score.py

"""AI scoring of pickup lines via Claude.

Composite = 0.20 * creativity + 0.30 * response_quality + 0.50 * hookup_probability
"""
from __future__ import annotations

import json
import logging
import re
import time
from typing import Optional

import anthropic

from store import Store

log = logging.getLogger(__name__)

SCORE_MODEL = "claude-haiku-4-5-20251001"

SCORE_PROMPT = """\
Score this pickup line/opener on three dimensions. Return JSON only.

Pickup line: "{line_text}"
Other person's response: {response}
Outcome hint: {outcome_hint}
Context / platform: {context}

Score each dimension 0.0–10.0:

creativity:
  How original, witty, or contextually clever is the line?
  0-3 = Generic/overused ("Are you tired? You've been running through my mind")
  4-6 = Decent, moderately creative
  7-10 = Highly original, personalized, genuinely clever

response_quality:
  How positively did the other person respond?
  0 = No reply / blocked / negative
  1-3 = Cold, one-word, disinterested
  4-6 = Neutral engagement
  7-9 = Enthusiastic, laughing, asking questions back
  10 = Exceptional (immediate date suggestion, very enthusiastic)
  If response is unknown → 5.0

hookup_probability:
  Estimated 0-10 probability this interaction leads to a date or hookup.
  Base on response_quality + outcome_hint + overall vibe.
  0 = Definitely not (blocked/negative)
  5 = Unknown/neutral
  10 = Confirmed hookup/date or very high probability

Return ONLY valid JSON — no markdown, no explanation:
{{"creativity": X.X, "response_quality": X.X, "hookup_probability": X.X}}
"""


def score_line(client: anthropic.Anthropic, line: dict) -> Optional[dict]:
    prompt = SCORE_PROMPT.format(
        line_text=line.get("line_text", ""),
        response=repr(line.get("response")) if line.get("response") else "unknown",
        outcome_hint=repr(line.get("outcome_hint")) if line.get("outcome_hint") else "unknown",
        context=line.get("context") or "dating app",
    )
    try:
        msg = client.messages.create(
            model=SCORE_MODEL,
            max_tokens=128,
            messages=[{"role": "user", "content": prompt}],
        )
        raw = msg.content[0].text.strip()
        raw = re.sub(r"^```(?:json)?\s*|\s*```$", "", raw, flags=re.MULTILINE).strip()
        scores = json.loads(raw)
        for k in ("creativity", "response_quality", "hookup_probability"):
            scores[k] = max(0.0, min(10.0, float(scores.get(k, 5.0))))
        return scores
    except Exception as e:
        log.warning("Scoring failed for line %s: %s", line.get("id"), e)
        return None


def score_all(
    store: Store,
    client: anthropic.Anthropic,
    limit: int = 200,
    delay: float = 0.3,
) -> int:
    """Score all unscored lines. Returns count scored."""
    unscored = store.unscored_lines(limit=limit)
    log.info("Scoring %d unscored lines with %s", len(unscored), SCORE_MODEL)
    count = 0
    for line in unscored:
        scores = score_line(client, line)
        if scores:
            store.upsert_score(line["id"], scores, SCORE_MODEL)
            count += 1
        time.sleep(delay)
    return count