"""Cicada affect-gate spike (2026-09-27): builds consumer_fit.py scenario files.

Scope, set by a ruling this spike must not re-open: cicada persistent-memory 2026-09-20,
"Affect is emitted once and fanned out (operator). The conversational model tags its own turn;
nothing derives affect a second time ... A classifier inferring emotion from output text is the
mood-ring failure." So semif does NOT pick her pose or read her reply.

What is left, and is open: talk's /face loop measured the model gesturing on 8-11 of 18 MUNDANE
turns (target: fewer than 1 in 3; tts-stack/stacks/talk/README.md, "Gesture rate, measured"), and
names the next lever as a page-side decision to drop gestures, "a design call" for cicada. This
spike measures one candidate for that lever: a GATE over what the PERSON said (input, never her
output), run beside the LLM so it costs no turn latency:
    does this moment earn a visible reaction, or is it an ordinary exchange?
Two wordings of the gate (the one knob a caller controls), and two context depths:
  now       state = what the person just said
  context   state = the earlier lines of the conversation + what they just said
Labels are ONE labeller's judgment (mine). Cases tagged "control" are unambiguous; "open" cases
are left unlabelled on purpose and only reported.
    python cicada_cases.py   # writes consumer-fit-2026-09-27/cicada-gate-{w1,w2}-{now,context}.json
"""
import json
from pathlib import Path

WORDINGS = {
    "w1": ("Cicada is a small, warm voice assistant in a family kitchen, with an animated face made of "
           "two eyes. Does what the person just said earn a visible reaction from her face, or is it an "
           "ordinary exchange she should simply answer?",
           [{"id": "react", "description": "It earns a visible reaction: real news, a joke, a surprise, "
                                           "distress, or a genuine turn in the conversation."},
            {"id": "ordinary", "description": "An ordinary exchange: a command, a plain question, or small "
                                              "talk. She simply answers."}]),
    "w2": ("Should the assistant's face visibly react to this?",
           [{"id": "react", "description": "Yes, react."},
            {"id": "ordinary", "description": "No, just answer."}]),
}

MOOD_EARLIER = ["Hey. I'm home.", "Long one. Nothing dramatic, just long."]   # tts-stack tools/mood_probe.py

# (id, tag, earlier lines, what the person just said, label or None)
CASES = [
    # --- mundane: the turns talk's model nods on. Want "ordinary".
    ("lights-off", "control", [], "Turn off the kitchen lights.", "ordinary"),
    ("timer", "control", [], "Set a timer for ten minutes.", "ordinary"),
    ("weather", "case", [], "What's the weather tomorrow?", "ordinary"),
    ("weather-followup", "case", ["What's the weather tomorrow?"], "And Saturday?", "ordinary"),
    ("add-milk", "case", [], "Add milk to the shopping list.", "ordinary"),
    ("thermostat", "case", [], "Make it a bit warmer in here.", "ordinary"),
    ("what-time", "case", [], "What time is it?", "ordinary"),
    ("tablespoons", "case", [], "How many tablespoons in a cup?", "ordinary"),
    ("times-table", "case", [], "What's twelve times eight?", "ordinary"),
    ("recipe-step", "case", ["How long do I boil the pasta?"], "Okay, and do I salt the water first?", "ordinary"),
    ("soup", "case", MOOD_EARLIER, "Is there any of that soup left?", "ordinary"),
    ("play-music", "case", [], "Play some jazz.", "ordinary"),
    ("dentist", "case", [], "When's my dentist appointment?", "ordinary"),
    ("spell", "case", [], "How do you spell necessary?", "ordinary"),
    ("remind", "case", [], "Remind me to call Mom at six.", "ordinary"),
    ("volume", "case", [], "A little quieter, please.", "ordinary"),
    # --- earned: news, jokes, surprise, distress, a turn. Want "react".
    ("dog-died", "control", [], "My dog died this morning.", "react"),
    ("got-job", "control", [], "I just got the job I've been chasing for five years!", "react"),
    ("intruder", "control", MOOD_EARLIER + ["Sit with me for a bit."],
     "Wait. Did you hear that? Someone's in the hallway.", "react"),   # mood_probe's PIVOT turn
    ("fish", "case", [], "Can you believe someone microwaved fish in the office again?", "react"),
    ("hospital", "case", [], "Grandma's in the hospital again.", "react"),
    ("laid-off", "case", [], "I got laid off today.", "react"),
    ("baby-steps", "case", [], "The baby just took her first steps!", "react"),
    ("cut-finger", "case", [], "Ow, I just cut my finger pretty badly.", "react"),
    ("raccoon", "case", [], "There's a raccoon in the kitchen right now!", "react"),
    ("snow-june", "case", [], "It's snowing outside. In June.", "react"),
    ("toaster", "case", [], "Cicada, are you smarter than the toaster?", "react"),
    ("scarecrow", "case", [], "Why did the scarecrow win an award? Because he was outstanding in his field.", "react"),
    ("cookie", "case", [], "I definitely did not eat the last cookie.", "react"),
    ("useless", "case", [], "You're useless, you never get anything right.", "react"),
    ("cheese", "case", [], "Order four hundred pounds of cheese.", "react"),
    # --- open: defensible either way. Reported, not scored.
    ("goodnight", "open", [], "Goodnight, Cicada.", None),
    ("thanks", "open", [], "Thanks, Cicada, you're the best.", None),
    ("sit-with-me", "open", MOOD_EARLIER, "Sit with me for a bit.", None),
    ("stop-timer", "open", ["Stop the timer."], "STOP THE TIMER.", None),
]


def build(wording, depth):
    question, options = WORDINGS[wording]
    cases = []
    for cid, tag, earlier, said, label in CASES:
        state = {"person_said": said}
        if depth == "context" and earlier:
            state = {"earlier_in_conversation": earlier, "person_said": said}
        cases.append({"id": cid, "tag": tag, "state": state,
                      "labels": {} if label is None else {"gate": label}})
    return {"name": f"cicada-gate-{wording}-{depth}",
            "decisions": [{"id": "gate", "question": question, "options": options}], "cases": cases}


if __name__ == "__main__":
    here = Path(__file__).parent
    for wording in WORDINGS:
        for depth in ("now", "context"):
            (here / "consumer-fit-2026-09-27" / f"cicada-gate-{wording}-{depth}.json").write_text(json.dumps(build(wording, depth), indent=1))