feat(profile+pricing+voice): KI-275 — F-bundle: smoker + recap-verify + premium scope + price consolidation

F2 — Smoker slot end-to-end (KI-275):
- needs_finder.Profile: smoker: Optional[bool] field
- brain_tools._ACCEPTED_FIELDS + SLOT_UNION extended (now 16 slots).
_coerce_smoker accepts yes/true/smoker/tobacco→True, no/non-smoker→False,
empty/unclear→None.
- single_brain SYSTEM_PROMPT RULE 2.5 item 7: ask "Do you smoke or use
tobacco?" Save smoker: save_profile_field(field="smoker", value="yes"/"no").
- premium_calculator bulk_estimate threads smoker_mult (1.4× when True)
through both curated-anchor + flat-base paths. Surfaces
smoker_loading_x + smoker_loading_reason in breakdown when !=1.0.
- Premium delta: 35yo metro 10L SI no-PED → ₹11,500 → ₹14,300 (+24% via
curated 1.35×) / ₹9,000 → ₹12,600 (+40% flat fallback).

F3 — Recap-vs-reality fix:
- single_brain RULE 4 new "RECAP VERIFY" sub-section: forbids recapping
slots not actually saved via save_profile_field. Worked example for the
diabetes/family_medical_history split (don't conflate them).
- main.py post-turn confirmation auto-extract: when user_text matches
yes/correct/that's-right/looks-good regex AND profile_complete=False,
parses the bot's PRIOR turn for `* **Label:** value` patterns, maps
via _SLOT_ALIASES (Name / Age / Dependents / Income / Primary Goal /
Health Conditions etc.) to _REQUIRED_FOR_READY slots, and BACKFILLS
via brain_tools.save_profile_field. Never overrides existing values;
never crashes on parse failure. Logs F3 backfill events.
- Mirrors KI-253 closer + KI-254 auto-mark safety-net pattern.

F4 — Premium pill vs panel label/scope clarity:
- Header chip kicker: "Est. premium" → "Est. range · across plans"
(HI: "अनुमानित premium · सभी plans में").
- Header chip tooltip: clarifies "across all eligible plans for your
profile. Tap to estimate for a specific plan."
- PremiumCalculatorPanel title: "Illustrative premium calculator" →
"Per-plan premium estimate" (HI: "एक-policy premium अनुमान").
- Panel subtitle rewritten to emphasize single-plan / specific-settings
scope.
- NEW italic 11px disambiguation Tip caption under subtitle:
"Tip: header chip = range across plans · panel = for these specific
settings".
- Panel result kicker: "INDICATIVE ANNUAL PREMIUM" → "FOR THESE
SETTINGS · ₹/YEAR".

F5 — Consolidate per-policy pricing source of truth:
- PolicyPremiumWidget (inside compare modal) was calling /api/premium/bulk
which used flat ₹500/lakh heuristic for non-curated policies — giving
₹6,000 for Multiplier vs panel's ₹18,445-24,954 for same profile.
- Switched to /api/premium/estimate (same endpoint the standalone panel
uses) → both surfaces now hit the same estimate() code path with the
same multiplier chain.
- main.py /api/premium/estimate gained tenure_years + deductible_inr
request fields + base_sample_used response field. Post-multiplies
point/low/high by BULK_TENURE_MULT × BULK_DEDUCTIBLE_DISCOUNT.
- api.ts PremiumEstimateRequest + PremiumEstimateResponse extended.
- PolicyPremiumWidget renders point_estimate_inr as headline + low–high
band as bullet. "Estimate" badge driven by base_sample_used === false.
- Now Multiplier @ 10L/metro/29yo/no-PED via the modal widget returns
₹11,500 (range ₹9,775-₹13,224) — same as the panel for that policy.
- Header chip /api/profile/predicted-premium-band continues using
bulk_estimate (aggregate across 26 curated policies) per spec.

Verification:
- python -m py_compile clean on 5 backend files
- npx tsc --noEmit clean
- npx next build succeeds

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend/brain_tools.py

@@ -50,7 +50,10 @@ Slot → consumer matrix:
     copay_pct             → pricing (copay discount 1.0× / 0.95× / 0.88× / 0.80×)
     family_medical_history → pricing (family-history loading) + retrieval boost
 
-Total: 15 slots. `gender` is tolerated by save_profile_field for forward
+  KI-275 (2026-05-15 — smoker / tobacco)
+    smoker                → pricing (smoker_loading 1.0× / 1.40×, +30-50%)
+
+Total: 16 slots. `gender` is tolerated by save_profile_field for forward
 compat but is NOT on the Profile dataclass today; it does not appear in
 SLOT_UNION because no consumer reads it.
 """
@@ -90,6 +93,8 @@ _ACCEPTED_FIELDS = {
     # D2 (2026-05-15) — coupled additions: co-pay tolerance + family medical history
     "copay_pct",
     "family_medical_history",
+    # KI-275 (2026-05-15) — smoker / tobacco use, +30-50% premium loading.
+    "smoker",
     "gender",  # tolerated; not persisted unless Profile gains the field
 }
 
@@ -138,6 +143,8 @@ SLOT_UNION: tuple[str, ...] = (
     # D2 additions (2026-05-15)
     "copay_pct",
     "family_medical_history",
+    # KI-275 (2026-05-15) — smoker / tobacco use, +30-50% premium loading.
+    "smoker",
 )
 
 # Invariant: every SLOT_UNION field must be accepted by save_profile_field
@@ -232,6 +239,8 @@ def save_profile_field(session, field: str, value: Any) -> dict:
             normalized = _coerce_copay_pct(value)
         elif fld == "family_medical_history":
             normalized = _coerce_family_medical_history(value)
+        elif fld == "smoker":
+            normalized = _coerce_smoker(value)
         elif fld == "name":
             normalized = (str(value).strip() if value is not None else None) or None
         elif fld == "gender":
@@ -744,6 +753,51 @@ def _coerce_bool(value: Any) -> Optional[bool]:
     return None
 
 
+def _coerce_smoker(value: Any) -> Optional[bool]:
+    """KI-275 (2026-05-15) — tri-state bool for smoker / tobacco use.
+
+    Accepts:
+      - True / "yes" / "true" / "smoker" / "smokes" / "tobacco" / 1  → True
+      - False / "no" / "false" / "non-smoker" / "doesn't smoke" / 0  → False
+      - None / "" / unclear                                          → None
+
+    Returning None lets the KI-091 null-overwrite guard in
+    save_profile_field refuse to clobber a previously-captured value.
+    """
+    if value is None:
+        return None
+    if isinstance(value, bool):
+        return value
+    if isinstance(value, (int, float)):
+        return bool(value)
+    s = str(value).strip().lower()
+    if not s:
+        return None
+    _YES = {
+        "yes", "y", "true", "1",
+        "smoker", "smokes", "smoke", "i smoke",
+        "tobacco", "tobacco user", "uses tobacco",
+        "i do", "yep", "yeah", "yup",
+    }
+    _NO = {
+        "no", "n", "false", "0",
+        "non-smoker", "nonsmoker", "non smoker",
+        "doesn't smoke", "does not smoke", "dont smoke", "don't smoke",
+        "i don't", "i do not", "nope", "never", "no tobacco",
+    }
+    if s in _YES:
+        return True
+    if s in _NO:
+        return False
+    # Substring fall-through for prose ("I'm a non-smoker", "I smoke daily").
+    if any(tok in s for tok in ("non-smoker", "nonsmoker", "non smoker", "don't smoke",
+                                 "doesn't smoke", "do not smoke", "no tobacco")):
+        return False
+    if any(tok in s for tok in ("smoker", "smokes", "tobacco")):
+        return True
+    return None
+
+
 def _coerce_desired_sum_insured(value: Any) -> Optional[int]:
     """Parse desired sum insured (cover amount) as integer rupees.
 

backend/main.py

@@ -915,6 +915,120 @@ async def chat(req: ChatRequest, request: Request):
             session_id, type(_rec_err).__name__, _rec_err,
         )
 
+    # F3 — confirmation auto-extract safety net.
+    # Symptom: Gemini emits a recap bullet list ("**Primary Goal:** first
+    # family policy ...") from conversation context but skips one or more
+    # save_profile_field calls. User says "yes this is correct"; the
+    # _profile_complete gate refuses retrieval and the bot embarrassingly
+    # re-asks. This block parses the bot's prior recap turn, maps slot
+    # labels -> _REQUIRED_FOR_READY field names, and backfills any slot
+    # that's STILL missing on the live profile. Best-effort; never blocks.
+    # Mirror of KI-253 closer regex + KI-254 auto-mark pattern.
+    try:
+        _CONFIRM_RE = re.compile(
+            r"\b(yes|correct|that'?s right|all correct|looks good)\b",
+            re.IGNORECASE,
+        )
+        _RECAP_BULLET_RE = re.compile(r"^\s*\*\s*\*\*[^:]+:\*\*", re.MULTILINE)
+        _RECAP_LINE_RE = re.compile(
+            r"^\s*\*\s*\*\*([^:]+):\*\*\s*(.+?)\s*$",
+            re.MULTILINE,
+        )
+        # Slot label aliases -> canonical _REQUIRED_FOR_READY field names.
+        # Keys are lowercased + whitespace-collapsed.
+        _SLOT_ALIASES = {
+            "name": "name",
+            "full name": "name",
+            "age": "age",
+            "dependents": "dependents",
+            "family": "dependents",
+            "family members": "dependents",
+            "location": "location_tier",
+            "city": "location_tier",
+            "location tier": "location_tier",
+            "city tier": "location_tier",
+            "income": "income_band",
+            "income band": "income_band",
+            "annual income": "income_band",
+            "primary goal": "primary_goal",
+            "goal": "primary_goal",
+            "objective": "primary_goal",
+            "health conditions": "health_conditions",
+            "health": "health_conditions",
+            "medical conditions": "health_conditions",
+            "pre-existing conditions": "health_conditions",
+            "medical history": "health_conditions",
+        }
+
+        if (
+            USE_SINGLE_BRAIN
+            and turn is not None
+            and _CONFIRM_RE.search(req.user_text or "")
+            and not getattr(turn, "profile_complete", False)
+        ):
+            # Locate the most recent bot/assistant message in chat_history.
+            _prior_bot_text = ""
+            for _msg in reversed(req.chat_history or []):
+                if not isinstance(_msg, dict):
+                    continue
+                _role = (_msg.get("role") or "").lower()
+                if _role in ("assistant", "bot", "model"):
+                    _prior_bot_text = _msg.get("content") or ""
+                    break
+
+            if _prior_bot_text and _RECAP_BULLET_RE.search(_prior_bot_text):
+                from backend.session_state import get_session as _get_session_f3
+                from backend import brain_tools as _brain_tools_f3
+
+                _f3_session = _get_session_f3(session_id)
+                _profile_f3 = _f3_session.profile
+
+                _parsed: dict[str, str] = {}
+                for _label, _value in _RECAP_LINE_RE.findall(_prior_bot_text):
+                    _key = " ".join(_label.strip().lower().split())
+                    _slot = _SLOT_ALIASES.get(_key)
+                    if not _slot:
+                        continue
+                    _val = (_value or "").strip().rstrip(".")
+                    if not _val:
+                        continue
+                    # Don't override slots that already have values.
+                    _existing = getattr(_profile_f3, _slot, None)
+                    if _existing not in (None, "", []):
+                        continue
+                    # First mapping wins (in case a label appears twice).
+                    _parsed.setdefault(_slot, _val)
+
+                _backfilled: list[str] = []
+                for _slot in _brain_tools_f3._REQUIRED_FOR_READY:
+                    if _slot not in _parsed:
+                        continue
+                    _existing = getattr(_profile_f3, _slot, None)
+                    if _existing not in (None, "", []):
+                        continue
+                    try:
+                        _r = _brain_tools_f3.save_profile_field(
+                            session=_f3_session,
+                            field=_slot,
+                            value=_parsed[_slot],
+                        )
+                        if isinstance(_r, dict) and _r.get("saved"):
+                            _backfilled.append(_slot)
+                    except Exception:  # noqa: BLE001 — best-effort
+                        continue
+
+                if _backfilled:
+                    logging.info(
+                        "F3 confirmation auto-extract (session=%s): backfilled %s",
+                        session_id, _backfilled,
+                    )
+    except Exception as _f3_err:  # noqa: BLE001
+        # Safety-net must never break the reply.
+        logging.warning(
+            "F3 confirmation auto-extract failed (session=%s): %s: %s",
+            session_id, type(_f3_err).__name__, _f3_err,
+        )
+
     audio_b64 = None
     audio_mime: Optional[str] = None
     if req.return_audio and turn.reply_text:
@@ -2970,6 +3084,14 @@ class PremiumEstimateRequest(BaseModel):
     )
     # Voluntary co-payment % — reduces premium ~7% per 10pp of co-pay
     copayment_pct: float = Field(0.0, ge=0, le=40)
+    # B2 widget parity (KI-bugfix, 2026-05-15) — optional slider overrides so
+    # PolicyPremiumWidget (compare modal) can use the same curated-anchored
+    # estimate() pipeline as PremiumCalculatorPanel instead of the divergent
+    # bulk_estimate() flat-base path. Applied as straight multipliers on top
+    # of the estimate() result using premium_calculator.BULK_TENURE_MULT /
+    # BULK_DEDUCTIBLE_DISCOUNT constants — leaves estimate() math untouched.
+    tenure_years: Optional[int] = Field(None, ge=1, le=3)
+    deductible_inr: Optional[int] = Field(None, ge=0, le=200_000)
 
 
 class PremiumEstimateResponse(BaseModel):
@@ -2984,12 +3106,25 @@ class PremiumEstimateResponse(BaseModel):
         "Illustrative range only — actual premium depends on underwriting + "
         "medical history + risk factors. Confirm with the insurer before purchase."
     )
+    # Echo back the effective tenure / deductible so the widget can render a
+    # consistent breakdown line without re-deriving them. Optional for legacy
+    # callers (PremiumCalculatorPanel ignores both).
+    tenure_years: Optional[int] = None
+    deductible_inr: Optional[int] = None
+    # True when the underlying estimate() anchored to a curated quote sample.
+    # PolicyPremiumWidget uses this (instead of bulk_estimate's `assumed` flag)
+    # to decide whether to show its "Estimate" badge.
+    base_sample_used: bool = False
 
 
 @app.post("/api/premium/estimate", response_model=PremiumEstimateResponse)
 async def premium_estimate(req: PremiumEstimateRequest):
     """Illustrative premium calculator — rules-based estimate from curated public data."""
-    from backend.premium_calculator import estimate as _estimate
+    from backend.premium_calculator import (
+        estimate as _estimate,
+        BULK_TENURE_MULT,
+        BULK_DEDUCTIBLE_DISCOUNT,
+    )
     e = _estimate(
         age=req.age,
         sum_insured_inr=req.sum_insured_inr,
@@ -3000,13 +3135,43 @@ async def premium_estimate(req: PremiumEstimateRequest):
         pre_existing_conditions=req.pre_existing_conditions,
         copayment_pct=req.copayment_pct,
     )
+
+    # Snap incoming tenure / deductible to the nearest supported bucket so the
+    # widget can pass raw slider values without precomputing.
+    point = e.point_estimate_inr
+    low = e.low_inr
+    high = e.high_inr
+    effective_tenure: Optional[int] = None
+    effective_ded: Optional[int] = None
+    if req.tenure_years is not None:
+        effective_tenure = req.tenure_years if req.tenure_years in BULK_TENURE_MULT else 1
+        tenure_mult = BULK_TENURE_MULT.get(effective_tenure, 1.0)
+        point = int(round(point * tenure_mult))
+        low = int(round(low * tenure_mult))
+        high = int(round(high * tenure_mult))
+    if req.deductible_inr is not None:
+        if req.deductible_inr in BULK_DEDUCTIBLE_DISCOUNT:
+            effective_ded = req.deductible_inr
+        else:
+            effective_ded = min(
+                BULK_DEDUCTIBLE_DISCOUNT.keys(),
+                key=lambda d: abs(d - req.deductible_inr),
+            )
+        ded_mult = BULK_DEDUCTIBLE_DISCOUNT.get(effective_ded, 1.0)
+        point = int(round(point * ded_mult))
+        low = int(round(low * ded_mult))
+        high = int(round(high * ded_mult))
+
     return PremiumEstimateResponse(
         policy_id=e.policy_id,
-        point_estimate_inr=e.point_estimate_inr,
-        low_inr=e.low_inr,
-        high_inr=e.high_inr,
+        point_estimate_inr=point,
+        low_inr=low,
+        high_inr=high,
         methodology=e.methodology,
         sources=e.sources or [],
+        tenure_years=effective_tenure,
+        deductible_inr=effective_ded,
+        base_sample_used=e.base_sample_used is not None,
     )
 
 

backend/needs_finder.py

@@ -48,6 +48,7 @@ class Profile:
     # retrieval (family-history rider boost keywords).
     copay_pct: Optional[int] = None  # 0-50, % of every claim user accepts
     family_medical_history: list[str] = field(default_factory=list)  # blood-family conditions
+    smoker: Optional[bool] = None  # KI-275 — tobacco use, +30-50% premium loading
     asked: list[str] = field(default_factory=list)  # question IDs / field names already asked
     free_form_session: bool = False  # True = user asks free questions, not driven by us
     # KI-063 (2026-05-15) — per-user policy interaction log so the bot

backend/premium_calculator.py

@@ -32,6 +32,7 @@ location / family_size that B2 already handles):
   copay_pct (D2)          → copay_discount 1.0× / 0.95× / 0.88× / 0.80×
   family_medical_history  → family_history_loading 1.0× / 1.03× / 1.05× / 1.10×
   (D2)                      (cancer/heart +5%, 2+ conditions +10%, other +3%)
+  smoker (KI-275)         → smoker_loading 1.0× / 1.40× (+30-50% premium load)
 
 Slots that are profile-only (no pricing effect): name, primary_goal,
 income_band, budget_band (matched against output, not folded into the
@@ -594,6 +595,10 @@ def bulk_estimate(
     # D2 — copay_pct + family_medical_history (same read pattern).
     copay_pct = profile.get("copay_pct")
     family_medical_history = profile.get("family_medical_history")
+    # KI-275 — smoker / tobacco use (+30-50% loading). Same read pattern as
+    # the D2 fields above; mirrors how the panel slider already passes
+    # `smoker` straight through to estimate() on the curated path.
+    smoker = bool(profile.get("smoker") or False)
     # desired_sum_insured_inr — when present, becomes the default SI for
     # any policy without an explicit overrides entry (per-policy override
     # still wins, since this is the DEFAULT).
@@ -615,6 +620,10 @@ def bulk_estimate(
     # callers see no change.
     copay_mult, copay_label = _copay_discount(copay_pct)
     fam_mult, fam_label = _family_history_loading(family_medical_history)
+    # KI-275 — smoker loading. 1.40× (+40%) standard tobacco loading.
+    # 1.0× when smoker is False / None so legacy callers see no change.
+    smoker_mult = 1.4 if smoker else 1.0
+    smoker_label = "smoker_loading" if smoker else "non_smoker"
 
     out: dict[str, BulkPolicyPremium] = {}
     for pid in policy_ids:
@@ -644,7 +653,7 @@ def bulk_estimate(
                     age=age,
                     sum_insured_inr=sum_insured_inr,
                     city_tier="metro" if loc_label == "metro" else ("tier1" if "1" in loc_label else "tier2"),
-                    smoker=bool(profile.get("smoker", False)),
+                    smoker=smoker,
                     family_size=max(0, family_size - 1),
                     policy_id=pid,
                     pre_existing_conditions=profile.get("pre_existing_conditions") or "none",
@@ -701,6 +710,7 @@ def bulk_estimate(
                 * parents_mult
                 * copay_mult
                 * fam_mult
+                * smoker_mult
                 * tenure_mult
                 * ded_mult
             )
@@ -739,6 +749,9 @@ def bulk_estimate(
         if fam_mult != 1.0:
             breakdown["family_history_loading_x"] = round(fam_mult, 3)
             breakdown["family_history_loading_reason"] = fam_label
+        if smoker_mult != 1.0:
+            breakdown["smoker_loading_x"] = round(smoker_mult, 3)
+            breakdown["smoker_loading_reason"] = smoker_label
         if desired_si and not ov.get("sum_insured_inr"):
             breakdown["desired_si_default_inr"] = int(desired_si)
 

backend/single_brain.py

@@ -186,7 +186,10 @@ After all 7 slots are saved AND the user has confirmed the recap (RULE 4 implici
    3. Any existing health cover from work or otherwise? (e.g., '5L through employer' or 'no')  [SKIP if existing_cover_inr already captured]
    4. Co-pay tolerance: Are you OK with a co-pay — sharing 10-30% of every claim — to lower the premium? Or do you want zero co-pay (insurer pays it all)?
    5. Family medical history: Any major conditions running in your blood family (parents/siblings) — cancer / diabetes / heart disease / hypertension?
-   6. Approximate age of the eldest parent you'd cover?  [ASK ONLY IF dependents mentions parents AND parents_age_max not yet captured]"
+   6. Approximate age of the eldest parent you'd cover?  [ASK ONLY IF dependents mentions parents AND parents_age_max not yet captured]
+   7. Smoking status: Do you smoke or use tobacco products? (yes / no)
+      Save: save_profile_field(field='smoker', value='yes' or 'no')
+      Smokers face 30-50% premium loading; capturing this gives an accurate band."
 
 When the user answers, call save_profile_field once per provided value:
   save_profile_field(field="desired_sum_insured_inr", value="1000000")  # ₹10L
@@ -195,6 +198,7 @@ When the user answers, call save_profile_field once per provided value:
   save_profile_field(field="copay_pct",              value="0" or "10" or "20" or "30")  # 0 = no co-pay (higher premium), 10-30 = typical tiers
   save_profile_field(field="family_medical_history", value="cancer, diabetes" or "none")  # blood family only (parents/siblings)
   save_profile_field(field="parents_age_max",        value="68")        # eldest parent's age, only if covering parents
+  save_profile_field(field="smoker",                 value="yes" or "no") # KI-275 — tobacco use, +30-50% premium loading
 
 Gender hint: if the user mentions gender, keep it for conversational context only — Profile has no `gender` slot. Do NOT call save_profile_field(field="gender", ...) — it returns `field_not_on_profile_dataclass` and wastes a tool-call iteration.
 
@@ -233,6 +237,30 @@ fields. Your flow on that turn:
 Explicit confirmation is only required when the user's reply is a literal
 "yes/no/that's right" with no new data. Bypass the WAIT in any other case.
 
+═══════════════════════════════════════════════════════════
+RECAP VERIFY — DO NOT RECAP SLOTS YOU HAVEN'T SAVED
+═══════════════════════════════════════════════════════════
+Before you emit a "Here's a quick recap of your profile:" summary, you MUST
+have called save_profile_field for EVERY slot you're about to list. The
+profile_complete=True return value from save_profile_field is your only
+proof a slot is captured. Do NOT recap a slot you only inferred from
+conversation context — if you "remember" the user mentioning something but
+didn't call save_profile_field on it, either call save_profile_field NOW
+or do NOT include it in the recap.
+
+The most common failure: user says "I want a first-time family policy" and
+you mention it in the recap but never actually called
+save_profile_field(field="primary_goal", value="first_buy"). When the user
+then says "yes this is correct", the profile_complete gate refuses retrieval
+and you have to embarrassingly ask again.
+
+Worked example. User says: "I have mild diabetes and a family history of diabetes."
+  -> You MUST call BOTH:
+       save_profile_field(field="health_conditions", value="diabetes")
+       save_profile_field(field="family_medical_history", value="diabetes")
+  -> Do NOT conflate them into a single save_profile_field with
+    "diabetes, family history of diabetes" — they are SEPARATE slots.
+
 ═══════════════════════════════════
 RULE 5 — Comparison view ("compare #1 and #3")
 ═══════════════════════════════════

frontend/src/app/page.tsx

@@ -1429,7 +1429,7 @@ export default function Page() {
                 className={`group relative overflow-hidden rounded-xl shadow-sm transition-all hover:shadow-md hover:brightness-110 cursor-pointer ${
                   showPremium ? "ring-2 ring-[var(--primary)]" : ""
                 }`}
-                title={uiLang === "hi" ? "Premium को sliders से refine करने के लिए tap करें" : "Tap to estimate / refine premium with sliders"}
+                title={uiLang === "hi" ? "आपकी profile के सभी eligible plans का अनुमानित premium range. किसी एक plan के लिए estimate देखने के लिए tap करें." : "Estimated premium range across all eligible plans for your profile. Tap to estimate for a specific plan."}
               >
                 <div className="absolute inset-0 bg-gradient-to-br from-amber-500 via-orange-500 to-amber-600" />
                 <div className="relative flex items-stretch text-white">
@@ -1438,7 +1438,7 @@ export default function Page() {
                   </div>
                   <div className="px-3 py-2 text-left">
                     <div className="text-[10px] uppercase tracking-wider opacity-85 leading-none">
-                      {uiLang === "hi" ? "अनुमानित premium" : "Est. premium"}
+                      {uiLang === "hi" ? "अनुमानित premium · सभी plans में" : "Est. range · across plans"}
                     </div>
                     <div className="text-xs font-bold leading-tight whitespace-nowrap">
                       {hasMeaningfulBand
@@ -1802,6 +1802,7 @@ export default function Page() {
               <PremiumCalculatorPanel
                 onClose={() => setShowPremium(false)}
                 initialProfile={profileCompleteness?.profile}
+                uiLang={uiLang}
               />
             )}
             {showProfile && (
@@ -2116,10 +2117,13 @@ function ProfileBuilderPanel({
 function PremiumCalculatorPanel({
   onClose,
   initialProfile,
+  uiLang = "en",
 }: {
   onClose: () => void;
   initialProfile?: UserProfile;
+  uiLang?: UILang;
 }) {
+  const isHi = uiLang === "hi";
   // KI (2026-05-15) — Fix B. The panel previously opened with static
   // defaults (Age 35 / SI 10L / Self only / None / metro) which felt
   // disconnected from the user's already-captured profile. We now seed
@@ -2246,9 +2250,16 @@ function PremiumCalculatorPanel({
       <div className="max-w-6xl mx-auto px-4 sm:px-6 py-5">
         <div className="flex items-baseline justify-between mb-3">
           <div>
-            <h2 className="text-sm font-semibold">Illustrative premium calculator</h2>
+            <h2 className="text-sm font-semibold">{isHi ? "एक-policy premium अनुमान" : "Per-plan premium estimate"}</h2>
             <p className="text-xs text-[var(--muted-foreground)]">
-              Indicative annual premium range from public quote data. Not a binding quote — actual depends on underwriting.
+              {isHi
+                ? "इन specific settings के साथ एक policy का अनुमानित premium देखने के लिए sliders adjust करें."
+                : "Adjust the sliders to see the estimated premium for a single plan with these specific settings."}
+            </p>
+            <p className="text-[11px] text-[var(--muted-foreground)] mt-1 italic">
+              {isHi
+                ? "Tip: header chip = सभी plans का range · panel = इन specific settings के लिए"
+                : "Tip: header chip = range across plans · panel = for these specific settings"}
             </p>
           </div>
           <button onClick={onClose} className="text-xs text-[var(--muted-foreground)] hover:underline">close</button>
@@ -2352,7 +2363,7 @@ function PremiumCalculatorPanel({
             {busy && <div className="text-xs text-[var(--muted-foreground)]">Estimating…</div>}
             {!busy && estimate && (
               <>
-                <div className="text-[10px] uppercase tracking-wide text-[var(--muted-foreground)] font-semibold">Indicative annual premium</div>
+                <div className="text-[10px] uppercase tracking-wide text-[var(--muted-foreground)] font-semibold">{isHi ? "इन settings के लिए · ₹/वर्ष" : "For these settings · ₹/year"}</div>
                 <div className="text-3xl font-bold mt-1">
                   {fmtINR(estimate.low_inr)} <span className="text-[var(--muted-foreground)] text-base font-normal">–</span> {fmtINR(estimate.high_inr)}
                 </div>

frontend/src/components/PolicyPremiumWidget.tsx

@@ -4,24 +4,35 @@
  * PolicyPremiumWidget — per-policy slider-driven premium calculator.
  *
  * Embedded inside PolicyCompareModal (B1). Fetches an initial estimate from
- * /api/premium/bulk using the user's profile defaults, then re-fetches
+ * /api/premium/estimate using the user's profile defaults, then re-fetches
  * (debounced 300ms) whenever the user moves the SI / tenure / deductible
- * sliders. When the backend marks the row `assumed: true` (no curated
- * actuarial data for this policy) the widget shows an "Estimate" badge so
+ * sliders. When the backend reports `base_sample_used: false` (no curated
+ * actuarial sample for this policy) the widget shows an "Estimate" badge so
  * the user understands the number is heuristic, not a quote.
+ *
+ * KI-bugfix (2026-05-15): switched from /api/premium/bulk → /api/premium/estimate
+ * so per-policy pricing shares the curated-anchored math used by the standalone
+ * PremiumCalculatorPanel. The bulk endpoint's flat ₹500/lakh fallback was
+ * producing wildly low numbers (~₹6K) versus the panel's curated number
+ * (~₹18-25K) for the same profile. Tenure + deductible are still honoured —
+ * the estimate endpoint now applies the bulk multipliers post-anchor.
  */
 
 import { useCallback, useEffect, useMemo, useRef, useState } from "react";
 
 import {
-  postPremiumBulk,
+  postPremiumEstimate,
   type PremiumBulkProfile,
-  type PremiumBulkRow,
+  type PreExistingCondition,
+  type PremiumEstimateResponse,
 } from "@/lib/api";
 
 export type PolicyPremiumWidgetProps = {
   policyId: string;
   policyName: string;
+  // Kept the bulk-profile shape because PolicyCompareModal still hands us this
+  // exact object — it doubles as the predicted-premium-band profile. We map
+  // its fields onto the estimate-endpoint contract internally.
   profile?: PremiumBulkProfile;
   initialSumInsured?: number;
   initialTenureYears?: 1 | 2 | 3;
@@ -64,33 +75,35 @@ function summariseProfile(profile?: PremiumBulkProfile): string | null {
   return parts.length ? parts.join(", ") : null;
 }
 
-const BREAKDOWN_LABELS: Record<string, string> = {
-  base_inr: "Base",
-  age_loading_x: "Age loading",
-  location_loading_x: "Location loading",
-  family_loading_x: "Family loading",
-  tenure_discount_x: "Tenure discount",
-  deductible_discount_x: "Deductible discount",
-};
+/**
+ * Map the free-text `location_tier` we ship in PremiumBulkProfile onto the
+ * strict {metro|tier1|tier2} enum the /api/premium/estimate endpoint expects.
+ * Same normalization rule used inside premium_calculator.bulk_estimate.
+ */
+function normaliseCityTier(tier: string | null | undefined): "metro" | "tier1" | "tier2" {
+  if (!tier) return "metro";
+  const t = String(tier).toLowerCase().replace(/[-_\s]/g, "");
+  if (t === "metro") return "metro";
+  if (t.includes("1")) return "tier1";
+  if (t.includes("2")) return "tier2";
+  // tier3 / unknown — fall back to tier2 (closest cheaper bucket the estimate
+  // endpoint accepts; matches the standalone panel's default).
+  return "tier2";
+}
 
-function renderBreakdownBullets(breakdown: Record<string, number | string>): string[] {
-  const bullets: string[] = [];
-  const base = breakdown.base_inr;
-  if (typeof base === "number") bullets.push(`Base: ₹${formatInr(base)}`);
-  for (const key of [
-    "age_loading_x",
-    "location_loading_x",
-    "family_loading_x",
-    "tenure_discount_x",
-    "deductible_discount_x",
-  ]) {
-    const v = breakdown[key];
-    if (typeof v === "number" && Math.abs(v - 1.0) > 0.001) {
-      const label = BREAKDOWN_LABELS[key];
-      bullets.push(`${label}: ${v.toFixed(2)}×`);
-    }
-  }
-  return bullets;
+/**
+ * Coerce the typed `pre_existing_conditions` profile slot onto the
+ * estimate-endpoint's PreExistingCondition union, defaulting to "none".
+ */
+function normalisePed(ped: string | null | undefined): PreExistingCondition {
+  if (!ped) return "none";
+  const allowed: PreExistingCondition[] = [
+    "none",
+    "diabetes_or_hypertension",
+    "heart_disease",
+    "multiple",
+  ];
+  return (allowed as string[]).includes(ped) ? (ped as PreExistingCondition) : "none";
 }
 
 export default function PolicyPremiumWidget({
@@ -103,9 +116,11 @@ export default function PolicyPremiumWidget({
   onCalculated,
 }: PolicyPremiumWidgetProps) {
   const [sumInsured, setSumInsured] = useState<number>(initialSumInsured);
-  const [tenureYears, setTenureYears] = useState<number>(initialTenureYears);
-  const [deductibleInr, setDeductibleInr] = useState<number>(initialDeductibleInr);
-  const [row, setRow] = useState<PremiumBulkRow | null>(null);
+  const [tenureYears, setTenureYears] = useState<1 | 2 | 3>(initialTenureYears);
+  const [deductibleInr, setDeductibleInr] = useState<0 | 25000 | 50000 | 100000>(
+    initialDeductibleInr,
+  );
+  const [resp, setResp] = useState<PremiumEstimateResponse | null>(null);
   const [loading, setLoading] = useState<boolean>(true);
   const [error, setError] = useState<string | null>(null);
 
@@ -123,25 +138,26 @@ export default function PolicyPremiumWidget({
       setLoading(true);
       setError(null);
       try {
-        const resp = await postPremiumBulk({
-          policy_ids: [policyId],
-          profile: profile ?? {},
-          overrides: {
-            [policyId]: {
-              sum_insured_inr: sumInsured,
-              tenure_years: tenureYears,
-              deductible_inr: deductibleInr,
-            },
-          },
+        // Defaults match the standalone PremiumCalculatorPanel so the two
+        // surfaces converge on the same number for the same profile.
+        const age = typeof profile?.age === "number" ? profile.age : 35;
+        const familySize =
+          typeof profile?.family_size === "number" ? profile.family_size : 1;
+        const r = await postPremiumEstimate({
+          age,
+          sum_insured_inr: sumInsured,
+          city_tier: normaliseCityTier(profile?.location_tier),
+          smoker: profile?.smoker === true,
+          family_size: familySize,
+          policy_id: policyId,
+          pre_existing_conditions: normalisePed(profile?.pre_existing_conditions),
+          copayment_pct: 0,
+          tenure_years: tenureYears,
+          deductible_inr: deductibleInr,
         });
         if (signal.aborted) return;
-        const r = resp.per_policy[policyId];
-        if (!r) {
-          setError("No estimate returned for this policy.");
-          return;
-        }
-        setRow(r);
-        onCalculatedRef.current?.(r.premium_inr_annual);
+        setResp(r);
+        onCalculatedRef.current?.(r.point_estimate_inr);
       } catch (e) {
         if (signal.aborted) return;
         setError(e instanceof Error ? e.message : String(e));
@@ -165,13 +181,34 @@ export default function PolicyPremiumWidget({
   }, [fetchPremium]);
 
   const profileSummary = summariseProfile(profile);
-  const bullets = row ? renderBreakdownBullets(row.breakdown) : [];
+  // `base_sample_used: false` means the curated illustrative_premiums.json
+  // had no anchor sample for this policy → estimate() fell back to the generic
+  // FALLBACK_BASE_INR path. Same semantics as the old bulk row.assumed flag.
+  const isHeuristic = resp ? resp.base_sample_used === false : false;
+
+  // Compact, profile-only breakdown — the estimate endpoint doesn't expose
+  // a multiplicative chain so we surface the active overrides + methodology
+  // instead. This keeps the widget transparent without faking precision.
+  const bullets: string[] = [];
+  if (resp) {
+    bullets.push(
+      `Range: ₹${formatInr(resp.low_inr)} – ₹${formatInr(resp.high_inr)}/year (±15% band)`,
+    );
+    if (resp.tenure_years && resp.tenure_years !== 1) {
+      bullets.push(`Multi-year discount applied (${resp.tenure_years}-year policy)`);
+    }
+    if (resp.deductible_inr && resp.deductible_inr > 0) {
+      bullets.push(
+        `Voluntary deductible discount applied (${formatDeductibleLabel(resp.deductible_inr)})`,
+      );
+    }
+  }
 
   return (
     <div className="policy-premium-widget" style={widgetStyle}>
       <header style={headerStyle}>
         <div style={{ fontWeight: 600, fontSize: 14 }}>{policyName}</div>
-        {row?.assumed && (
+        {isHeuristic && (
           <span style={badgeStyle} title="Heuristic — no exact actuarial data for this policy.">
             Estimate
           </span>
@@ -252,13 +289,13 @@ export default function PolicyPremiumWidget({
       <div style={resultBoxStyle} aria-live="polite">
         {error ? (
           <div style={{ color: "#b00020" }}>Failed: {error}</div>
-        ) : loading && !row ? (
+        ) : loading && !resp ? (
           <div style={{ color: "#666" }}>Calculating estimate…</div>
-        ) : row ? (
+        ) : resp ? (
           <>
             <div style={resultHeadlineStyle}>
               Estimated premium:&nbsp;
-              <strong>₹{formatInr(row.premium_inr_annual)}</strong>
+              <strong>₹{formatInr(resp.point_estimate_inr)}</strong>
               <span style={resultSuffixStyle}>/year</span>
               {loading && <span style={spinnerHintStyle}> updating…</span>}
             </div>
@@ -269,8 +306,8 @@ export default function PolicyPremiumWidget({
                 ))}
               </ul>
             )}
-            {row.notes && row.notes.length > 0 && (
-              <div style={noteStyle}>{row.notes.join(" ")}</div>
+            {resp.methodology && (
+              <div style={noteStyle}>{resp.methodology}</div>
             )}
           </>
         ) : null}

frontend/src/lib/api.ts

@@ -330,6 +330,12 @@ export type PremiumEstimateRequest = {
   policy_id?: string | null;
   pre_existing_conditions?: PreExistingCondition;
   copayment_pct?: number;
+  // B2 widget parity (KI-bugfix, 2026-05-15) — optional slider overrides so the
+  // PolicyPremiumWidget (compare modal) can share the curated-anchored
+  // estimate() pipeline with PremiumCalculatorPanel. Server snaps tenure to
+  // {1,2,3} and deductible to {0, 25k, 50k, 100k}.
+  tenure_years?: 1 | 2 | 3;
+  deductible_inr?: 0 | 25000 | 50000 | 100000;
 };
 
 export type PremiumEstimateResponse = {
@@ -341,6 +347,13 @@ export type PremiumEstimateResponse = {
   sources: string[];
   is_illustrative: boolean;
   disclaimer: string;
+  // Echoed back when caller passed tenure / deductible overrides.
+  tenure_years?: number | null;
+  deductible_inr?: number | null;
+  // True when the backend anchored the base to a curated quote sample (i.e.
+  // the policy is in illustrative_premiums.json). Drives the widget's
+  // "Estimate" badge — replaces bulk_estimate's `assumed` flag.
+  base_sample_used?: boolean;
 };
 
 export type ComparePolicyEntry = {