config.py

import os 
import logging
from datetime import datetime 

RELEVANT_TOPICS = {"finance", "business", "geopolitics", "technology", "economics"}
CANDIDATE_LABELS = list(RELEVANT_TOPICS) + ["sports", "entertainment", "lifestyle", "health", "weather", "other"]

RSS_FEEDS = {
    "Yahoo Finance": "https://finance.yahoo.com/rss/topstories",
    "CNBC Business": "https://www.cnbc.com/id/10001147/device/rss/rss.html",
    "Financial Times": "https://www.ft.com/?format=rss",
    "BBC Business": "http://feeds.bbci.co.uk/news/business/rss.xml",
    "CNN Politics": "http://rss.cnn.com/rss/cnn_allpolitics.rss",
}

"""
Zero-shot scores rarely concentrate on a single label when a headline
plausibly fits several relevant categories at once (e.g. "business" vs
"economics" vs "finance"), so genuinely relevant stories often have no
single label clearing a high bar even though their combined relevant mass
is high. We therefore sum scores across all RELEVANT_TOPICS labels and
compare that combined mass against MIN_RELEVANT_MASS, rather than gating
on the single top label clearing MIN_TOPIC_SCORE 
"""
MIN_RELEVANT_MASS = 0.35

"""
Kept for the cluster-name relevance re-check, which only has a short
LLM-generated phrase to work with and benefits from a slightly looser bar.
"""
MIN_TOPIC_SCORE = 0.35

DISCORD_MAX_LEN = 2000 # discord api limit on chars in a single message 

BASE_DIR    = os.path.dirname(os.path.abspath(__file__))
LOGS_DIR    = os.path.join(BASE_DIR, "logs")
REPORTS_DIR = os.path.join(BASE_DIR, "reports")

os.makedirs(LOGS_DIR, exist_ok=True)
os.makedirs(REPORTS_DIR, exist_ok=True)

LOG_DATE_STR = datetime.now().strftime("%Y-%m-%d")
LOG_PATH = os.path.join(LOGS_DIR, f"Sentibot_{LOG_DATE_STR}.log")
logging.basicConfig(filename=LOG_PATH, level=logging.INFO, format="%(asctime)s  %(levelname)-8s  %(message)s", datefmt="%Y-%m-%d %H:%M:%S", encoding="utf-8", filemode="w")
logger = logging.getLogger("Sentibot")