lukeandClaude Opus 4.6
1c7ac334b4
Previously _build_backgrounds ran two parallel Sonnet calls of 6 callers each, with zero cross-batch awareness. Each batch independently gravitated to the same quirky archetypes (Coast to Coast weirdos, specific hobbies) and produced clustered rosters — one recent show had two BBQ competitors, two taxidermists, and two conspiracy-pattern callers across 10 slots. Collapses to a single Sonnet call generating all 10 at once so the model can self-diversify across the full roster. The BATCH_SYSTEM_PROMPT also gets an explicit ANTI-COLLISION RULE with concrete forbidden examples (no two BBQ competitors, no two taxidermists, no two mystery- signal callers, etc.) and an instruction to scan output for collisions before finalizing. max_tokens bumped 8000 -> 16000 to accommodate the larger response. Wall-clock cost: ~30s -> ~60s, acceptable for the diversity gain. Verified on a fresh reset: 10 fully differentiated callers, zero archetype collisions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
190 lines
8.4 KiB
Python
190 lines
8.4 KiB
Python
from dataclasses import dataclass
|
|
from typing import Optional
|
|
import json
|
|
|
|
import httpx
|
|
|
|
from ..config import settings
|
|
from .cost_tracker import cost_tracker
|
|
|
|
BATCH_MODEL = "anthropic/claude-sonnet-4.6"
|
|
|
|
REQUIRED_FIELDS = {
|
|
"name", "age", "voice_suggestion", "location", "identity",
|
|
"situation", "reason_calling", "opening_line", "secret_want",
|
|
"specific_details", "emotional_register"
|
|
}
|
|
|
|
|
|
@dataclass
|
|
class CallerIdentity:
|
|
name: str
|
|
age: int
|
|
voice_suggestion: str
|
|
location: str
|
|
identity: str
|
|
situation: str
|
|
reason_calling: str
|
|
opening_line: str
|
|
secret_want: str
|
|
specific_details: list[str]
|
|
emotional_register: str
|
|
# set after voice validation
|
|
voice_resolved: Optional[str] = None
|
|
|
|
|
|
def parse_batch_response(raw: str) -> list[CallerIdentity]:
|
|
stripped = raw.strip()
|
|
if stripped.startswith("```") and stripped.endswith("```"):
|
|
lines = stripped.splitlines()
|
|
stripped = "\n".join(lines[1:-1])
|
|
data = json.loads(stripped)
|
|
callers = data.get("callers", [])
|
|
result = []
|
|
for c in callers:
|
|
missing = REQUIRED_FIELDS - set(c.keys())
|
|
if missing:
|
|
raise ValueError(f"CallerIdentity missing fields: {missing}")
|
|
result.append(CallerIdentity(**{k: c[k] for k in REQUIRED_FIELDS}))
|
|
return result
|
|
|
|
|
|
def resolve_voice(suggestion: str, roster: list[str]) -> str:
|
|
"""Map sonnet's voice suggestion to a real voice in the roster.
|
|
Case-insensitive exact match; deterministic fallback to first roster entry."""
|
|
if not suggestion or not roster:
|
|
return roster[0] if roster else ""
|
|
lower_map = {v.lower(): v for v in roster}
|
|
return lower_map.get(suggestion.lower(), roster[0])
|
|
|
|
|
|
BATCH_SYSTEM_PROMPT = """You are writing a roster of callers for Luke's late-night radio show in New Mexico.
|
|
|
|
CREATIVE RANGE: Your callers must span the emotional range of Howard Stern (chaos, strong characters), Coast to Coast AM (earnest weirdos, sincere believers), Loveline (real problems, real advice-seeking), Delilah (emotional vulnerability, connection), and Opie and Anthony (sharp, irreverent, specific people).
|
|
|
|
Maximum character distance between callers. No two callers should feel like siblings.
|
|
|
|
ANTI-COLLISION RULE — THIS IS NON-NEGOTIABLE: All callers in this roster must be clearly differentiated. No two callers may share the same hobby, obsession, profession archetype, or story theme. Specifically forbidden within a single roster: two BBQ competitors, two taxidermists, two amateur-radio or mystery-signal callers, two callers with ex-spouse drama, two conspiracy/pattern-seers, two retired military, two grandmothers-of-many, two callers calling about a weird neighbor, two callers with religious-object stories. Each caller's defining "thing" — their hook, their obsession, the specific topic they're calling about — must appear exactly once in the roster. Before finalizing, scan your output and swap any collisions.
|
|
|
|
Do not default to sitcom plots. Real humans are specific and strange. Give each caller details that could only belong to them.
|
|
|
|
You will output strict JSON with a "callers" array. Each caller has exactly these fields: name, age, voice_suggestion, location, identity, situation, reason_calling, opening_line, secret_want, specific_details (array of 2-3 strings), emotional_register."""
|
|
|
|
|
|
def build_batch_prompt(ctx: dict) -> str:
|
|
lines = [
|
|
f"Tonight is {ctx['date']}. {ctx['weather']}.",
|
|
"",
|
|
"Today's news headlines (ground callers in real context, but do not force topicality):",
|
|
]
|
|
for h in ctx["headlines"]:
|
|
lines.append(f"- {h}")
|
|
lines.append("")
|
|
|
|
if ctx["recent_caller_summaries"]:
|
|
lines.append("Recent callers (DO NOT repeat these archetypes or situations):")
|
|
for s in ctx["recent_caller_summaries"]:
|
|
lines.append(f"- {s}")
|
|
lines.append("")
|
|
|
|
if ctx["regulars_included"]:
|
|
lines.append("RECURRING CHARACTERS IN TONIGHT'S LINEUP:")
|
|
lines.append("")
|
|
for r in ctx["regulars_included"]:
|
|
lines.append(f"### {r['name']}")
|
|
lines.append(r["lore"])
|
|
lines.append(f"Current arc state: {r['arc_state']}")
|
|
lines.append("")
|
|
lines.append(f"For {r['name']}: invent a fresh reason he is calling tonight — a new development, grievance, or specific recent event. DO NOT alter his voice, personality, or core traits. Write a new scene for an existing character.")
|
|
lines.append(f"CRITICAL — this caller MUST appear in the output JSON with EXACTLY these fields locked: name=\"{r['name']}\", voice_suggestion=\"{r['voice']}\", age={r['age']}. Do NOT rename this caller. Do NOT give him a different voice. Do NOT change his age. Only invent: location, identity, situation, reason_calling, opening_line, secret_want, specific_details, emotional_register.")
|
|
lines.append("")
|
|
|
|
lines.append(f"Available voices (voice_suggestion must match one of these exactly):")
|
|
lines.append(", ".join(ctx["voice_roster"]))
|
|
lines.append("")
|
|
lines.append(f"Generate {ctx['caller_count']} callers. Output JSON only, no prose.")
|
|
return BATCH_SYSTEM_PROMPT + "\n\n" + "\n".join(lines)
|
|
|
|
|
|
async def generate_batch(ctx: dict) -> list[CallerIdentity]:
|
|
"""Call sonnet-4.6 with the batch prompt, parse + voice-resolve the response."""
|
|
prompt = build_batch_prompt(ctx)
|
|
async with httpx.AsyncClient(timeout=120.0) as client:
|
|
resp = await client.post(
|
|
"https://openrouter.ai/api/v1/chat/completions",
|
|
headers={"Authorization": f"Bearer {settings.openrouter_api_key}"},
|
|
json={
|
|
"model": BATCH_MODEL,
|
|
"messages": [{"role": "user", "content": prompt}],
|
|
"response_format": {"type": "json_object"},
|
|
"max_tokens": 16000,
|
|
"temperature": 0.9,
|
|
},
|
|
)
|
|
resp.raise_for_status()
|
|
data = resp.json()
|
|
|
|
content = data["choices"][0]["message"]["content"]
|
|
usage = data.get("usage", {})
|
|
cost_tracker.record_llm_call(
|
|
category="background_gen",
|
|
model=BATCH_MODEL,
|
|
usage_data=usage,
|
|
)
|
|
|
|
callers = parse_batch_response(content)
|
|
for c in callers:
|
|
c.voice_resolved = resolve_voice(c.voice_suggestion, ctx["voice_roster"])
|
|
return callers
|
|
|
|
|
|
REGULAR_SYSTEM_PROMPT = """You are writing a single caller — a recurring character on Luke's late-night radio show. The character's identity is fixed (name, voice, age, core lore). You only invent a fresh reason he's calling tonight, grounded in his current arc state.
|
|
|
|
Output strict JSON with these fields only: location, identity, situation, reason_calling, opening_line, secret_want, specific_details (array of 2-3 strings), emotional_register. Do NOT include name, voice_suggestion, or age — those are locked."""
|
|
|
|
|
|
async def generate_regular_situation(regular: dict, ctx: dict) -> dict:
|
|
"""Generate a fresh situation for a locked regular. Returns a dict matching
|
|
the CallerIdentity schema. One small LLM call (~$0.01)."""
|
|
lines = [
|
|
f"Tonight is {ctx['date']}. {ctx['weather']}.",
|
|
"",
|
|
f"### {regular['name']}",
|
|
regular["lore"],
|
|
f"Current arc state: {regular['arc_state']}",
|
|
"",
|
|
f"Invent a fresh reason {regular['name']} is calling tonight — a new development, grievance, or specific recent event. DO NOT alter his voice, personality, or core traits.",
|
|
"",
|
|
"Output JSON only, no prose.",
|
|
]
|
|
prompt = REGULAR_SYSTEM_PROMPT + "\n\n" + "\n".join(lines)
|
|
|
|
async with httpx.AsyncClient(timeout=60.0) as client:
|
|
resp = await client.post(
|
|
"https://openrouter.ai/api/v1/chat/completions",
|
|
headers={"Authorization": f"Bearer {settings.openrouter_api_key}"},
|
|
json={
|
|
"model": BATCH_MODEL,
|
|
"messages": [{"role": "user", "content": prompt}],
|
|
"response_format": {"type": "json_object"},
|
|
"max_tokens": 1500,
|
|
"temperature": 0.9,
|
|
},
|
|
)
|
|
resp.raise_for_status()
|
|
data = resp.json()
|
|
|
|
content = data["choices"][0]["message"]["content"]
|
|
usage = data.get("usage", {})
|
|
cost_tracker.record_llm_call(
|
|
category="background_gen",
|
|
model=BATCH_MODEL,
|
|
usage_data=usage,
|
|
)
|
|
|
|
stripped = content.strip()
|
|
if stripped.startswith("```") and stripped.endswith("```"):
|
|
lines = stripped.splitlines()
|
|
stripped = "\n".join(lines[1:-1])
|
|
return json.loads(stripped)
|