"""Paths and shared constants for search evaluation."""
  
  from pathlib import Path
  
  _PKG_DIR = Path(__file__).resolve().parent
  _SCRIPTS_EVAL_DIR = _PKG_DIR.parent
  PROJECT_ROOT = _SCRIPTS_EVAL_DIR.parents[1]
  

  # Canonical English labels (must match LLM prompt output in prompts._CLASSIFY_TEMPLATE_EN)
  RELEVANCE_EXACT = "Exact Match"
  RELEVANCE_HIGH = "High Relevant"
  RELEVANCE_LOW = "Low Relevant"

  RELEVANCE_IRRELEVANT = "Irrelevant"

  
  VALID_LABELS = frozenset({RELEVANCE_EXACT, RELEVANCE_HIGH, RELEVANCE_LOW, RELEVANCE_IRRELEVANT})
  
  # Precision / MAP "positive" set (all non-irrelevant tiers)
  RELEVANCE_NON_IRRELEVANT = frozenset({RELEVANCE_EXACT, RELEVANCE_HIGH, RELEVANCE_LOW})
  
  _LEGACY_LABEL_MAP = {
      "Exact": RELEVANCE_EXACT,
      "Partial": RELEVANCE_HIGH,
  }
  
  
  def normalize_stored_label(label: str) -> str:
      """Map legacy 3-way SQLite labels to current 4-way strings; pass through canonical labels."""
      s = str(label).strip()
      if s in VALID_LABELS:
          return s
      return _LEGACY_LABEL_MAP.get(s, s)
  

  
  DEFAULT_ARTIFACT_ROOT = PROJECT_ROOT / "artifacts" / "search_evaluation"
  DEFAULT_QUERY_FILE = _SCRIPTS_EVAL_DIR / "queries" / "queries.txt"
  

  # Judge LLM (eval_framework only; override via CLI --judge-model / constructor kwargs)
  DEFAULT_JUDGE_MODEL = "qwen3.5-flash"
  DEFAULT_JUDGE_ENABLE_THINKING = True
  DEFAULT_JUDGE_DASHSCOPE_BATCH = True
  DEFAULT_JUDGE_BATCH_COMPLETION_WINDOW = "24h"
  DEFAULT_JUDGE_BATCH_POLL_INTERVAL_SEC = 10.0

  
  # Rebuild annotation pool (build --force-refresh-labels): search recall + full-corpus rerank + LLM batches
  DEFAULT_SEARCH_RECALL_TOP_K = 500
  DEFAULT_RERANK_HIGH_THRESHOLD = 0.5
  DEFAULT_RERANK_HIGH_SKIP_COUNT = 1000
  DEFAULT_REBUILD_LLM_BATCH_SIZE = 50

  DEFAULT_REBUILD_MIN_LLM_BATCHES = 20

  DEFAULT_REBUILD_MAX_LLM_BATCHES = 40
  DEFAULT_REBUILD_IRRELEVANT_STOP_RATIO = 0.92
  DEFAULT_REBUILD_IRRELEVANT_STOP_STREAK = 3