training room",
    "The delivery schedule", "The weekly safety walk", "The recycling skip",
    "The car park resurfacing", "The uniform order", "The first aid cabinet",
]
_VERBS = [
    "was inspected on", "will be reviewed on", "is booked for", "was discussed at the meeting on",
    "has been rescheduled to", "was signed off on", "is due for maintenance on", "was repainted on",
]
_DATES = [
    "Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "the 3rd", "the 12th",
    "the 19th", "the 26th", "the first of the month", "the last working day",
]
_TAILS = [
    "and no further action is needed.", "and the notes are on the shared drive.",
    "pending a quote from the contractor.", "with two staff attending.",
    "and the manager has been informed.", "as part of the quarterly plan.",
    "and the checklist was completed.", "following the usual procedure.",
    "and the supplier confirmed by phone.", "with photos attached to the ticket.",
]
_POLICY = [
    "Hard hats and boots must be worn beyond the yellow line in the yard.",
    "Forklift keys are returned to the office at the end of each shift.",
    "Visitors sign in at reception and are accompanied at all times.",
    "The canteen fridge is cleared every Friday afternoon.",
    "Annual leave requests go through the online portal with two weeks notice.",
    "Timber offcuts under one metre go in the blue skip, not the green one.",
    "Fire assembly point is the far corner of the staff car park.",
    "Deliveries are not accepted after four thirty without prior arrangement.",
    "Radios are charged overnight in the dispatch office.",
    "Spill kits are checked on the first Monday of each month.",
]


def make_padding(n_tokens: int, seed: int = 7) -> str:
    """Return roughly n_tokens of neutral text as numbered paragraphs."""
    if n_tokens <= 0:
        return ""
    rng = random.Random(seed)
    target_chars = n_tokens * CHARS_PER_TOKEN
    paragraphs = []
    total = 0
    i = 1
    while total < target_chars:
        sentences = [
            f"{rng.choice(_SUBJECTS)} {rng.choice(_VERBS)} {rng.choice(_DATES)} {rng.choice(_TAILS)}"
            for _ in range(rng.randint(3, 6))
        ]
        if rng.random() < 0.3:
            sentences.append(rng.choice(_POLICY))
        para = f"Note {i}. " + " ".join(sentences)
        paragraphs.append(para)
        total += len(para) + 2
        i += 1
    return "\n\n".join(paragraphs)


def split_padding(padding: str, fraction: float) -> tuple[str, str]:
    """Split padding at a paragraph boundary near the given fraction of its length."""
    if not padding:
        return "", ""
    cut = int(len(padding) * fraction)
    boundary = padding.rfind("\n\n", 0, cut)
    if boundary == -1:
        boundary = padding.find("\n\n", cut)
    if boundary == -1:
        return "", padding
    return padding[:boundary], padding[boundary + 2:]