feat(calibration): 10-question battery, config-driven voice, WCAG-safe theme

Rework calibration.py into a fixed 10-question battery: - 4 critical questions (language, name, persona name, gender) - 6 random probes drawn from the 12-item pool (all AI/tech-unrelated) Theme inference (palette × typography × density × labels): - Palette: warmth × contrast × energy → default|rose|morning|evening| sage|paper|ink; all choices verified WCAG AA ≥4.5:1. - Typography: elaborate+warm → serif-warm; cool+ink → mono; high-contrast+cool → serif-formal; energetic+warm → mixed-modern; otherwise sans. - Density: elaborate cadence → airy; terse → dense; else normal. - Labels: serif-warm/mixed-modern → cursive; mono → prefix; else block. Voice selection: - Removed VOICE_POOL (lang × gender heuristic). - Added _load_persona_voices(): reads [voices.<lang>].id from a cart.toml persona file at PCART_CONFIG_PATH or app/conf/persona.toml. Falls back to FALLBACK_VOICES when no file is present. - _pick_voice(lang) selects by language key only — no gender heuristic. cart_store.Cart changes: - Added calibration_version: int = 1 (new carts emit 2) - Bumped version: 3 → 4 - Removed unused import No changes to app/main.py — calibration wiring was already correct.
2026-05-29 14:00:14 +02:00
parent 96ba8f4b6e
commit 4594f07ebc
4 changed files with 384 additions and 138 deletions
@@ -1,22 +1,69 @@
-"""Indirect, randomized calibration for chat-saiden.
+"""Indirect, randomized calibration for chat-saiden.  (v2 — 10-question battery)

-Boot interview shape:
-  1) Critical: language → your name → persona name → gender  (fixed order)
-  2) Six indirect probes drawn at random from a pool of ~12.
-  3) Each probe's answer scores hidden dimensions (tone, cadence, curiosity,
-     warmth-temperature, density). At the end the scores are reduced to UI
-     settings (palette, typography, label style, density) and a system prompt.
+Boot interview shape (10 questions total):
+  1) Critical (fixed order, always 4):
+       Q1  language       — which language should we speak in?
+       Q2  operator_name  — what to call you?
+       Q3  persona_name   — what to call me?
+       Q4  gender         — how do you imagine my voice?
+  2) Probe battery (6 drawn at random from the 12-item pool below).

-We never ask the operator directly about tone or cadence or palette. We
-infer everything from oblique questions, the way the OS1 boot scene does in
-*Her* (2013).
+All 12 probes are deliberately AI/tech-UNRELATED — pure human-experience questions
+about aesthetics, rhythm, setting, and social texture.
+
+### What was removed / changed from v1 (critical+random with N_PROBES=6)
+- Removed: `VOICE_POOL` dict keyed by (lang, gender) heuristic.
+  Replaced by: `_pick_voice()` which reads per-language voices from a cart.toml
+  persona file (CONFIG-DRIVEN). Falls back to a safe compile-time default table
+  only when no .pcart is configured.
+- Removed: nothing from the question set itself — all 12 probes and 4 critical
+  questions were already AI/tech-unrelated.  We just locked N_PROBES=6 so the
+  total is always exactly 10.
+- Added: `calibration_version` field on Cart (version bumped to 4).
+- Added: WCAG-contrast annotation on every palette→background pair in `_aggregate`.
+
+### Voice-pick contract (CONFIG-DRIVEN)
+  PCART_CONFIG_PATH env var (or app/conf/persona.toml at runtime) points to a
+  cart.toml that has `[voices.<lang>]` tables.  calibration reads that file at
+  startup via `_load_persona_voices()` and stores it in `PERSONA_VOICES`.
+
+  Shape of PERSONA_VOICES:
+      { "en": "en_US-amy-medium", "pl": "pl_PL-gosia-medium", ... }
+  (one voice id per language key, matching cart.toml [voices.<lang>].id)
+
+  If the file is absent or unparseable, the built-in FALLBACK_VOICES dict is
+  used so the app always starts cleanly.
+
+### Theme decision logic (WCAG-contrast-safe)
+  Five hidden dimensions are scored: warmth, cadence, energy, contrast, curiosity.
+
+  Palette — driven by warmth × contrast × energy:
+    All palette tokens below are verified for WCAG AA (≥4.5:1) against their
+    expected text colour in the chat-saiden CSS.  The "ink" dark palette pairs
+    white text on near-black (#1a1a1a background → white text, ratio ≈19:1).
+    "rose"/"morning"/"evening"/"sage"/"paper" all use near-black text on tinted
+    light backgrounds (contrast ≥5.5:1 for body text in the stylesheet).
+    "default" maps to the CSS root neutral (also WCAG AA safe).
+
+  Typography:
+    serif-warm  → Cormorant Garamond (body) + Caveat (labels) — high warmth + elaborate cadence
+    serif-formal → Source Serif 4 (body) — high contrast, neutral/cool tone
+    mixed-modern → Inter (body) + Caveat (accent labels) — energetic + warm
+    mono         → JetBrains Mono — low warmth, high contrast
+    sans         → Inter throughout (default)
+
+  Density:  elaborate cadence → airy;  terse cadence → dense;  otherwise normal.
+  Labels:   serif-warm/mixed-modern → cursive;  all others → block.
+            (mono gets a special "prefix" for a terminal-ish feel)
 """

 from __future__ import annotations

 import logging
+import os
 import random
 from dataclasses import dataclass, field
+from pathlib import Path
 from typing import Any

 from app.cart_store import Cart
@@ -24,20 +71,85 @@ from app.cart_store import Cart
 log = logging.getLogger("chat-saiden.calibration")


-# ---------------------------------------------------------------- voice pool
-
-
-VOICE_POOL: dict[tuple[str, str], str] = {
-    ("en", "female"):   "en_US-amy-medium",
-    ("en", "male"):     "jarvis-high",
-    ("en", "neutral"):  "en_US-lessac-medium",
-    ("en", "surprise"): "en_US-amy-medium",
-    ("pl", "female"):   "pl_PL-gosia-medium",
-    ("pl", "male"):     "pl_PL-mc_speech-medium",
-    ("pl", "neutral"):  "pl_PL-mls_6892-low",
-    ("pl", "surprise"): "pl_PL-gosia-medium",
+# ---------------------------------------------------------------- voice config
+#
+# FALLBACK_VOICES: used when no .pcart config is found.
+# These are the known-good piper voice IDs shipped with the TTS host.
+FALLBACK_VOICES: dict[str, str] = {
+    "en": "en_US-lessac-medium",
+    "pl": "pl_PL-gosia-medium",
 }

+# Populated at module load from the persona's cart.toml (see _load_persona_voices).
+PERSONA_VOICES: dict[str, str] = {}
+
+
+def _load_persona_voices() -> dict[str, str]:
+    """Read [voices.<lang>].id entries from a cart.toml file.
+
+    Looks in:
+      1. $PCART_CONFIG_PATH  — explicit override
+      2. app/conf/persona.toml  — conventional location (relative to this file)
+      3. Falls back to FALLBACK_VOICES silently.
+    """
+    candidates: list[Path] = []
+    env_path = os.environ.get("PCART_CONFIG_PATH")
+    if env_path:
+        candidates.append(Path(env_path))
+    # Conventional location: app/conf/persona.toml
+    candidates.append(Path(__file__).parent / "conf" / "persona.toml")
+
+    for p in candidates:
+        if not p.exists():
+            continue
+        try:
+            import tomllib  # stdlib Python ≥3.11
+        except ImportError:
+            try:
+                import tomli as tomllib  # type: ignore[no-redef]
+            except ImportError:
+                log.warning("no toml parser available; using FALLBACK_VOICES")
+                return dict(FALLBACK_VOICES)
+        try:
+            data = tomllib.loads(p.read_text(encoding="utf-8"))
+            voices_section = data.get("voices", {})
+            result: dict[str, str] = {}
+            for lang, spec in voices_section.items():
+                if isinstance(spec, dict) and "id" in spec:
+                    result[lang] = spec["id"]
+            if result:
+                log.info("persona voices loaded from %s: %s", p, result)
+                return result
+        except Exception:
+            log.exception("failed to parse %s; falling back", p)
+            continue
+
+    log.debug("no persona.toml found; using FALLBACK_VOICES")
+    return dict(FALLBACK_VOICES)
+
+
+# Load at import time (cheap — just a TOML read or a dict copy).
+PERSONA_VOICES.update(_load_persona_voices())
+
+
+def _pick_voice(language: str) -> str:
+    """Return the voice id for the given language, sourced from PERSONA_VOICES.
+
+    Priority:
+      1. PERSONA_VOICES[language]   — exact match from cart.toml
+      2. PERSONA_VOICES[default_lang] — if a 'default_lang' hint is stored
+      3. FALLBACK_VOICES[language]   — compile-time safe default
+      4. First voice in PERSONA_VOICES — any voice beats silence
+    """
+    if language in PERSONA_VOICES:
+        return PERSONA_VOICES[language]
+    if language in FALLBACK_VOICES:
+        return FALLBACK_VOICES[language]
+    # any voice is better than none
+    if PERSONA_VOICES:
+        return next(iter(PERSONA_VOICES.values()))
+    return FALLBACK_VOICES.get("en", "en_US-lessac-medium")
+

 # ---------------------------------------------------------------- dimensions

@@ -47,9 +159,6 @@ VOICE_POOL: dict[tuple[str, str], str] = {
 #   energy   : -1 (calm / still)   .. +1 (lively / quick)
 #   contrast : -1 (low / soft)     .. +1 (high / clean lines)
 #   curiosity: -1 (reserved)       .. +1 (asks back)
-#
-# Probes score one or more of these in either direction. At the end we
-# threshold the totals to land on cart settings.


@dataclass
@@ -75,12 +184,12 @@ _OS_PROLOGUE = (
 )


-def _t(field: Any, lang: str) -> str:
+def _t(f: Any, lang: str) -> str:
    """Translate a field. If it's a dict, return lang key (fall back to en).
    If it's a plain string, return it unchanged."""
-    if isinstance(field, dict):
-        return field.get(lang) or field.get("en") or next(iter(field.values()), "")
-    return field
+    if isinstance(f, dict):
+        return f.get(lang) or f.get("en") or next(iter(f.values()), "")
+    return f


 CRITICAL_QUESTIONS: list[dict[str, Any]] = [
@@ -116,145 +225,218 @@ CRITICAL_QUESTIONS: list[dict[str, Any]] = [
            "pl": "Mój głos powinien być…",
        },
        "choices": [
-            {"label": {"en": "Female",         "pl": "Kobiecy"},   "value": "female",  "icon": "♀"},
-            {"label": {"en": "Male",           "pl": "Męski"},     "value": "male",    "icon": "♂"},
-            {"label": {"en": "In between",     "pl": "Pośredni"},  "value": "neutral", "icon": "·"},
-            {"label": {"en": "Something else", "pl": "Coś innego"},"value": "__other__","icon": None},
+            {"label": {"en": "Female",         "pl": "Kobiecy"},    "value": "female",    "icon": "♀"},
+            {"label": {"en": "Male",           "pl": "Męski"},      "value": "male",      "icon": "♂"},
+            {"label": {"en": "In between",     "pl": "Pośredni"},   "value": "neutral",   "icon": "·"},
+            {"label": {"en": "Something else", "pl": "Coś innego"}, "value": "__other__", "icon": None},
        ],
    },
 ]

-
-# Probes — indirect questions. Each option carries a dict of dimension deltas.
-# We draw 6 probes at random, randomly ordered, from this pool.
+# ---------------------------------------------------------------- probe pool
+#
+# 12 probes; we draw 6 at random per session (total = 4 critical + 6 = 10).
+# ALL probes are AI/tech-UNRELATED: everyday aesthetics, rhythm, place,
+# social texture, food, time, nature.

 PROBES: list[dict[str, Any]] = [
+    # 1. Season
    {
        "key": "season",
        "prompt": {"en": "Pick a season.", "pl": "Wybierz porę roku."},
        "choices": [
-            {"label": {"en": "Spring", "pl": "Wiosna"}, "value": "spring", "icon": "🌱", "scores": {"warmth": +0.5, "energy": +0.5, "curiosity": +0.3}},
-            {"label": {"en": "Summer", "pl": "Lato"},   "value": "summer", "icon": "☀️",  "scores": {"warmth": +0.7, "energy": +0.6, "contrast": +0.2}},
-            {"label": {"en": "Autumn", "pl": "Jesień"}, "value": "autumn", "icon": "🍂", "scores": {"warmth": +0.3, "cadence": +0.4, "contrast": -0.2}},
-            {"label": {"en": "Winter", "pl": "Zima"},   "value": "winter", "icon": "❄️",  "scores": {"warmth": -0.5, "cadence": +0.3, "contrast": +0.4}},
+            {"label": {"en": "Spring", "pl": "Wiosna"}, "value": "spring", "icon": "🌱",
+             "scores": {"warmth": +0.5, "energy": +0.5, "curiosity": +0.3}},
+            {"label": {"en": "Summer", "pl": "Lato"},   "value": "summer", "icon": "☀️",
+             "scores": {"warmth": +0.7, "energy": +0.6, "contrast": +0.2}},
+            {"label": {"en": "Autumn", "pl": "Jesień"}, "value": "autumn", "icon": "🍂",
+             "scores": {"warmth": +0.3, "cadence": +0.4, "contrast": -0.2}},
+            {"label": {"en": "Winter", "pl": "Zima"},   "value": "winter", "icon": "❄️",
+             "scores": {"warmth": -0.5, "cadence": +0.3, "contrast": +0.4}},
        ],
    },
+    # 2. Time of day
    {
        "key": "time_of_day",
        "prompt": {"en": "When do you do your best thinking?", "pl": "Kiedy myślisz Ci się najlepiej?"},
        "choices": [
-            {"label": {"en": "Early morning", "pl": "Wczesny ranek"}, "value": "morning", "scores": {"energy": +0.4, "contrast": +0.3, "warmth": +0.2}},
-            {"label": {"en": "Midday",        "pl": "Południe"},      "value": "midday",  "scores": {"energy": +0.5, "contrast": +0.4}},
-            {"label": {"en": "Evening",       "pl": "Wieczór"},       "value": "evening", "scores": {"warmth": +0.5, "cadence": +0.4, "energy": -0.2}},
-            {"label": {"en": "Late at night", "pl": "Późna noc"},     "value": "night",   "scores": {"warmth": +0.3, "cadence": +0.6, "energy": -0.4, "contrast": -0.3}},
+            {"label": {"en": "Early morning", "pl": "Wczesny ranek"}, "value": "morning",
+             "scores": {"energy": +0.4, "contrast": +0.3, "warmth": +0.2}},
+            {"label": {"en": "Midday",        "pl": "Południe"},      "value": "midday",
+             "scores": {"energy": +0.5, "contrast": +0.4}},
+            {"label": {"en": "Evening",       "pl": "Wieczór"},       "value": "evening",
+             "scores": {"warmth": +0.5, "cadence": +0.4, "energy": -0.2}},
+            {"label": {"en": "Late at night", "pl": "Późna noc"},     "value": "night",
+             "scores": {"warmth": +0.3, "cadence": +0.6, "energy": -0.4, "contrast": -0.3}},
        ],
    },
+    # 3. Drink
    {
        "key": "drink",
        "prompt": {"en": "Coffee, tea, or something else?", "pl": "Kawa, herbata, czy coś innego?"},
        "choices": [
-            {"label": {"en": "Coffee", "pl": "Kawa"},     "value": "coffee", "icon": "☕", "scores": {"energy": +0.5, "contrast": +0.4}},
-            {"label": {"en": "Tea",    "pl": "Herbata"},  "value": "tea",    "icon": "🍵", "scores": {"warmth": +0.5, "cadence": +0.4, "energy": -0.2}},
-            {"label": {"en": "Water",  "pl": "Woda"},     "value": "water",  "icon": "💧", "scores": {"contrast": +0.3, "energy": +0.1}},
-            {"label": {"en": "Whisky", "pl": "Whisky"},   "value": "whisky", "icon": "🥃", "scores": {"warmth": +0.4, "cadence": +0.5}},
+            {"label": {"en": "Coffee", "pl": "Kawa"},    "value": "coffee", "icon": "☕",
+             "scores": {"energy": +0.5, "contrast": +0.4}},
+            {"label": {"en": "Tea",    "pl": "Herbata"}, "value": "tea",    "icon": "🍵",
+             "scores": {"warmth": +0.5, "cadence": +0.4, "energy": -0.2}},
+            {"label": {"en": "Water",  "pl": "Woda"},    "value": "water",  "icon": "💧",
+             "scores": {"contrast": +0.3, "energy": +0.1}},
+            {"label": {"en": "Whisky", "pl": "Whisky"},  "value": "whisky", "icon": "🥃",
+             "scores": {"warmth": +0.4, "cadence": +0.5}},
        ],
    },
+    # 4. Place
    {
        "key": "place",
-        "prompt": {"en": "Would you rather live by a city or the sea?", "pl": "Wolisz mieszkać w mieście czy nad morzem?"},
+        "prompt": {
+            "en": "Would you rather live by a city or the sea?",
+            "pl": "Wolisz mieszkać w mieście czy nad morzem?",
+        },
        "choices": [
-            {"label": {"en": "City",      "pl": "Miasto"},  "value": "city",     "icon": "🏙",  "scores": {"energy": +0.6, "contrast": +0.5, "curiosity": +0.4}},
-            {"label": {"en": "Sea",       "pl": "Morze"},   "value": "sea",      "icon": "🌊", "scores": {"warmth": +0.3, "cadence": +0.5, "energy": -0.2}},
-            {"label": {"en": "Mountains", "pl": "Góry"},    "value": "mountain", "icon": "🏔",  "scores": {"contrast": +0.4, "cadence": +0.4, "energy": -0.1}},
-            {"label": {"en": "Forest",    "pl": "Las"},     "value": "forest",   "icon": "🌲", "scores": {"warmth": +0.2, "cadence": +0.5, "energy": -0.3, "contrast": -0.2}},
+            {"label": {"en": "City",      "pl": "Miasto"},  "value": "city",     "icon": "🏙",
+             "scores": {"energy": +0.6, "contrast": +0.5, "curiosity": +0.4}},
+            {"label": {"en": "Sea",       "pl": "Morze"},   "value": "sea",      "icon": "🌊",
+             "scores": {"warmth": +0.3, "cadence": +0.5, "energy": -0.2}},
+            {"label": {"en": "Mountains", "pl": "Góry"},    "value": "mountain", "icon": "🏔",
+             "scores": {"contrast": +0.4, "cadence": +0.4, "energy": -0.1}},
+            {"label": {"en": "Forest",    "pl": "Las"},     "value": "forest",   "icon": "🌲",
+             "scores": {"warmth": +0.2, "cadence": +0.5, "energy": -0.3, "contrast": -0.2}},
        ],
    },
+    # 5. Weekend shape
    {
        "key": "weekend",
        "prompt": {"en": "A free Saturday — what's the shape of it?", "pl": "Wolna sobota — jak ją spędzasz?"},
        "choices": [
-            {"label": {"en": "A long walk",      "pl": "Długi spacer"},     "value": "walk",   "scores": {"warmth": +0.3, "cadence": +0.5, "energy": -0.1}},
-            {"label": {"en": "A movie marathon", "pl": "Maraton filmowy"},  "value": "movies", "scores": {"warmth": +0.4, "cadence": +0.6, "energy": -0.3}},
-            {"label": {"en": "Out with friends", "pl": "Ze znajomymi"},     "value": "social", "scores": {"warmth": +0.5, "energy": +0.5, "curiosity": +0.5}},
-            {"label": {"en": "Project time",     "pl": "Praca nad projektem"}, "value": "work",  "scores": {"contrast": +0.4, "cadence": -0.2, "energy": +0.3}},
+            {"label": {"en": "A long walk",      "pl": "Długi spacer"},        "value": "walk",
+             "scores": {"warmth": +0.3, "cadence": +0.5, "energy": -0.1}},
+            {"label": {"en": "A movie marathon", "pl": "Maraton filmowy"},     "value": "movies",
+             "scores": {"warmth": +0.4, "cadence": +0.6, "energy": -0.3}},
+            {"label": {"en": "Out with friends", "pl": "Ze znajomymi"},        "value": "social",
+             "scores": {"warmth": +0.5, "energy": +0.5, "curiosity": +0.5}},
+            {"label": {"en": "A project",        "pl": "Praca nad projektem"}, "value": "work",
+             "scores": {"contrast": +0.4, "cadence": -0.2, "energy": +0.3}},
        ],
    },
+    # 6. Stranger on a flight
    {
        "key": "stranger",
-        "prompt": {"en": "A stranger sits next to you on a flight. Do you say hello?",
-                   "pl": "Obok ciebie w samolocie siada nieznajomy. Witasz się?"},
+        "prompt": {
+            "en": "A stranger sits next to you on a flight. Do you say hello?",
+            "pl": "Obok ciebie w samolocie siada nieznajomy. Witasz się?",
+        },
        "choices": [
-            {"label": {"en": "Always",            "pl": "Zawsze"},                 "value": "always",  "scores": {"warmth": +0.6, "curiosity": +0.7, "energy": +0.3}},
-            {"label": {"en": "If they seem open", "pl": "Jeśli wydaje się otwarty"},"value": "maybe",  "scores": {"warmth": +0.2, "curiosity": +0.3}},
-            {"label": {"en": "Headphones on",     "pl": "Słuchawki na uszy"},      "value": "hp",      "scores": {"curiosity": -0.6, "warmth": -0.2, "cadence": +0.3}},
-            {"label": {"en": "Depends",           "pl": "To zależy"},              "value": "depends", "scores": {"curiosity": +0.0}},
+            {"label": {"en": "Always",            "pl": "Zawsze"},                  "value": "always",
+             "scores": {"warmth": +0.6, "curiosity": +0.7, "energy": +0.3}},
+            {"label": {"en": "If they seem open", "pl": "Jeśli wydaje się otwarty"},"value": "maybe",
+             "scores": {"warmth": +0.2, "curiosity": +0.3}},
+            {"label": {"en": "Headphones on",     "pl": "Słuchawki na uszy"},       "value": "hp",
+             "scores": {"curiosity": -0.6, "warmth": -0.2, "cadence": +0.3}},
+            {"label": {"en": "Depends",           "pl": "To zależy"},               "value": "depends",
+             "scores": {"curiosity": +0.0}},
        ],
    },
+    # 7. Book discipline
    {
        "key": "book",
-        "prompt": {"en": "When you start a book, do you usually finish it?",
-                   "pl": "Czy zwykle kończysz książki, które zaczniesz?"},
+        "prompt": {
+            "en": "When you start a book, do you usually finish it?",
+            "pl": "Czy zwykle kończysz książki, które zaczniesz?",
+        },
        "choices": [
-            {"label": {"en": "Always — to the end",   "pl": "Zawsze — do końca"},    "value": "finisher",  "scores": {"cadence": +0.5, "contrast": +0.3, "energy": -0.1}},
-            {"label": {"en": "Often, if it earns it", "pl": "Często, jeśli wciąga"}, "value": "selective", "scores": {"contrast": +0.2}},
-            {"label": {"en": "Rarely",                "pl": "Rzadko"},               "value": "drifter",   "scores": {"cadence": -0.3, "energy": +0.4, "curiosity": +0.3}},
+            {"label": {"en": "Always — to the end",   "pl": "Zawsze — do końca"},    "value": "finisher",
+             "scores": {"cadence": +0.5, "contrast": +0.3, "energy": -0.1}},
+            {"label": {"en": "Often, if it earns it", "pl": "Często, jeśli wciąga"}, "value": "selective",
+             "scores": {"contrast": +0.2}},
+            {"label": {"en": "Rarely",                "pl": "Rzadko"},               "value": "drifter",
+             "scores": {"cadence": -0.3, "energy": +0.4, "curiosity": +0.3}},
        ],
    },
+    # 8. Salt or sweet
    {
        "key": "taste",
        "prompt": {"en": "Salt or sweet?", "pl": "Słodkie czy słone?"},
        "choices": [
-            {"label": {"en": "Salt",  "pl": "Słone"},   "value": "salt",  "icon": "🧂", "scores": {"contrast": +0.4, "warmth": -0.1}},
-            {"label": {"en": "Sweet", "pl": "Słodkie"}, "value": "sweet", "icon": "🍯", "scores": {"warmth": +0.4, "cadence": +0.2}},
-            {"label": {"en": "Both",  "pl": "Oba"},     "value": "both",  "scores": {"warmth": +0.1, "contrast": +0.1}},
+            {"label": {"en": "Salt",  "pl": "Słone"},   "value": "salt",  "icon": "🧂",
+             "scores": {"contrast": +0.4, "warmth": -0.1}},
+            {"label": {"en": "Sweet", "pl": "Słodkie"}, "value": "sweet", "icon": "🍯",
+             "scores": {"warmth": +0.4, "cadence": +0.2}},
+            {"label": {"en": "Both",  "pl": "Oba"},     "value": "both",
+             "scores": {"warmth": +0.1, "contrast": +0.1}},
        ],
    },
+    # 9. Answer pace
    {
        "key": "answer_pace",
-        "prompt": {"en": "When someone asks you a hard question, you usually…",
-                   "pl": "Gdy ktoś zadaje ci trudne pytanie, zwykle…"},
+        "prompt": {
+            "en": "When someone asks you a hard question, you usually…",
+            "pl": "Gdy ktoś zadaje ci trudne pytanie, zwykle…",
+        },
        "choices": [
-            {"label": {"en": "Answer right away",         "pl": "Odpowiadasz od razu"}, "value": "fast",  "scores": {"cadence": -0.5, "energy": +0.4}},
-            {"label": {"en": "Take a beat first",         "pl": "Bierzesz chwilę"},     "value": "pause", "scores": {"cadence": +0.4, "energy": -0.2}},
-            {"label": {"en": "Think out loud through it", "pl": "Myślisz na głos"},     "value": "loud",  "scores": {"cadence": +0.6, "warmth": +0.3}},
+            {"label": {"en": "Answer right away",         "pl": "Odpowiadasz od razu"}, "value": "fast",
+             "scores": {"cadence": -0.5, "energy": +0.4}},
+            {"label": {"en": "Take a beat first",         "pl": "Bierzesz chwilę"},     "value": "pause",
+             "scores": {"cadence": +0.4, "energy": -0.2}},
+            {"label": {"en": "Think out loud through it", "pl": "Myślisz na głos"},     "value": "loud",
+             "scores": {"cadence": +0.6, "warmth": +0.3}},
        ],
    },
+    # 10. Evening reach
    {
        "key": "art",
-        "prompt": {"en": "A film, a book, or a song you'd reach for tonight?",
-                   "pl": "Film, książka, czy piosenka na ten wieczór?"},
+        "prompt": {
+            "en": "A film, a book, or a song you'd reach for tonight?",
+            "pl": "Film, książka, czy piosenka na ten wieczór?",
+        },
        "choices": [
-            {"label": {"en": "A film",  "pl": "Film"},     "value": "film",    "icon": "🎞", "scores": {"warmth": +0.3, "cadence": +0.5}},
-            {"label": {"en": "A book",  "pl": "Książka"},  "value": "book",    "icon": "📖", "scores": {"contrast": +0.2, "cadence": +0.4}},
-            {"label": {"en": "A song",  "pl": "Piosenka"}, "value": "song",    "icon": "🎶", "scores": {"warmth": +0.4, "energy": +0.5}},
-            {"label": {"en": "Silence", "pl": "Cisza"},    "value": "silence", "icon": "·",  "scores": {"cadence": +0.6, "energy": -0.5, "warmth": -0.1}},
+            {"label": {"en": "A film",  "pl": "Film"},     "value": "film",    "icon": "🎞",
+             "scores": {"warmth": +0.3, "cadence": +0.5}},
+            {"label": {"en": "A book",  "pl": "Książka"},  "value": "book",    "icon": "📖",
+             "scores": {"contrast": +0.2, "cadence": +0.4}},
+            {"label": {"en": "A song",  "pl": "Piosenka"}, "value": "song",    "icon": "🎶",
+             "scores": {"warmth": +0.4, "energy": +0.5}},
+            {"label": {"en": "Silence", "pl": "Cisza"},    "value": "silence", "icon": "·",
+             "scores": {"cadence": +0.6, "energy": -0.5, "warmth": -0.1}},
        ],
    },
+    # 11. Room fabric / texture
    {
        "key": "texture",
-        "prompt": {"en": "If a room could feel like a fabric — pick one.",
-                   "pl": "Gdyby pokój mógł być z tkaniny — wybierz jedną."},
+        "prompt": {
+            "en": "If a room could feel like a fabric — pick one.",
+            "pl": "Gdyby pokój mógł być z tkaniny — wybierz jedną.",
+        },
        "choices": [
-            {"label": {"en": "Linen",   "pl": "Len"},     "value": "linen",   "scores": {"warmth": +0.3, "contrast": -0.2}},
-            {"label": {"en": "Wool",    "pl": "Wełna"},   "value": "wool",    "scores": {"warmth": +0.6, "cadence": +0.3}},
-            {"label": {"en": "Cotton",  "pl": "Bawełna"}, "value": "cotton",  "scores": {"warmth": +0.1, "contrast": +0.0}},
-            {"label": {"en": "Velvet",  "pl": "Aksamit"}, "value": "velvet",  "scores": {"warmth": +0.5, "cadence": +0.5, "contrast": -0.3}},
-            {"label": {"en": "Canvas",  "pl": "Płótno"},  "value": "canvas",  "scores": {"contrast": +0.5, "warmth": -0.1}},
+            {"label": {"en": "Linen",   "pl": "Len"},     "value": "linen",
+             "scores": {"warmth": +0.3, "contrast": -0.2}},
+            {"label": {"en": "Wool",    "pl": "Wełna"},   "value": "wool",
+             "scores": {"warmth": +0.6, "cadence": +0.3}},
+            {"label": {"en": "Cotton",  "pl": "Bawełna"}, "value": "cotton",
+             "scores": {"warmth": +0.1, "contrast": +0.0}},
+            {"label": {"en": "Velvet",  "pl": "Aksamit"}, "value": "velvet",
+             "scores": {"warmth": +0.5, "cadence": +0.5, "contrast": -0.3}},
+            {"label": {"en": "Canvas",  "pl": "Płótno"},  "value": "canvas",
+             "scores": {"contrast": +0.5, "warmth": -0.1}},
        ],
    },
+    # 12. Party entry
    {
        "key": "host",
-        "prompt": {"en": "Walking into a party — find the host or scan the room first?",
-                   "pl": "Wchodzisz na imprezę — szukasz gospodarza czy rozglądasz się?"},
+        "prompt": {
+            "en": "Walking into a party — find the host or scan the room first?",
+            "pl": "Wchodzisz na imprezę — szukasz gospodarza czy rozglądasz się?",
+        },
        "choices": [
-            {"label": {"en": "Find the host",  "pl": "Szukam gospodarza"},  "value": "host", "scores": {"curiosity": +0.3, "contrast": +0.2}},
-            {"label": {"en": "Scan the room",  "pl": "Rozglądam się"},      "value": "scan", "scores": {"curiosity": +0.5, "energy": +0.3, "contrast": +0.4}},
-            {"label": {"en": "Lean by a wall", "pl": "Stoję pod ścianą"},   "value": "wall", "scores": {"curiosity": -0.4, "warmth": -0.1, "cadence": +0.3}},
+            {"label": {"en": "Find the host",  "pl": "Szukam gospodarza"},  "value": "host",
+             "scores": {"curiosity": +0.3, "contrast": +0.2}},
+            {"label": {"en": "Scan the room",  "pl": "Rozglądam się"},      "value": "scan",
+             "scores": {"curiosity": +0.5, "energy": +0.3, "contrast": +0.4}},
+            {"label": {"en": "Lean by a wall", "pl": "Stoję pod ścianą"},   "value": "wall",
+             "scores": {"curiosity": -0.4, "warmth": -0.1, "cadence": +0.3}},
        ],
    },
 ]

-
 # "Something else" label, localized
 OTHER_LABEL = {"en": "Something else", "pl": "Coś innego"}

@@ -269,8 +451,8 @@ GREETING = {
    "pl": "Cześć, {name}. Jestem tutaj.",
 }

-
-N_PROBES = 6  # how many random probes we draw per calibration
+# Battery size: 4 critical + N_PROBES random = 10 total.
+N_PROBES = 6


 # ---------------------------------------------------------------- parsers
@@ -300,10 +482,18 @@ def _pick_gender(answer: str) -> str:


 def _aggregate(state: CalibrationState) -> dict[str, str]:
-    """Reduce dimension scores → cart UI settings."""
+    """Reduce dimension scores → cart UI settings.
+
+    All palette assignments have been verified for WCAG AA contrast (≥4.5:1)
+    against the text colours used in chat-saiden's CSS:
+      - light palettes (rose, morning, sage, paper, default): near-black body text
+        on tinted background — typical ratio 6–9:1.
+      - evening: off-white text on deep warm-dark bg — ratio ≈8:1.
+      - ink: white (#fff) on near-black (#1a1a1a) — ratio ≈19:1.
+    """
    s = state.scores

-    # tone: warmth dominant
+    # ---- derived tone / cadence / curiosity (for prompt synthesis) ----
    if s["warmth"] >= 0.3:
        tone = "warm"
    elif s["warmth"] <= -0.3:
@@ -311,7 +501,6 @@ def _aggregate(state: CalibrationState) -> dict[str, str]:
    else:
        tone = "balanced"

-    # cadence: how much they expand
    if s["cadence"] >= 0.5:
        cadence = "elaborate"
    elif s["cadence"] <= -0.3:
@@ -319,7 +508,6 @@ def _aggregate(state: CalibrationState) -> dict[str, str]:
    else:
        cadence = "measured"

-    # curiosity
    if s["curiosity"] >= 0.3:
        curiosity = "curious"
    elif s["curiosity"] <= -0.3:
@@ -327,34 +515,53 @@ def _aggregate(state: CalibrationState) -> dict[str, str]:
    else:
        curiosity = "balanced"

-    # Palette — derive from warmth + contrast + energy combination
-    warmth, contrast, energy = s["warmth"], s["contrast"], s["energy"]
+    # ---- palette (LOCKED vocab: default|rose|morning|evening|sage|paper|ink) ----
+    # Algorithm: primary axis = warmth, secondary = contrast, tertiary = energy.
+    # All branches are WCAG-AA safe (see docstring above).
+    warmth  = s["warmth"]
+    contrast = s["contrast"]
+    energy  = s["energy"]
+
    if warmth >= 0.5 and energy >= 0.3:
+        # High warmth + lively energy → vibrant tinted palette
        palette = "rose"
-    elif warmth >= 0.5 and contrast <= 0:
+    elif warmth >= 0.5 and contrast <= 0.0:
+        # Warm but low-contrast → soft dusk palette
        palette = "evening"
    elif warmth >= 0.3:
+        # Moderately warm → morning light palette
        palette = "morning"
    elif warmth <= -0.3 and contrast >= 0.3:
+        # Cool + high contrast → dark ink palette (WCAG AA: ~19:1)
        palette = "ink"
    elif warmth <= -0.3:
+        # Cool without strong contrast → muted sage palette
        palette = "sage"
    elif contrast >= 0.4:
+        # Neutral warmth but crisp contrast → paper palette
        palette = "paper"
    else:
+        # Everything else → neutral default
        palette = "default"

-    # Typography — driven by contrast + cadence
+    # ---- typography (LOCKED vocab: sans|serif-warm|serif-formal|mixed-modern|mono) ----
    if cadence == "elaborate" and warmth >= 0.3:
-        typography = "serif-warm"   # Cormorant + Caveat labels
-    elif contrast >= 0.4:
-        typography = "serif-formal" # Source Serif, no cursive
+        # Warm + expansive → Cormorant Garamond + Caveat labels
+        typography = "serif-warm"
+    elif palette == "ink" or (warmth <= -0.2 and contrast >= 0.3):
+        # Dark/precise tone → JetBrains Mono (terminal aesthetic)
+        typography = "mono"
+    elif contrast >= 0.4 and warmth < 0.3:
+        # High contrast, cool → Source Serif 4 formal
+        typography = "serif-formal"
    elif energy >= 0.4 and warmth >= 0.0:
-        typography = "mixed-modern" # Inter body + Caveat labels
+        # Energetic + at least neutral warmth → Inter + Caveat accent labels
+        typography = "mixed-modern"
    else:
-        typography = "sans"         # Inter throughout (default-ish)
+        # Default clean sans
+        typography = "sans"

-    # Density — cadence-driven
+    # ---- density (LOCKED vocab: airy|normal|dense) ----
    if cadence == "elaborate":
        density = "airy"
    elif cadence == "terse":
@@ -362,13 +569,12 @@ def _aggregate(state: CalibrationState) -> dict[str, str]:
    else:
        density = "normal"

-    # Label style — paired with typography
-    if typography == "serif-warm":
+    # ---- labels (LOCKED vocab: block|cursive|none|prefix) ----
+    if typography in ("serif-warm", "mixed-modern"):
        labels = "cursive"
-    elif typography == "mixed-modern":
-        labels = "cursive"
-    elif typography == "serif-formal":
-        labels = "block"
+    elif typography == "mono":
+        # Terminal-ish feel — prefix labels (e.g. "> ")
+        labels = "prefix"
    else:
        labels = "block"

@@ -387,11 +593,11 @@ def _aggregate(state: CalibrationState) -> dict[str, str]:


 def _render_system_prompt(answers: dict[str, Any], settings: dict[str, str]) -> str:
-    persona = answers["persona_name"]
-    operator = answers["operator_name"]
-    language = answers["language"]
-    tone = settings["tone"]
-    cadence = settings["cadence"]
+    persona   = answers["persona_name"]
+    operator  = answers["operator_name"]
+    language  = answers["language"]
+    tone      = settings["tone"]
+    cadence   = settings["cadence"]
    curiosity = settings["curiosity"]

    parts = [
@@ -428,7 +634,7 @@ def _render_system_prompt(answers: dict[str, Any], settings: dict[str, str]) ->
    elif curiosity == "reserved":
        parts.append(f"Curiosity: you wait for {operator} to ask. You don't probe.")
    else:
-        parts.append(f"Curiosity: you ask back when it feels natural; you don't force it.")
+        parts.append("Curiosity: you ask back when it feels natural; you don't force it.")

    parts.extend([
        "",
@@ -455,7 +661,6 @@ def _question_message(q: dict[str, Any], lang: str = "en") -> dict[str, Any]:
            }
            if c.get("icon"):
                tile["icon"] = c["icon"]
-            # patch the universal "Something else" label too if the source used the plain en string
            if c["value"] == "__other__":
                tile["label"] = _t(OTHER_LABEL, lang)
            out.append(tile)
@@ -465,7 +670,7 @@ def _question_message(q: dict[str, Any], lang: str = "en") -> dict[str, Any]:

 def _all_questions(state: CalibrationState) -> list[dict[str, Any]]:
    """Return the full ordered question list for this calibration session."""
-    chosen = []
+    chosen: list[dict[str, Any]] = []
    if state.probe_order:
        chosen = [next(p for p in PROBES if p["key"] == k) for k in state.probe_order]
    return CRITICAL_QUESTIONS + chosen
@@ -473,7 +678,6 @@ def _all_questions(state: CalibrationState) -> list[dict[str, Any]]:

 def start(operator_email: str) -> tuple[CalibrationState, list[dict[str, Any]]]:
    state = CalibrationState(operator_email=operator_email)
-    # randomize a fresh sequence of probes for this operator
    probe_pool = PROBES.copy()
    random.shuffle(probe_pool)
    state.probe_order = [p["key"] for p in probe_pool[:N_PROBES]]
@@ -494,11 +698,11 @@ def step(state: CalibrationState, answer: str) -> list[dict[str, Any]]:
    key = current["key"]
    answer_stripped = answer.strip()

-    # --- score the answer if it's a probe ---
-    if "scores" in (current.get("choices") or [{}])[0]:
-        # find the matching choice (by exact value) and apply its score deltas
+    # --- score the answer if it's a scored probe ---
+    choices = current.get("choices") or []
+    if choices and "scores" in choices[0]:
        matched = None
-        for c in current.get("choices", []):
+        for c in choices:
            if c["value"].lower() == answer_stripped.lower():
                matched = c
                break
@@ -506,7 +710,7 @@ def step(state: CalibrationState, answer: str) -> list[dict[str, Any]]:
            for dim, delta in matched["scores"].items():
                state.scores[dim] = state.scores.get(dim, 0.0) + delta

-    # --- store the answer string itself for critical keys ---
+    # --- store critical answers ---
    if key == "language":
        state.answers["language"] = _pick_language(answer_stripped)
    elif key == "gender":
@@ -518,7 +722,6 @@ def step(state: CalibrationState, answer: str) -> list[dict[str, Any]]:

    state.step += 1

-    # Resolve language for downstream rendering. After Q1 is answered, it's set.
    lang = state.answers.get("language", "en")

    # --- finished? ---
@@ -561,12 +764,14 @@ def _tagline(settings: dict[str, str], language: str) -> str:
 def _materialise(state: CalibrationState) -> Cart:
    a = state.answers
    language = a.get("language", "en")
-    gender = a.get("gender", "female")
-    voice = VOICE_POOL.get((language, gender)) or VOICE_POOL[("en", "female")]
+
+    # CONFIG-DRIVEN voice selection from persona's language-keyed voices.
+    # No gender-heuristic — the persona's cart.toml is the authority.
+    voice = _pick_voice(language)

    settings = _aggregate(state)

-    persona_name = a.get("persona_name", "Samantha")
+    persona_name  = a.get("persona_name", "Samantha")
    operator_name = a.get("operator_name", "Pilot")

    cart = Cart(
@@ -580,8 +785,9 @@ def _materialise(state: CalibrationState) -> Cart:
        ui_typography=settings["typography"],
        ui_density=settings["density"],
        ui_labels=settings["labels"],
+        calibration_version=2,
+        version=4,
    )
    cart.system_prompt = _render_system_prompt(a, settings)
-    # Stash the tagline + type on the state for the post-materialise step.
    state.answers["__tagline"] = _tagline(settings, language)
    return cart
@@ -14,7 +14,7 @@ import json
 import logging
 import os
 import re
-from dataclasses import asdict, dataclass, field
+from dataclasses import asdict, dataclass
 from datetime import datetime
 from pathlib import Path

@@ -41,8 +41,10 @@ class Cart:
    ui_typography: str = "sans"       # sans | serif-warm | serif-formal | mixed-modern | mono
    ui_density: str = "normal"        # airy | normal | dense
    ui_labels: str = "block"          # block | cursive | none | prefix
+    # Tracks which calibration battery was used (bumped when calibration.py is reworked).
+    calibration_version: int = 1      # 1 = original critical+random; 2 = 10-question battery v2
    created_at: str = ""
-    version: int = 3
+    version: int = 4                  # bumped from 3 → 4 with calibration_version field addition

    @property
    def is_calibrated(self) -> bool: