backend/core/analysis_knobs.py

from __future__ import annotations

from typing import Any, List, Optional


DEFAULT_BOTTOM_UP_INSTRUCTIONS: str = (
    "You are performing bottom-up thematic analysis. Identify the most prominent themes in the conversation, with no a priori concepts; the themes are emergent from the text."
)

DEFAULT_TOP_DOWN_INSTRUCTIONS: str = (
    "You are performing top-down coding using the provided codebook categories. For each category, extract the most relevant codes/items supported by evidence."
)

DEFAULT_RUBRIC_INSTRUCTIONS: str = (
    "You are performing a care experience rubric analysis. Summarize the overall experience and list the key reasons for each bucket."
)


DEFAULT_BOTTOM_UP_ATTRIBUTES: List[str] = [
    "Prefer fewer, higher-confidence items.",
    "For themes: include a short code label (1-3 words) in addition to the longer summary.",
]

DEFAULT_RUBRIC_ATTRIBUTES: List[str] = [
    "Prefer fewer, higher-confidence items.",
    "For care experience: do not duplicate the same evidence_id across positive/negative/mixed/neutral. If a sentence supports both positive and negative interpretations, put it in care_experience.mixed.",
]

DEFAULT_TOP_DOWN_ATTRIBUTES: List[str] = [
    "Prefer fewer, higher-confidence items.",
    "For top_down_codebook categories: include a short code label (1-3 words) and cite evidence.",
]


def normalize_analysis_attributes(attributes: Any) -> List[str]:
    if not isinstance(attributes, list):
        return []
    cleaned = []
    for item in attributes:
        if not isinstance(item, str):
            continue
        stripped = item.strip()
        if stripped:
            cleaned.append(stripped)
    return cleaned


def compile_analysis_rules_block(attributes: Any, *, defaults: Optional[List[str]] = None) -> str:
    rules = normalize_analysis_attributes(attributes)
    if not rules:
        rules = list(defaults or [])
    bullets = "\n".join(f"- {line}" for line in rules)
    return "Rules:\n" + bullets + "\n"