tests/test_rag.py

"""Smoke + behavior tests for the AgroSense RAG core.

Run:  python -m pytest        (or)   python tests/test_rag.py
These run fully offline against the hashing embedder + numpy vector store.
"""
from __future__ import annotations

import os
import sys
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
os.environ.setdefault("AGROSENSE_EMBEDDING_BACKEND", "hashing")  # deterministic, offline

from agrosense import RAGEngine
from agrosense.embeddings import HashingEmbedder


def test_hashing_embedder_is_normalized():
    emb = HashingEmbedder(dim=128)
    vecs = emb.encode(["sandy soil maize fertilizer", "paddy disease control"])
    norms = (vecs ** 2).sum(axis=1) ** 0.5
    assert vecs.shape == (2, 128)
    assert all(abs(n - 1.0) < 1e-5 for n in norms)


def _engine() -> RAGEngine:
    return RAGEngine()


def test_engine_indexes_kb():
    assert _engine().num_documents >= 5


def test_maize_query_retrieves_maize_and_cites():
    ans = _engine().answer(
        "My soil is sandy, rainfall 900 mm, I want to grow maize. "
        "What fertilizer and pest management should I follow?"
    )
    crops = {c.crop for c in ans.citations}
    assert "Maize" in crops, f"expected Maize in citations, got {crops}"
    assert ans.citations, "answer must include at least one citation"
    assert "Sources" in ans.text


def test_answer_is_grounded_no_empty():
    ans = _engine().answer("fertilizer schedule for mint aromatic crop")
    assert "Mint" in {c.crop for c in ans.citations}
    assert "50:75:50" in ans.text  # the exact basal dose from the KB


def test_irrelevant_query_returns_fallback_not_fabrication():
    ans = _engine().answer("what is the capital of France")
    # An off-domain question must NOT produce confident crop citations.
    assert ans.citations == [], f"expected no citations, got {ans.citations}"
    assert "could not find relevant guidance" in ans.text.lower()


def test_gate_rejects_label_word_only_query():
    # "fertilizer" is a field LABEL but appears in no field VALUE; a query that
    # only matches on the label must not return fabricated crop advice.
    ans = _engine().answer("fertilizer advice for my spaceship")
    assert ans.citations == [], f"expected no citations, got {ans.citations}"
    assert "could not find relevant guidance" in ans.text.lower()


def test_weather_advisories_are_grounded_in_numbers():
    # Build a forecast by hand (no network) and check advisories follow the data.
    from agrosense.weather import DailyWeather, WeatherForecast
    wf = WeatherForecast(
        location_name="Test", latitude=0.0, longitude=0.0,
        current_temp_c=30.0, current_precip_mm=0.0, current_humidity=50.0,
        daily=[
            DailyWeather("2026-06-01", tmax_c=39.0, tmin_c=24.0, precip_mm=0.0, precip_prob=10.0),
            DailyWeather("2026-06-02", tmax_c=35.0, tmin_c=23.0, precip_mm=20.0, precip_prob=80.0),
        ],
    )
    advisories = " ".join(wf.advisories()).lower()
    assert "rain likely" in advisories          # because day 2 has 80% / 20mm
    assert "high daytime temperatures" in advisories  # because day 1 tmax 39
    assert "Local weather" in wf.to_context()


def test_weather_offline_returns_none_not_error():
    # No location -> engine attaches no weather and never raises.
    ans = _engine().answer("paddy disease control", location=None)
    assert ans.weather is None
    assert "Sources" in ans.text


def test_engine_answer_accepts_location_param_gracefully():
    # A location is requested; offline this returns None weather but still answers
    # from the KB. (When online, weather is attached — covered by the live check.)
    ans = _engine().answer("fertilizer for maize", location="Nowhere-Place-XYZ-123")
    assert isinstance(ans.to_dict(), dict)
    assert ans.citations  # KB answer is unaffected by weather availability


def test_satellite_imagery_urls_built_offline():
    # URL construction needs no network and must produce valid GIBS WMS links.
    from datetime import date
    from agrosense.satellite import SatelliteClient
    client = SatelliteClient(buffer_deg=0.25)
    img = client.imagery(15.85, 74.50, today=date(2026, 5, 29))
    urls = img["imagery"]
    assert "MODIS_Terra_CorrectedReflectance_TrueColor" in urls["true_color"]
    assert "MODIS_Terra_NDVI_8Day" in urls["ndvi"]
    assert "BBOX=15.6,74.25,16.1,74.75" in urls["true_color"]  # lat-buf,lon-buf,...
    assert img["ndvi_date"] == "2026-05-11"   # today - 18 days
    assert "worldview.earthdata.nasa.gov" in urls["worldview"]


def test_agroclimate_notes_grounded_and_filter_fill_values():
    from agrosense.satellite import AgroClimate, _avg
    # NASA POWER fill value (-999) must be excluded from averages.
    assert _avg([20.0, -999.0, 22.0]) == 21.0
    ac = AgroClimate(start="2026-04-15", end="2026-05-15", days=30,
                     avg_solar_mj=22.0, avg_tmax_c=33.0, avg_tmin_c=22.0,
                     total_precip_mm=2.0)
    notes = " ".join(ac.notes()).lower()
    assert "ample solar radiation" in notes      # solar 22 >= 20
    assert "little rainfall" in notes             # 2 mm over 30 days


def test_satellite_offline_and_flag_off_attach_nothing():
    # include_satellite defaults False -> no satellite work, no network.
    ans = _engine().answer("maize fertilizer", location=None)
    assert ans.satellite is None


# --- field-level numeric NDVI (Earth Engine seam), mock-tested without creds ---

_EE_SAMPLE = {
    "type": "FeatureCollection",
    "features": [
        {"properties": {"date": "2026-05-20", "ndvi": 0.71, "cloud": 5}},
        {"properties": {"date": "2026-05-01", "ndvi": 0.52, "cloud": 12}},
        {"properties": {"date": "2026-05-10", "ndvi": None, "cloud": 90}},   # skipped
        {"properties": {"date": "2026-05-15", "ndvi": 0.66, "cloud": 8}},
    ],
}


def test_ndvi_parse_skips_nulls_and_sorts():
    from agrosense.ndvi import parse_ee_features
    obs = parse_ee_features(_EE_SAMPLE)
    assert [o.date for o in obs] == ["2026-05-01", "2026-05-15", "2026-05-20"]
    assert [o.ndvi for o in obs] == [0.52, 0.66, 0.71]


def test_ndvi_summarize_trend_status_and_notes():
    from agrosense.ndvi import NDVIResult, parse_ee_features, summarize, classify_ndvi
    assert classify_ndvi(0.7) == "healthy/dense"
    assert classify_ndvi(0.4) == "moderate"
    assert classify_ndvi(0.1) == "sparse/stressed"

    res = NDVIResult(location_name="Field", latitude=15.0, longitude=74.0,
                     buffer_m=100, observations=parse_ee_features(_EE_SAMPLE))
    summarize(res)
    assert res.latest == 0.71 and res.latest_date == "2026-05-20"
    assert res.status == "healthy/dense"
    assert res.trend == "rising"           # 0.52 early -> 0.66/0.71 late
    notes = " ".join(res.notes()).lower()
    assert "dense and healthy" in notes and "trending up" in notes


def test_ndvi_provider_disabled_without_credentials():
    # With no EE config (auto backend), the provider must not activate -> graceful None.
    import importlib
    from agrosense import config, ndvi
    assert config.NDVI_BACKEND == "auto" and not config.earthengine_configured()
    assert ndvi.get_ndvi_provider() is None
    assert ndvi.EarthEngineNDVIProvider().available() is False
    importlib.reload(ndvi)  # leave module clean


class _FakeNDVIProvider:
    """Stands in for Earth Engine so the seam can be tested without creds/network."""
    def fetch(self, lat, lon, location_name="", **kw):
        from agrosense.ndvi import NDVIResult, parse_ee_features, summarize
        res = NDVIResult(location_name=location_name, latitude=lat, longitude=lon,
                         buffer_m=100, observations=parse_ee_features(_EE_SAMPLE))
        return summarize(res)


def test_satellite_client_uses_injected_ndvi_provider():
    from agrosense.satellite import SatelliteClient
    client = SatelliteClient(ndvi_provider=_FakeNDVIProvider())
    nd = client.numeric_ndvi(15.85, 74.50, location_name="Test Field")
    assert nd is not None and nd.latest == 0.71 and nd.status == "healthy/dense"
    # And it must serialize + appear in the satellite context text.
    from agrosense.satellite import SatelliteReport
    rep = SatelliteReport(location_name="Test", latitude=15.85, longitude=74.50,
                          truecolor_date="2026-05-26", ndvi_date="2026-05-11",
                          imagery={"true_color": "x", "ndvi": "y"}, numeric_ndvi=nd)
    d = rep.to_dict()
    assert d["numeric_ndvi"]["latest"] == 0.71
    assert "Field NDVI" in rep.to_context()


def test_satellite_client_no_provider_returns_none():
    from agrosense.satellite import SatelliteClient
    client = SatelliteClient(ndvi_provider=None)
    assert client.numeric_ndvi(15.85, 74.50) is None


class _FakeMask:
    """Records the SCL classes a 'keep' mask excludes, via And-chaining."""
    def __init__(self, excluded):
        self.excluded = set(excluded)

    def And(self, other):   # noqa: N802 - mirrors the Earth Engine API name
        return _FakeMask(self.excluded | other.excluded)


class _FakeSCL:
    def neq(self, cls):
        return _FakeMask([cls])


class _FakeImage:
    def __init__(self):
        self.masked_with = None

    def select(self, band):
        assert band == "SCL"
        return _FakeSCL()

    def updateMask(self, mask):   # noqa: N802 - mirrors the Earth Engine API name
        self.masked_with = mask
        return self


def test_scl_cloud_mask_excludes_exactly_intended_classes():
    # Verifies the per-pixel cloud-mask composition WITHOUT Earth Engine: the
    # built mask must exclude exactly cloud/shadow/cirrus/snow SCL classes.
    from agrosense.ndvi import S2_CLOUD_SCL_CLASSES, apply_cloud_mask, build_scl_mask
    assert S2_CLOUD_SCL_CLASSES == (3, 8, 9, 10, 11)
    mask = build_scl_mask(_FakeSCL())
    assert mask.excluded == {3, 8, 9, 10, 11}
    img = _FakeImage()
    out = apply_cloud_mask(img)
    assert out is img and img.masked_with.excluded == {3, 8, 9, 10, 11}


# --- multilingual (translation), mock-tested without a translation backend ---

class _FakeTranslator:
    name = "fake"
    def translate(self, text, source, target):
        return f"[{target}] {text}"


class _RaisingTranslator:
    name = "boom"
    def translate(self, text, source, target):
        raise RuntimeError("no backend")


def test_supported_languages_cover_report_phase2():
    from agrosense.translation import SUPPORTED_LANGUAGES
    for code in ("en", "hi", "kn", "te", "mr", "bn", "ml"):
        assert code in SUPPORTED_LANGUAGES
    from agrosense.translation import IdentityTranslator
    assert IdentityTranslator().translate("hello", "en", "hi") == "hello"


def test_english_bypasses_translation():
    e = _engine()
    e._translator = _FakeTranslator()
    ans = e.answer("maize fertilizer", language="en")
    assert ans.language == "en" and ans.translation_backend is None
    assert not ans.text.startswith("[")          # no translation marker


def test_non_english_translates_query_in_and_answer_out():
    e = _engine()
    e._translator = _FakeTranslator()
    ans = e.answer("maize fertilizer", language="hi")
    assert ans.language == "hi" and ans.translation_backend == "fake"
    assert ans.text.startswith("[hi] ")          # answer translated to target
    assert ans.citations                          # retrieval still worked (query->en)
    assert "Maize" in {c.crop for c in ans.citations}


class _FakeAutoTranslator:
    """Mimics a backend that auto-detects (deep-translator). Records call sources."""
    name = "fake-auto"
    supports_auto = True
    def __init__(self):
        self.calls = []
    def translate(self, text, source, target):
        self.calls.append((source, target))
        if target == "en":         # query-in: auto-detected English -> unchanged
            return text
        return f"[{target}] {text}"


def test_query_in_uses_auto_detect_and_preserves_retrieval():
    # Regression: an already-English query under a non-English answer language must
    # NOT be mistranslated as if it were that language (which broke retrieval).
    e = _engine()
    ft = _FakeAutoTranslator()
    e._translator = ft
    ans = e.answer("maize fertilizer", language="hi")
    assert ("auto", "en") in ft.calls          # query translated with auto-detect
    assert "Maize" in {c.crop for c in ans.citations}   # retrieval preserved
    assert ans.text.startswith("[hi] ")        # answer still translated out


def test_translation_failure_falls_back_to_english():
    e = _engine()
    e._translator = _RaisingTranslator()
    ans = e.answer("maize fertilizer", language="hi")
    assert ans.translation_backend is None        # failed -> not applied
    assert "Sources" in ans.text                  # English grounded answer preserved


# --- market prices, mock-tested without a data.gov.in key ---

_PRICES_PAYLOAD = {
    "records": [
        {"market": "Belgaum", "commodity": "Tomato", "variety": "Local",
         "state": "Karnataka", "district": "Belagavi", "arrival_date": "2026-05-28",
         "min_price": "800", "max_price": "1400", "modal_price": "1100"},
        {"market": "Hubli", "commodity": "Tomato", "variety": "Hybrid",
         "state": "Karnataka", "district": "Dharwad", "arrival_date": "2026-05-28",
         "min_price": "900", "max_price": "1600", "modal_price": "1500"},
        {"market": "BadData", "commodity": "Tomato", "variety": "",
         "state": "Karnataka", "district": "", "arrival_date": "2026-05-28",
         "min_price": "NA", "max_price": None, "modal_price": "x"},  # bad -> None
    ]
}


def test_prices_parse_and_summary_grounded():
    from agrosense.prices import PriceReport, parse_records
    rows = parse_records(_PRICES_PAYLOAD)
    assert len(rows) == 3
    assert rows[2].modal_price is None and rows[2].min_price is None  # bad row cleaned
    rep = PriceReport(commodity="Tomato", state="Karnataka", records=rows)
    s = rep.summary()
    assert s["modal_min"] == 1100 and s["modal_max"] == 1500 and s["count"] == 3
    notes = " ".join(rep.notes())
    assert "Modal price" in notes and "vary notably" in notes  # spread > 20%


def test_price_client_disabled_without_key():
    from agrosense.prices import MarketPriceClient, get_price_client
    from agrosense import config
    assert MarketPriceClient(api_key=None).available() is False
    assert MarketPriceClient(api_key=None).fetch(commodity="Tomato") is None
    if not config.prices_configured():
        assert get_price_client() is None


class _FakePriceClient:
    def fetch(self, commodity=None, state=None, market=None, limit=None):
        from agrosense.prices import PriceReport, parse_records
        return PriceReport(commodity=commodity, state=state,
                           records=parse_records(_PRICES_PAYLOAD))


def test_engine_attaches_prices_for_retrieved_crop():
    e = _engine()
    e._prices = _FakePriceClient()
    ans = e.answer("price of tomato", include_prices=True)
    assert ans.prices is not None
    assert ans.prices["commodity"] == "Tomato"     # derived from top retrieved crop
    assert ans.prices["summary"]["modal_max"] == 1500


def test_engine_no_prices_when_flag_off():
    e = _engine()
    e._prices = _FakePriceClient()
    ans = e.answer("price of tomato", include_prices=False)
    assert ans.prices is None


# --- decision-fusion advisories (pure fuse(), no network) ---

def _mk_weather(daily_specs):
    from agrosense.weather import DailyWeather, WeatherForecast
    daily = [DailyWeather(date=f"2026-06-0{i+1}", tmax_c=t, tmin_c=tn,
                          precip_mm=mm, precip_prob=pp)
             for i, (t, tn, mm, pp) in enumerate(daily_specs)]
    return WeatherForecast(location_name="Test", latitude=15.0, longitude=74.0,
                           current_temp_c=28.0, current_precip_mm=0.0,
                           current_humidity=60.0, daily=daily)


def _mk_sat(total_precip=None, solar=22.0, ndvi_status=None, ndvi_trend=None):
    from agrosense.satellite import AgroClimate, SatelliteReport
    from agrosense.ndvi import NDVIResult
    ac = None
    if total_precip is not None:
        ac = AgroClimate(start="2026-05-01", end="2026-05-31", days=30,
                         avg_solar_mj=solar, avg_tmax_c=34.0, avg_tmin_c=22.0,
                         total_precip_mm=total_precip)
    nd = None
    if ndvi_status is not None:
        nd = NDVIResult(location_name="Test", latitude=15.0, longitude=74.0, buffer_m=100,
                        latest=0.7, mean=0.65, trend=ndvi_trend, status=ndvi_status)
    return SatelliteReport(location_name="Test", latitude=15.0, longitude=74.0,
                           truecolor_date="2026-05-26", ndvi_date="2026-05-11",
                           imagery={"ndvi": "u"}, agroclimate=ac, numeric_ndvi=nd)


def _titles(report):
    return {a.title for a in report.advisories}


def test_fusion_rain_plus_falling_ndvi_flags_drainage_high():
    from agrosense.fusion import fuse
    w = _mk_weather([(32, 23, 25, 90), (31, 23, 20, 85)])   # heavy rain both days
    sat = _mk_sat(total_precip=80, ndvi_status="moderate", ndvi_trend="falling")
    rep = fuse(w, sat, location_name="Test")
    titles = _titles(rep)
    assert "Drainage & disease check" in titles
    assert "Hold sprays & N top-dressing" in titles
    assert rep.advisories[0].urgency == "high"            # sorted high-first
    assert "Good field-work window" not in titles          # it's not dry


def test_fusion_dry_low_rain_sparse_ndvi_flags_irrigation_and_stress():
    from agrosense.fusion import fuse
    w = _mk_weather([(36, 24, 0.0, 5), (37, 24, 0.0, 0)])  # dry
    sat = _mk_sat(total_precip=3, ndvi_status="sparse/stressed", ndvi_trend="falling")
    rep = fuse(w, sat, location_name="Test")
    titles = _titles(rep)
    assert "Irrigate soon" in titles                       # dry + low recent rain
    assert "Likely crop stress over the field" in titles   # sparse NDVI + dry
    assert all(a.urgency in ("high", "medium", "low") for a in rep.advisories)


def test_fusion_healthy_rising_ndvi_is_low_urgency_only():
    from agrosense.fusion import fuse
    w = _mk_weather([(30, 22, 2.0, 30), (29, 22, 3.0, 40)])  # not rainy, not dry
    sat = _mk_sat(total_precip=50, ndvi_status="healthy/dense", ndvi_trend="rising")
    rep = fuse(w, sat, location_name="Test")
    titles = _titles(rep)
    assert "Canopy healthy" in titles
    assert all(a.urgency != "high" for a in rep.advisories)  # nothing urgent


def test_fusion_no_contradictory_maintain_with_urgent_irrigation():
    # A healthy, rising crop during a dry spell with low recent rain: must NOT show
    # "Canopy healthy / maintain" alongside a HIGH "Irrigate soon".
    from agrosense.fusion import fuse
    w = _mk_weather([(36, 24, 0.0, 5), (37, 24, 0.0, 0)])     # dry
    sat = _mk_sat(total_precip=3, ndvi_status="healthy/dense", ndvi_trend="rising")
    rep = fuse(w, sat, location_name="Test")
    titles = _titles(rep)
    assert "Irrigate soon" in titles                          # urgent need stands
    assert "Canopy healthy" not in titles                     # suppressed (would contradict)


def test_fusion_hint_when_ndvi_not_configured():
    from agrosense.fusion import fuse
    w = _mk_weather([(30, 22, 2.0, 30)])
    sat = _mk_sat(total_precip=50, ndvi_status=None)         # no numeric NDVI
    rep = fuse(w, sat, location_name="Test")
    assert "Enable field-level NDVI" in _titles(rep)


def test_fusion_handles_missing_signals():
    from agrosense.fusion import fuse
    rep = fuse(None, None, location_name="Nowhere", latitude=1.0, longitude=2.0)
    assert rep.advisories == [] and rep.location_name == "Nowhere"
    assert "No actionable signals" in rep.to_context()


def test_engine_fusion_requires_location():
    assert _engine().get_fusion_advisories() is None


# --- location environment profile (pure parsers, no network) ---

def test_compass_and_aqi_category():
    from agrosense.environment import compass, aqi_category
    assert compass(0) == "N" and compass(90) == "E" and compass(180) == "S"
    assert compass(270) == "W" and compass(266) == "W" and compass(None) is None
    assert aqi_category(30) == "Good" and aqi_category(75) == "Moderate"
    assert aqi_category(180) == "Unhealthy" and aqi_category(400) == "Hazardous"
    assert aqi_category(None) is None


def test_parse_air_quality_and_pollen_region_note():
    from agrosense.environment import parse_air_quality, parse_pollen
    payload = {"current": {"european_aqi": 18, "us_aqi": 34, "pm2_5": 7.5,
                           "pm10": 9.6, "ozone": 40, "grass_pollen": None}}
    aq = parse_air_quality(payload)
    assert aq.us_aqi == 34 and aq.pm2_5 == 7.5 and aq.category == "Good"
    # India: pollen is null -> not available, with an explanatory note.
    pol = parse_pollen(payload)
    assert pol.available is False and "Europe" in pol.note
    # Europe-style payload with real pollen values -> available.
    pol2 = parse_pollen({"current": {"grass_pollen": 12, "birch_pollen": 3}})
    assert pol2.available is True and pol2.values["grass"] == 12


def test_engine_environment_requires_location():
    assert _engine().get_environment() is None


class _FakeEnvClient:
    def profile(self, place=None, latitude=None, longitude=None):
        from agrosense.environment import EnvironmentProfile, Sunlight, Wind
        return EnvironmentProfile(
            location_name=place or "pt", latitude=15.0, longitude=74.0,
            elevation_m=769.0, population=490045, humidity_pct=92.0,
            sunlight=Sunlight(sunshine_hours=11.5, uv_index_max=8.95),
            wind=Wind(speed_kmh=12.6, direction_deg=266.0, direction_compass="W"))


def _find_adv(rep, prefix):
    return next((a for a in rep.advisories if a.title.startswith(prefix)), None)


def test_crop_profiles_and_stage_normalization():
    from agrosense.crop_profiles import get_crop_profile, normalize_stage
    assert get_crop_profile("Tomato").heat_stress_c == 32      # heat-sensitive
    assert get_crop_profile("Cotton").heat_stress_c == 40      # heat-tolerant
    assert get_crop_profile("unknown").name == "default"
    assert normalize_stage("Flowering") == "flowering"
    assert normalize_stage("bogus") is None and normalize_stage(None) is None


def test_fusion_per_crop_heat_threshold():
    from agrosense.fusion import fuse
    w = _mk_weather([(34, 24, 0.0, 0)])                        # 34°C, dry
    sat = _mk_sat(total_precip=50)
    # Tomato (threshold 32) -> heat advisory; Cotton (threshold 40) -> none.
    assert _find_adv(fuse(w, sat, crop="tomato"), "Heat stress") is not None
    assert _find_adv(fuse(w, sat, crop="cotton"), "Heat stress") is None


def test_fusion_stage_escalates_urgency():
    from agrosense.fusion import fuse
    w = _mk_weather([(39, 24, 0.0, 0)])                        # above maize heat threshold (38)
    sat = _mk_sat(total_precip=50)
    veg = _find_adv(fuse(w, sat, crop="maize", stage="vegetative"), "Heat stress")
    flo = _find_adv(fuse(w, sat, crop="maize", stage="flowering"), "Heat stress")
    assert veg.urgency == "medium"           # not a yield-critical stage
    assert flo.urgency == "high"             # flowering escalates


def test_fusion_maturity_rain_flags_harvest_quality():
    from agrosense.fusion import fuse
    w = _mk_weather([(30, 22, 25, 90)])                        # heavy rain
    rep = fuse(w, _mk_sat(total_precip=50), crop="wheat", stage="maturity")
    assert _find_adv(rep, "Harvest-quality risk") is not None
    assert rep.crop == "Wheat" and rep.stage == "maturity"


def test_fusion_stage_note_surfaced():
    from agrosense.fusion import fuse
    rep = fuse(_mk_weather([(30, 22, 2.0, 30)]), _mk_sat(total_precip=50),
               crop="paddy", stage="flowering")
    note = _find_adv(rep, "Stage watch")
    assert note is not None and note.urgency == "high"   # flowering is yield-critical


# --- planetary positions (pure ephemeris, no network) ---

def test_planetary_sun_declination_matches_season():
    from datetime import datetime, timezone
    from agrosense.planetary import compute_positions
    # June solstice: solar declination ~ +23.4°.
    jun = compute_positions(15.0, 74.0, when=datetime(2026, 6, 21, 12, 0, tzinfo=timezone.utc))
    sun = next(b for b in jun.bodies if b.name == "Sun")
    assert 22.5 < sun.dec_deg < 23.7, sun.dec_deg
    # March equinox: solar declination ~ 0°.
    mar = compute_positions(15.0, 74.0, when=datetime(2026, 3, 20, 12, 0, tzinfo=timezone.utc))
    sun2 = next(b for b in mar.bodies if b.name == "Sun")
    assert abs(sun2.dec_deg) < 1.5, sun2.dec_deg


def test_lunar_day_tithi_mapping_and_header():
    from datetime import datetime, timezone
    from agrosense.planetary import tithi_from_elongation, lunar_day
    # Boundaries: 0 deg -> Shukla Pratipada (1); 180 -> Krishna Pratipada (16).
    assert tithi_from_elongation(0) == {"tithi": 1, "paksha": "Shukla",
                                        "name": "Pratipada", "label": "Shukla Pratipada"}
    assert tithi_from_elongation(179.9)["name"] == "Purnima"          # full-moon tithi (15)
    assert tithi_from_elongation(180)["label"] == "Krishna Pratipada"  # tithi 16
    assert tithi_from_elongation(354)["name"] == "Amavasya" and tithi_from_elongation(354)["tithi"] == 30
    # Real instant: tithi in range, label is "<Paksha> <Name>".
    ld = lunar_day(datetime(2026, 5, 30, 12, 0, tzinfo=timezone.utc))
    assert 1 <= ld["tithi"] <= 30 and ld["paksha"] in ("Shukla", "Krishna")
    assert ld["label"].startswith(ld["paksha"])
    # The date header carries the lunar day for the UI top bar.
    from agrosense.calendars import datetime_header
    h = datetime_header(datetime(2026, 5, 30, 12, 0, tzinfo=timezone.utc))
    assert h["lunar_day"] and ("Shukla" in h["lunar_day"] or "Krishna" in h["lunar_day"])


def test_traditional_suitability_rules():
    from agrosense.traditional import astrological_suitability
    # Favourable nakshatra + waxing moon -> Favourable for sowing.
    good = astrological_suitability("sowing", {"nakshatra": "Rohini", "paksha": "Shukla",
                                               "karana": "Bava", "tithi_number": 5})
    assert good["verdict"] == "Favourable" and any("Rohini" in r for r in good["reasons"])
    # Vishti (Bhadra) karana -> postpone new work even if nakshatra is good.
    vishti = astrological_suitability("sowing", {"nakshatra": "Rohini", "paksha": "Shukla",
                                                 "karana": "Vishti", "tithi_number": 5})
    assert vishti["verdict"] == "Better to postpone"
    # Amavasya -> avoid sowing.
    amav = astrological_suitability("sowing", {"nakshatra": "Hasta", "paksha": "Krishna",
                                               "karana": "Naga", "tithi_number": 30})
    assert amav["verdict"] == "Better to postpone" and any("Amavasya" in r for r in amav["reasons"])
    # Waning moon favours pest control / harvest.
    pest = astrological_suitability("pest_control", {"nakshatra": "Chitra", "paksha": "Krishna",
                                                     "karana": "Gara", "tithi_number": 22})
    assert "favourable" in pest["verdict"].lower()


def test_traditional_practices_region_and_engine():
    from agrosense.traditional import practices_for
    # Karnataka 'Akkadi' (region-specific) ranks ahead of All-India for mixed cropping.
    ka = practices_for("mixed_cropping", region="Karnataka")
    assert ka and ka[0]["region"] == "Karnataka"
    sow = practices_for("sowing")
    assert any("Beejamrit" in p["title"] or "sowing" in p["title"].lower() or
               "Nakshatra" in p["title"] for p in sow)
    # Engine assembles panchang + suitability + practices + disclaimer.
    out = _engine().traditional_advice(activity="sowing")
    assert out["activity"] == "sowing"
    assert set(("vaara", "tithi", "nakshatra", "yoga", "karana")) <= set(out["panchang"])
    assert out["astrology"]["verdict"] and out["practices"]
    assert "complementary" in out["disclaimer"].lower()


def test_panchang_full():
    from datetime import datetime, timezone
    from agrosense.planetary import panchang, NAKSHATRAS, YOGAS
    from agrosense.calendars import datetime_header, VAARA
    p = panchang(datetime(2026, 5, 30, 12, 0, tzinfo=timezone.utc))
    assert p["nakshatra"] in NAKSHATRAS and 1 <= p["nakshatra_number"] <= 27
    assert p["yoga"] in YOGAS
    assert p["karana"] in (["Kimstughna", "Shakuni", "Chatushpada", "Naga",
                            "Bava", "Balava", "Kaulava", "Taitila", "Gara", "Vanija", "Vishti"])
    # Header panchang has all five limbs (vaara from the local weekday).
    h = datetime_header(datetime(2026, 5, 30, 12, 0, tzinfo=timezone.utc))
    pg = h["panchang"]
    assert pg["vaara"] in VAARA
    assert all(k in pg for k in ("vaara", "tithi", "paksha", "nakshatra", "yoga", "karana"))


def test_planetary_report_shape_and_ranges():
    from datetime import datetime, timezone
    from agrosense.planetary import compute_positions
    rep = compute_positions(15.85, 74.50, when=datetime(2026, 5, 29, 18, 30, tzinfo=timezone.utc),
                            location_name="Belagavi")
    names = [b.name for b in rep.bodies]
    assert names == ["Sun", "Moon", "Mercury", "Venus", "Mars", "Jupiter", "Saturn"]
    for b in rep.bodies:
        assert -90.0 <= b.altitude_deg <= 90.0
        assert 0.0 <= b.azimuth_deg < 360.0
        assert b.above_horizon == (b.altitude_deg > 0)
    assert 0.0 <= rep.moon_phase["illumination"] <= 1.0
    assert isinstance(rep.moon_phase["name"], str)
    assert "Planetary positions" in rep.to_context()


def test_engine_planetary_requires_location():
    assert _engine().get_planetary() is None


# --- evaluation harness (pure metrics + smoke run) ---

def test_eval_metric_functions():
    from agrosense.evaluation import (parse_fact_lines, retrieval_hit, retrieval_mrr,
                                      answer_relevance, faithfulness, _percentile)
    ans = "**Maize** [1]\n- Fertilizer: NPK 120:60:40\n- Pest management: scout weekly\n"
    facts = parse_fact_lines(ans)
    assert facts == ["NPK 120:60:40", "scout weekly"]
    assert retrieval_hit(["Maize", "Wheat"], "Maize") is True
    assert retrieval_hit(["Wheat"], "Maize") is False
    assert retrieval_mrr(["Wheat", "Maize"], "Maize") == 0.5
    assert retrieval_mrr(["Maize"], "Maize") == 1.0
    assert answer_relevance(ans, ["fertilizer", "pest"]) == 1.0
    assert answer_relevance(ans, ["disease"]) == 0.0
    # grounded: both fact values appear in grounding text -> 1.0
    grounding = "Fertilizer: NPK 120:60:40. Pest management: scout weekly every day."
    assert faithfulness(ans, grounding) == 1.0
    # hallucinated: a fact value not present -> < 1.0
    assert faithfulness(ans, "Fertilizer: NPK 120:60:40") == 0.5
    assert _percentile([1, 2, 3, 4], 50) in (2, 3)


def test_evaluate_smoke_over_engine():
    from agrosense.evaluation import EvalCase, evaluate
    cases = [
        EvalCase("fertilizer for maize", expected_crop="Maize", intents=["fertilizer"]),
        EvalCase("what is the capital of France", in_domain=False),
    ]
    report = evaluate(_engine(), cases)
    m = report["metrics"]
    assert m["n_in_domain"] == 1 and m["n_out_of_domain"] == 1
    assert m["context_relevance"] == 1.0          # maize query retrieves Maize
    assert m["faithfulness"] == 1.0               # extractive is grounded
    assert m["ood_accuracy"] == 1.0               # off-domain -> fallback
    assert "all_targets_met" in report and isinstance(report["all_targets_met"], bool)


# --- real ground water (CGWB/data.gov.in), pure logic + graceful gating ---

_GW_PAYLOAD = {"records": [
    {"station_name": "Well-A", "latitude": "15.80", "longitude": "74.50",
     "data_value": "8.5", "state": "Karnataka", "arrival_date": "2026-03-01"},
    {"station_name": "Well-B", "latitude": "16.50", "longitude": "75.20",
     "data_value": "12.0", "state": "Karnataka", "arrival_date": "2026-03-01"},
    {"station_name": "Bad", "latitude": "", "longitude": "", "data_value": "9"},  # dropped
]}


def test_groundwater_haversine_parse_and_nearest():
    from agrosense.groundwater import haversine_km, parse_stations, nearest_station
    # ~111 km per degree of latitude near the equator.
    assert 105 < haversine_km(15.0, 74.0, 16.0, 74.0) < 115
    stations = parse_stations(_GW_PAYLOAD)
    assert len(stations) == 2 and stations[0].depth_m == 8.5   # bad row dropped
    s, dist = nearest_station(stations, 15.85, 74.50)
    assert s.name == "Well-A" and dist < 10                    # closest well


def test_groundwater_client_disabled_without_config():
    from agrosense.groundwater import GroundwaterClient, get_groundwater_client
    from agrosense import config
    assert GroundwaterClient(resource_id=None, api_key=None).available() is False
    assert GroundwaterClient(resource_id=None, api_key=None).level(15.0, 74.0) is None
    if not config.groundwater_configured():
        assert get_groundwater_client() is None


class _FakeGWClient:
    def level(self, lat, lon):
        from agrosense.groundwater import GroundwaterReading
        return GroundwaterReading(depth_m=8.5, station_name="Well-A", distance_km=6.2,
                                  latitude=15.80, longitude=74.50, state="Karnataka",
                                  date="2026-03-01")


def test_environment_uses_real_groundwater_when_available():
    from agrosense.environment import EnvironmentClient
    client = EnvironmentClient(groundwater_client=_FakeGWClient())
    reading = client.groundwater_reading(15.85, 74.50)
    assert reading is not None and reading.depth_m == 8.5
    # And EnvironmentClient with no provider -> None (proxy path).
    assert EnvironmentClient(groundwater_client=None).groundwater_reading(15.0, 74.0) is None


# --- calendars (Gregorian + Indian National / Saka + IST) ---

def test_indian_national_calendar_anchors():
    from datetime import date
    from agrosense.calendars import indian_national_date
    # Chaitra 1 of Saka 1946 falls on 21 March 2024 (a leap year).
    assert indian_national_date(date(2024, 3, 21)) == (1946, 1, 1)
    # Chaitra 1 of Saka 1945 falls on 22 March 2023 (non-leap).
    assert indian_national_date(date(2023, 3, 22)) == (1945, 1, 1)
    # Day before Chaitra 1 belongs to the previous Saka year, last month Phalguna(12).
    y, m, d = indian_national_date(date(2023, 3, 21))
    assert y == 1944 and m == 12
    # 29 May 2026 -> 8 Jyaishtha 1948.
    assert indian_national_date(date(2026, 5, 29)) == (1948, 3, 8)


def test_ist_and_header():
    from datetime import datetime, timezone
    from agrosense.calendars import ist_now, datetime_header
    noon_utc = datetime(2026, 5, 29, 12, 0, tzinfo=timezone.utc)
    ist = ist_now(noon_utc)
    assert (ist.hour, ist.minute) == (17, 30)            # UTC+5:30
    h = datetime_header(noon_utc)
    assert "2026" in h["gregorian"] and "Saka" in h["indian_national"]
    assert h["ist_time"] == "17:30"


# --- Reuters news feed parsing (pure) ---

def test_news_parse_feed_and_clean_title():
    from agrosense.news import parse_feed
    xml = (
        '<rss><channel>'
        '<item><title>India monsoon weakest in 11 years - Reuters</title>'
        '<link>http://x/1</link><pubDate>Thu, 29 May 2026</pubDate></item>'
        '<item><title>Wheat crop outlook - Reuters</title>'
        '<link>http://x/2</link></item>'
        '</channel></rss>'
    )
    items = parse_feed(xml)
    assert len(items) == 2
    assert items[0].title == "India monsoon weakest in 11 years"   # " - Reuters" stripped
    assert items[0].link == "http://x/1"
    assert items[0].source == "Google News"                        # default (general)
    # Source is overridable (used for the scoped Reuters feed).
    assert parse_feed(xml, source="Reuters")[0].source == "Reuters"
    assert parse_feed(xml, limit=1) == items[:1]
    assert parse_feed("not xml at all") == []                       # graceful


def test_news_url_region_and_topic_building():
    from agrosense.news import build_news_url, LOCALES, TOPICS
    # General top stories for India.
    url, src = build_news_url(None, "IN")
    assert "news.google.com/rss?" in url and "ceid=IN%3Aen" in url and src == "Google News"
    # Topic search for the US locale.
    url2, _ = build_news_url(TOPICS["Business"], "US")
    assert "/rss/search?" in url2 and "q=business" in url2 and "ceid=US%3Aen" in url2
    # Reuters site filter -> source labelled Reuters.
    assert build_news_url("agriculture site:reuters.com", "IN")[1] == "Reuters"
    # Unknown region falls back to India.
    assert "ceid=IN%3Aen" in build_news_url(None, "ZZ")[0]
    assert "United States" in [v[3] for v in LOCALES.values()]


# --- commodity prices (Yahoo futures + Agmarknet), pure parse + graceful gating ---

def _yahoo_payload(price, prev, currency="USD"):
    return {"chart": {"result": [{"meta": {
        "regularMarketPrice": price, "chartPreviousClose": prev, "currency": currency}}]}}


def test_commodity_parse_and_currency():
    from agrosense.commodities import parse_yahoo_quote, currency_symbol
    assert currency_symbol("USD") == "$" and currency_symbol("USX") == "¢"
    q = parse_yahoo_quote(_yahoo_payload(4593.8, 4500.4), "Gold", "oz")
    assert q.price == 4593.8 and q.unit == "oz" and q.currency == "$"
    assert q.change_pct == round((4593.8 - 4500.4) / 4500.4 * 100, 2)
    # Coffee in US cents.
    qc = parse_yahoo_quote(_yahoo_payload(266.0, 274.0, "USX"), "Coffee", "lb")
    assert qc.currency == "¢" and qc.change_pct < 0
    # Bad / missing data -> None.
    assert parse_yahoo_quote({"chart": {"result": [{"meta": {}}]}}, "X", "oz") is None
    assert parse_yahoo_quote({"bad": 1}, "X", "oz") is None


class _FakeYahoo:
    def quote(self, symbol):
        return _yahoo_payload(100.0, 80.0)


class _FakeAgmarknet:
    def fetch(self, commodity=None, state=None, market=None, limit=None):
        from agrosense.prices import MarketPrice, PriceReport
        return PriceReport(commodity=commodity, state=None, records=[
            MarketPrice(market="m", commodity=commodity, variety="", state="",
                        district="", arrival_date="", min_price=100, max_price=200,
                        modal_price=150)])


def test_commodities_client_lists_all_six_with_graceful_gating():
    from agrosense.commodities import CommoditiesClient
    # No Agmarknet key -> Arecanut/Coconut unavailable, globals priced.
    items = CommoditiesClient(yahoo=_FakeYahoo(), prices=None).quotes()
    names = [c.name for c in items]
    assert names == ["Gold", "Silver", "Crude Oil", "Coffee", "Arecanut", "Coconut"]
    assert all(c.price == 100.0 and c.change_pct == 25.0 for c in items[:4])
    assert items[4].price is None and items[5].price is None      # n/a without key
    # With an Agmarknet provider, the mandi commodities get a modal price.
    items2 = CommoditiesClient(yahoo=_FakeYahoo(), prices=_FakeAgmarknet()).quotes()
    assert items2[4].name == "Arecanut" and items2[4].price == 150.0
    assert items2[4].currency == "₹" and items2[4].unit == "quintal"


# --- hazards: EONET events + FIRMS fires (pure parse) ---

_EONET_PAYLOAD = {"events": [
    {"title": "Flood near field", "categories": [{"title": "Floods"}],
     "geometry": [{"date": "2026-05-01", "type": "Point", "coordinates": [74.6, 15.9]}],
     "link": "http://x/a"},
    {"title": "Distant storm", "categories": [{"title": "Severe Storms"}],
     "geometry": [{"date": "2026-05-02", "type": "Point", "coordinates": [80.0, 13.0]}]},
    {"title": "Polygon fire zone", "categories": [{"title": "Wildfires"}],
     "geometry": [{"date": "2026-05-03", "type": "Polygon",
                   "coordinates": [[[74.55, 15.88], [74.6, 15.9], [74.5, 15.8]]]}]},
]}


def test_eonet_parse_filters_radius_and_sorts():
    from agrosense.hazards import parse_eonet_events
    evs = parse_eonet_events(_EONET_PAYLOAD, lat=15.85, lon=74.50, radius_km=500)
    titles = [e.title for e in evs]
    assert "Distant storm" not in titles            # ~700 km away, filtered out
    assert "Flood near field" in titles and "Polygon fire zone" in titles
    assert evs[0].distance_km <= evs[1].distance_km  # sorted by distance ascending
    flood = next(e for e in evs if e.title == "Flood near field")
    assert flood.latitude == 15.9 and flood.longitude == 74.6   # [lon,lat] handled


def test_firms_parse_csv():
    from agrosense.hazards import parse_firms_csv, FirmsClient
    csv = ("latitude,longitude,bright_ti4,acq_date,acq_time,confidence,frp,daynight\n"
           "15.90,74.60,330.1,2026-05-28,1200,n,5.2,D\n"
           "16.50,75.00,310.0,2026-05-28,1206,h,3.1,D\n"
           "bad,row,only\n")
    fires = parse_firms_csv(csv, lat=15.85, lon=74.50)
    assert len(fires) == 2                           # bad row skipped
    assert fires[0].distance_km <= fires[1].distance_km
    assert fires[0].frp == 5.2 and fires[0].confidence == "n"
    # No key -> client unavailable, fires() returns None.
    assert FirmsClient(map_key=None).available() is False
    assert FirmsClient(map_key=None).fires(15.0, 74.0) is None


def test_engine_hazards_requires_location():
    assert _engine().get_hazards() is None


# --- vision: plant disease + identification (pure logic + heuristic) ---

def test_vision_softmax_topk_and_health():
    from agrosense.vision import softmax, top_k, health_from_colors
    p = softmax([2.0, 1.0, 0.0])
    assert abs(sum(p) - 1.0) < 1e-6 and p[0] > p[1] > p[2]
    tk = top_k([0.1, 0.7, 0.2], ["a", "b", "c"], k=2)
    assert tk[0]["label"] == "b" and len(tk) == 2
    assert "Healthy" in health_from_colors({"green": 0.8, "yellow": 0.05, "brown": 0.05})[0]
    assert "stress" in health_from_colors({"green": 0.3, "yellow": 0.3, "brown": 0.2})[0].lower()
    assert "Inconclusive" in health_from_colors({"green": 0.3, "other": 0.7})[0]


def _png_bytes(color):
    import io
    from PIL import Image
    buf = io.BytesIO()
    Image.new("RGB", (16, 16), color).save(buf, format="PNG")
    return buf.getvalue()


def test_vision_heuristic_disease_and_plant_fallback():
    from agrosense.vision import PlantVision
    pv = PlantVision(disease_model=None, plant_model=None)   # no trained model
    green = pv.predict_disease(_png_bytes((20, 180, 40)))
    assert green.backend == "heuristic" and "Healthy" in green.label
    brown = pv.predict_disease(_png_bytes((150, 90, 40)))
    assert "stress" in brown.label.lower() or "Inconclusive" in brown.label
    plant = pv.predict_plant(_png_bytes((20, 180, 40)))
    assert plant.label == "Unknown" and "trained model" in plant.note
    pest = pv.predict_pest(_png_bytes((20, 180, 40)))
    assert pest.task == "pest" and pest.label == "Unknown"
    assert "IP102" in pest.note and pest.backend == "heuristic"


class _StubPestModel:
    def predict(self, image_bytes):
        from agrosense.vision import Prediction
        return Prediction("pest", "Brown planthopper", 0.91,
                          top_k=[{"label": "Brown planthopper", "prob": 0.91}],
                          backend="keras:pest.keras", note="Trained CNN inference.")


def test_vision_pest_uses_trained_model_when_present():
    from agrosense.vision import PlantVision
    pv = PlantVision(pest_model=_StubPestModel())
    p = pv.predict_pest(_png_bytes((90, 70, 40)))
    assert p.label == "Brown planthopper" and p.backend.startswith("keras")


def test_vision_factory_without_models():
    from agrosense.vision import get_plant_vision, PlantVision
    pv = get_plant_vision()
    assert isinstance(pv, PlantVision)
    # No model env configured -> all three backends None (heuristic path).
    assert pv._disease is None and pv._plant is None and pv._pest is None


# --- plant telemedicine (pure logic + engine consult) ---

def test_telemedicine_intents_severity_health():
    from agrosense.telemedicine import (symptom_intents, assess_severity, derive_health)
    assert "disease" in symptom_intents("yellow spots and blight on leaves")
    assert "pest" in symptom_intents("holes chewed by a caterpillar")
    assert "nutrition" in symptom_intents("pale stunted plants")
    assert symptom_intents("") == []
    assert assess_severity("spots spreading rapidly everywhere", 0.3) == "high"   # 2+ words
    assert assess_severity("a few spots", 0.4) == "low"
    assert assess_severity(None, 0.9) == "high"                                    # high conf
    assert derive_health("Healthy foliage", 0.8, False) == "Likely healthy"
    assert derive_health("Suspected disease", 0.4, True) == "Needs attention"


def test_telemedicine_build_prescription_is_grounded_and_cited():
    from agrosense.telemedicine import build_prescription
    meta = {"recommended_fertilizer": "NPK 120:80:60", "disease_prevention": "Spray Mancozeb 0.25%",
            "pest_management": "Yellow sticky traps", "source": "ICAR Bulletin"}
    items = build_prescription(meta, ["disease", "pest"])
    cats = [i.category for i in items]
    assert "Treatment (disease)" in cats and "Treatment (pest)" in cats
    assert "Cultural / IPM" in cats and "Monitoring" in cats
    # Dosage text comes verbatim from the KB field, and every item is cited.
    assert any("Mancozeb 0.25%" in i.instruction for i in items)
    assert all(i.source == "ICAR Bulletin" for i in items)
    # No intents -> defaults to disease + nutrition.
    assert "Nutrition / fertilizer" in [i.category for i in build_prescription(meta, [])]


def test_ipm_label_parse_and_lookup():
    from agrosense.ipm import parse_vision_label, lookup_ipm
    # PlantVillage-style class label -> (crop, condition).
    assert parse_vision_label("Tomato___Late_blight") == ("Tomato", "Late blight")
    assert parse_vision_label("aphid") == (None, "aphid")
    # Class label drives the IPM entry.
    e = lookup_ipm("Tomato___Late_blight")
    assert e and e["condition"] == "Late blight" and "Mancozeb" in e["treatment"]
    # Symptom text naming a pest also maps.
    assert lookup_ipm("lots of fall armyworm in the whorl")["condition"] == "Fall armyworm"
    # Most specific (longest) key wins; nonsense -> None.
    assert lookup_ipm("powdery mildew on leaves")["condition"] == "Powdery mildew"
    assert lookup_ipm("the weather is nice today") is None


def test_consultation_is_diagnosis_driven_with_ipm():
    # Symptom names a specific disease -> targeted IPM prescription + updated diagnosis.
    ans = _engine().plant_consultation(crop="Tomato",
                                       symptoms="late blight lesions spreading on leaves")
    assert "Late blight" in ans["diagnosis"]
    cats = [p["category"] for p in ans["prescription"]]
    assert any(c.startswith("Treatment (diagnosed: Late blight)") for c in cats)
    # The targeted IPM treatment text + its source are present and cited.
    rx = " ".join(p["instruction"] for p in ans["prescription"])
    assert "Mancozeb" in rx or "Chlorothalonil" in rx
    assert any("IPM Guide" in c for c in ans["citations"])


def test_engine_plant_consultation_grounded():
    ans = _engine().plant_consultation(crop="Tomato",
                                       symptoms="yellow spots spreading on the lower leaves")
    assert ans["crop"] == "Tomato" and ans["diagnosis_basis"] == "symptoms"
    assert ans["health_status"] == "Needs attention"
    assert ans["severity"] in ("moderate", "high")
    assert len(ans["prescription"]) >= 3 and ans["citations"]
    assert "disclaimer" in ans and "extension officer" in ans["disclaimer"]
    assert ans["weather_note"] is None              # no location given


# --- farmers' digital clubs ---

def test_club_validate_and_store():
    import tempfile
    from pathlib import Path
    from agrosense.clubs import validate_club, ClubStore
    assert validate_club("X", "location", "Karnataka") is None
    assert "name" in validate_club("", "location", "KA")
    assert "location" in validate_club("X", "bogus", "KA").lower()
    assert "commodity" in validate_club("X", "commodity", "").lower()
    with tempfile.TemporaryDirectory() as d:
        store = ClubStore(path=Path(d) / "clubs.json", video_base="https://v.example")
        # seeded with defaults on first use
        assert len(store.all()) >= 4
        assert any(c.type == "location" for c in store.all())
        # create + room url + creator auto-joined
        c = store.create("Mango Growers", "commodity", "Mango", "desc", creator="Asha")
        assert c.id == "mango-growers" and "Asha" in c.members
        assert store.public(c)["room_url"] == "https://v.example/AgroSense-Club-mango-growers"
        # join + post
        store.join(c.id, "Ravi")
        store.add_post(c.id, "Ravi", "First mango harvest done!", link="http://x")
        got = store.get(c.id)
        assert "Ravi" in got.members and got.posts[-1]["text"].startswith("First mango")
        # filter by type/key/search
        assert all(x.type == "commodity" for x in store.filter(ctype="commodity"))
        assert any(x.key == "Mango" for x in store.filter(key="mango"))
        assert any("Mango" in x.name for x in store.filter(search="mango"))


def test_engine_clubs_flow():
    from agrosense.config import DATA_DIR
    from pathlib import Path
    p = Path(DATA_DIR) / "clubs.json"
    original = p.read_text(encoding="utf-8") if p.exists() else None
    try:
        e = _engine()
        loc = e.list_clubs(ctype="location")
        assert loc and all(c["type"] == "location" for c in loc)
        club = e.create_club("Test Banana Club", "commodity", "Banana", "d", creator="Meena")
        assert club["member_count"] == 1 and "Banana" in club["key"]
        joined = e.join_club(club["id"], "Karthik")
        assert joined["member_count"] == 2
        posted = e.post_to_club(club["id"], "Karthik", "Panama wilt spreading — advice?")
        assert posted["posts"][-1]["author"] == "Karthik"
        assert e.get_club(club["id"])["room_url"].endswith("AgroSense-Club-" + club["id"])
    finally:
        if original is None:
            p.unlink(missing_ok=True)
        else:
            p.write_text(original, encoding="utf-8")


def test_trading_validate_and_store():
    import tempfile
    from pathlib import Path
    from agrosense.trading import validate_listing, TradingStore
    assert validate_listing("sell", "Tomato", 500, 18) is None
    assert "Type" in validate_listing("lease", "Tomato", 1, 1)
    assert "Commodity" in validate_listing("sell", "", 1, 1)
    assert "Quantity" in validate_listing("sell", "Tomato", 0, 1)
    assert "Quantity" in validate_listing("sell", "Tomato", "abc", 1)
    assert "Price" in validate_listing("sell", "Tomato", 1, -5)
    with tempfile.TemporaryDirectory() as d:
        store = TradingStore(path=Path(d) / "m.json", video_base="https://v.example")
        assert len(store.all()) >= 3                       # seeded with demos
        assert any(l.type == "buy" for l in store.all()) and any(l.type == "sell" for l in store.all())
        l = store.create({"type": "sell", "commodity": "Mango", "quantity": "20",
                          "unit": "quintal", "price": "4500", "location": "Ratnagiri",
                          "state": "Maharashtra", "seller": "Asha"})
        assert l.id == "mango-sell" and l.quantity == 20.0 and l.price == 4500.0
        assert store.public(l)["room_url"] == "https://v.example/AgroSense-Deal-mango-sell"
        # only open listings by default; closed hidden unless include_closed
        store.set_status(l.id, "closed")
        assert all(x.id != "mango-sell" for x in store.filter())
        assert any(x.id == "mango-sell" for x in store.filter(include_closed=True))
        # inquiry blocked on a closed listing
        try:
            store.add_inquiry(l.id, "Buyer", "999", "interested")
            assert False, "expected closed-listing error"
        except ValueError:
            pass
        # reopen + inquire + filters
        store.set_status(l.id, "open")
        got = store.add_inquiry(l.id, "Buyer", "99999", "Can do 4400?", quantity="10")
        assert got.inquiries[-1]["contact"] == "99999" and got.inquiries[-1]["quantity"] == 10.0
        assert all(x.type == "sell" for x in store.filter(ltype="sell"))
        assert any(x.commodity == "Mango" for x in store.filter(commodity="mang"))
        assert any(x.id == "mango-sell" for x in store.filter(state="maharashtra"))
        # empty inquiry rejected
        try:
            store.add_inquiry(l.id, "X", "", "")
            assert False, "expected empty-inquiry error"
        except ValueError:
            pass


def test_engine_trading_flow():
    from agrosense.config import DATA_DIR
    from pathlib import Path
    p = Path(DATA_DIR) / "market_listings.json"
    original = p.read_text(encoding="utf-8") if p.exists() else None
    try:
        e = _engine()
        sells = e.list_listings(ltype="sell")
        assert sells and all(l["type"] == "sell" for l in sells)
        created = e.create_listing({"type": "sell", "commodity": "Banana", "quantity": 12,
                                    "unit": "dozen", "price": 60, "location": "Theni",
                                    "state": "Tamil Nadu", "seller": "Murugan"})
        assert created["commodity"] == "Banana" and created["status"] == "open"
        assert created["room_url"].endswith("AgroSense-Deal-" + created["id"])
        inq = e.inquire_listing(created["id"], "Wholesaler", "98765", "Bulk order?", quantity=100)
        assert inq["inquiry_count"] == 1
        closed = e.close_listing(created["id"])
        assert closed["status"] == "closed"
        # closed listing drops out of the default browse
        assert all(l["id"] != created["id"] for l in e.list_listings())
        # bad input -> ValueError
        try:
            e.create_listing({"type": "sell", "commodity": "", "quantity": 1, "price": 1})
            assert False, "expected ValueError"
        except ValueError:
            pass
    finally:
        if original is None:
            p.unlink(missing_ok=True)
        else:
            p.write_text(original, encoding="utf-8")


# --- live agri-doctor consultation ---

def _experts():
    from agrosense.consultation import ExpertDirectory
    return ExpertDirectory.load().experts


def test_consultation_expert_matching_and_room_url():
    from agrosense.consultation import match_expert, room_url
    experts = _experts()
    # Disease -> pathology expert; pest -> entomology; honor language when possible.
    assert "pathology" in match_expert(experts, "disease").tags
    assert "entomology" in match_expert(experts, "pest").tags
    kn = match_expert(experts, "disease", "Kannada")
    assert kn is not None and "Kannada" in kn.languages
    # Unknown area falls back to an available expert (general/any), never None here.
    assert match_expert(experts, "zzz") is not None
    assert room_url("https://meet.jit.si/", "C0007") == "https://meet.jit.si/AgroSense-Consult-C0007"


def test_consultation_service_lifecycle():
    from agrosense.consultation import ConsultationService, ExpertDirectory
    svc = ConsultationService(ExpertDirectory.load(), video_base="https://v.example")
    req = svc.create("Asha", "Tomato", "AI: late blight", channel="video", area="disease")
    assert req.id == "C0001" and req.status == "assigned" and req.expert is not None
    assert req.room_url == "https://v.example/AgroSense-Consult-C0001"
    assert req.messages and req.messages[0].sender == "system"
    # Chat append + retrieval.
    svc.add_message(req.id, "farmer", "Leaves have brown lesions")
    got = svc.get(req.id)
    assert got.messages[-1].text == "Leaves have brown lesions"
    assert svc.get("nope") is None


def test_engine_request_live_consult_attaches_ai_context():
    e = _engine()
    out = e.request_live_consult(farmer_name="Ravi", crop="Maize",
                                 symptoms="fall armyworm larvae in the whorl",
                                 channel="video", language="Hindi")
    assert out["status"] == "assigned" and out["expert"] is not None
    assert out["room_url"].endswith("AgroSense-Consult-" + out["id"])
    assert "ai_consult" in out and "Fall armyworm" in out["ai_consult"]["diagnosis"]
    # Pest symptom should route to the entomology expert.
    assert "entomology" in out["expert"]["tags"]
    # Chat round-trip through the engine.
    updated = e.add_consult_message(out["id"], "farmer", "Spreading fast")
    assert updated["messages"][-1]["text"] == "Spreading fast"


# --- expert notifications ---

_REQ = {"id": "C0001", "farmer_name": "Asha", "crop": "Tomato",
        "summary": "AI: Late blight", "channel": "video",
        "room_url": "https://meet.jit.si/AgroSense-Consult-C0001",
        "expert": {"name": "Dr. Anjali Rao", "specialization": "Plant Pathology"}}


def test_notification_message_and_log():
    from agrosense.notifications import build_consult_notification, LogNotifier
    subject, body = build_consult_notification(_REQ)
    assert "C0001" in subject and "Tomato" in subject
    assert "Dr. Anjali Rao" in body and "meet.jit.si" in body
    log = LogNotifier()
    n = log.send(subject, body, {})
    assert n.ok and n.channel == "log" and log.sent == [n]


def test_webhook_notifier_gating_and_post():
    from agrosense.notifications import WebhookNotifier
    assert WebhookNotifier(url=None).available() is False
    assert WebhookNotifier(url=None).send("s", "b", {}) is None
    captured = {}

    def fake_post(url, json_payload, timeout):
        captured["url"] = url
        captured["payload"] = json_payload

    wh = WebhookNotifier(url="https://hook.example/x", poster=fake_post)
    n = wh.send("subj", "body", {"consultation_id": "C0001"})
    assert n.ok and captured["url"] == "https://hook.example/x"
    assert captured["payload"]["consultation_id"] == "C0001" and captured["payload"]["subject"] == "subj"

    def boom(url, json_payload, timeout):
        raise RuntimeError("down")

    assert WebhookNotifier(url="https://x", poster=boom).send("s", "b", {}).ok is False


def test_notifier_default_is_log_only_and_history():
    from agrosense.notifications import Notifier
    nf = Notifier()        # no webhook/email env -> log backend only
    out = nf.notify_consult(_REQ)
    assert [n.channel for n in out] == ["log"]
    assert len(nf.history()) == 1


def test_notifier_with_injected_webhook():
    from agrosense.notifications import Notifier, LogNotifier, WebhookNotifier
    posted = []
    wh = WebhookNotifier(url="https://hook/x", poster=lambda u, p, t: posted.append(p))
    nf = Notifier(backends=[LogNotifier(), wh])
    out = nf.notify_consult(_REQ)
    assert {n.channel for n in out} == {"log", "webhook"} and posted


def test_engine_consult_records_notification():
    e = _engine()
    out = e.request_live_consult(crop="Tomato", symptoms="late blight on leaves")
    assert out.get("notifications") and out["notifications"][0]["channel"] == "log"
    # System message records that the expert was notified.
    assert any("notified" in m["text"].lower() for m in out["messages"])
    assert e.get_notifications()        # audit history is non-empty


# --- internet radio (Radio Browser) ---

def test_radio_parse_stations():
    from agrosense.radio import parse_stations
    data = [
        {"name": "AIR Dharwad", "url": "http://x/1", "url_resolved": "https://x/1r",
         "codec": "AAC", "bitrate": 64, "state": "Karnataka", "votes": 10},
        {"name": "Radio Indigo 91.9 FM", "url": "https://x/2", "codec": "MP3",
         "bitrate": 128, "state": "Karnataka"},
        {"name": "AIR Dharwad", "url": "https://x/dup"},          # duplicate name -> dropped
        {"name": "No URL station", "url": "", "url_resolved": ""},  # no url -> dropped
        {"name": "", "url": "https://x/3"},                        # no name -> dropped
    ]
    out = parse_stations(data, limit=10)
    assert [s["name"] for s in out] == ["AIR Dharwad", "Radio Indigo 91.9 FM"]
    assert out[0]["url"] == "https://x/1r"          # prefers url_resolved
    assert out[1]["codec"] == "MP3" and out[1]["bitrate"] == 128
    assert parse_stations(data, limit=1)[0]["name"] == "AIR Dharwad"  # limit honoured
    assert parse_stations([]) == []


# --- plant-doctor onboarding & verification ---

def test_doctor_validate_and_tags():
    from agrosense.doctors import validate_application, derive_tags
    ok = {"name": "Dr X", "specialization": "Plant Pathology", "region": "KVK",
          "contact": "x@y.com", "languages": "English, Hindi", "credentials": "PhD",
          "registration_no": "ICAR-12345"}
    assert validate_application(ok) is None
    assert "name" in validate_application({**ok, "name": ""})
    assert "language" in validate_application({**ok, "languages": ""})
    assert "Credentials" in validate_application({**ok, "credentials": ""})
    assert "registration" in validate_application({**ok, "registration_no": ""}).lower()
    assert "format" in validate_application({**ok, "registration_no": "x@"}).lower()
    assert "pathology" in derive_tags("Plant Pathology") and "agronomy" in derive_tags("x")
    assert "entomology" in derive_tags("Entomology / pests")


def test_doctor_ratings_summary_and_range():
    import json, tempfile
    from pathlib import Path
    from agrosense.doctors import DoctorRegistry
    with tempfile.TemporaryDirectory() as d:
        seed = Path(d) / "seed.json"
        seed.write_text(json.dumps([{"id": "v1", "name": "Doc V", "specialization": "Agronomy",
            "tags": ["agronomy"], "languages": ["English"], "region": "KVK", "available": True}]),
            encoding="utf-8")
        reg = DoctorRegistry(path=Path(d) / "docs.json", seed_path=seed)
        assert reg.get("v1").rating_summary() == (None, 0)
        reg.add_rating("v1", 5, "excellent")
        reg.add_rating("v1", 4)
        avg, count = reg.get("v1").rating_summary()
        assert avg == 4.5 and count == 2
        prof = reg.get("v1").public_profile()
        assert prof["rating_avg"] == 4.5 and prof["rating_count"] == 2
        assert "contact" in prof and prof["ratings"][0]["comment"] == "excellent"
        for bad in (0, 6, "x"):
            try:
                reg.add_rating("v1", bad); assert False
            except ValueError:
                pass
        try:
            reg.add_rating("nope", 5); assert False
        except KeyError:
            pass


def test_doctor_registry_onboard_and_verify():
    import json, tempfile
    from pathlib import Path
    from agrosense.doctors import DoctorRegistry
    with tempfile.TemporaryDirectory() as d:
        seed = Path(d) / "seed.json"
        seed.write_text(json.dumps([{"id": "e1", "name": "Seed Doc",
            "specialization": "Agronomy", "tags": ["agronomy"], "languages": ["English"],
            "region": "KVK", "available": True}]), encoding="utf-8")
        reg = DoctorRegistry(path=Path(d) / "docs.json", seed_path=seed)
        assert len(reg.verified()) == 1                         # seed pre-verified
        doc = reg.onboard({"name": "Dr Asha", "specialization": "Entomology",
                           "region": "Pune", "contact": "a@b.com",
                           "languages": "English, Hindi", "credentials": "PhD Ento",
                           "registration_no": "REG-1234"})
        assert doc.status == "pending" and "entomology" in doc.tags
        assert len(reg.verified()) == 1                         # pending not routable yet
        reg.set_status(doc.id, "verified")
        assert len(reg.verified()) == 2 and reg.get(doc.id).status == "verified"
        reg.set_status(doc.id, "rejected", "incomplete")
        assert len(reg.verified()) == 1
        try:
            reg.onboard({"name": "No Creds", "specialization": "x", "region": "y",
                         "contact": "z", "languages": "English"}); assert False
        except ValueError:
            pass
        assert reg.get("nope") is None


def test_engine_doctor_onboard_then_verify_enters_directory():
    from agrosense.config import DATA_DIR
    from pathlib import Path
    p = Path(DATA_DIR) / "plant_doctors.json"
    original = p.read_text(encoding="utf-8") if p.exists() else None
    try:
        e = _engine()
        before = len(e.list_experts())
        doc = e.onboard_doctor({"name": "Dr Neem", "specialization": "Entomology",
                                "region": "Nagpur", "contact": "n@e.com",
                                "languages": "English, Marathi", "credentials": "PhD",
                                "registration_no": "ICAR-7788"})
        assert doc["status"] == "pending"
        # pending -> appears in admin listing but NOT in the verified consult directory
        assert any(x["id"] == doc["id"] for x in e.list_doctors("pending"))
        assert all(x["id"] != doc["id"] for x in e.list_experts())
        # cannot rate / view an unverified doctor
        assert e.get_doctor_profile(doc["id"]) is None
        # verify -> now in the directory (routable) with a public profile
        e.verify_doctor(doc["id"], approve=True)
        assert any(x["id"] == doc["id"] for x in e.list_experts())
        assert len(e.list_experts()) == before + 1
        # rate the now-verified doctor and see it reflected in the profile + directory
        prof = e.rate_doctor(doc["id"], 5, "very helpful")
        assert prof["rating_avg"] == 5.0 and prof["rating_count"] == 1
        listed = next(x for x in e.list_experts() if x["id"] == doc["id"])
        assert listed["rating_avg"] == 5.0 and listed["registration_no"] == "ICAR-7788"
    finally:
        if original is None:
            p.unlink(missing_ok=True)
        else:
            p.write_text(original, encoding="utf-8")


# --- knowledge-base admin (CRUD + live reload) ---

def test_kb_validate_entry():
    from agrosense.kb_admin import validate_entry
    assert validate_entry({"crop": "Paddy", "source": "ICAR"}) is None
    assert "crop" in validate_entry({"source": "x"})
    assert "source" in validate_entry({"crop": "x"})
    assert "rainfall_mm" in validate_entry({"crop": "x", "source": "y", "rainfall_mm": "lots"})


def test_kbstore_crud_on_tempfile():
    import json
    import tempfile
    from pathlib import Path
    from agrosense.kb_admin import KBStore
    with tempfile.TemporaryDirectory() as d:
        p = Path(d) / "kb.json"
        p.write_text("[]", encoding="utf-8")
        store = KBStore(p)
        e = store.add({"crop": "Quinoa", "soil_type": "Sandy", "source": "Test",
                       "rainfall_mm": "500"})
        assert e["id"] == "quinoa-001" and e["rainfall_mm"] == 500     # id + coercion
        assert len(store.entries()) == 1
        store.update("quinoa-001", {"crop": "Quinoa", "source": "Test2", "soil_type": "Loamy"})
        assert store.entries()[0]["soil_type"] == "Loamy"
        assert store.entries()[0]["source"] == "Test2"
        try:
            store.add({"crop": ""})
            assert False, "expected ValueError"
        except ValueError:
            pass
        assert store.delete("quinoa-001") is True and store.entries() == []
        assert store.delete("nope") is False
        assert json.loads(p.read_text(encoding="utf-8")) == []         # persisted


def test_engine_kb_admin_add_retrieve_delete_with_reload():
    from agrosense.config import KB_PATH
    from pathlib import Path
    original = Path(KB_PATH).read_text(encoding="utf-8")
    try:
        e = _engine()
        before = e.num_documents
        added = e.add_kb_entry({"crop": "Dragonfruit", "soil_type": "Sandy loam",
                                "recommended_fertilizer": "NPK 60:60:60",
                                "source": "Admin test"})
        assert e.num_documents == before + 1                       # index rebuilt
        ans = e.answer("dragonfruit fertilizer")
        assert "Dragonfruit" in {c.crop for c in ans.citations}    # now retrievable
        e.delete_kb_entry(added["id"])
        assert e.num_documents == before
    finally:
        Path(KB_PATH).write_text(original, encoding="utf-8")


def test_engine_environment_profile_serializes_all_fields():
    e = _engine()
    e._environment = _FakeEnvClient()
    d = e.get_environment(location="Belagavi").to_dict()
    for key in ("latitude", "longitude", "elevation_m", "population", "humidity_pct",
                "sunlight", "wind", "air_quality", "pollen", "groundwater"):
        assert key in d
    assert d["elevation_m"] == 769.0 and d["wind"]["direction_compass"] == "W"


def test_subsidies_filter_get_and_updates():
    from agrosense import subsidies
    subsidies._CACHE = None
    all_schemes = subsidies.list_schemes()
    assert len(all_schemes) >= 10
    central = subsidies.list_schemes(level="central")
    state = subsidies.list_schemes(level="state")
    assert central and state and all(s["level"] == "central" for s in central)
    assert len(central) + len(state) == len(all_schemes)
    # state filter keeps central (apply everywhere) + the matching state's schemes
    ka = subsidies.list_schemes(state="Karnataka")
    assert any(s["level"] == "central" for s in ka)
    assert any(s["state"] == "Karnataka" for s in ka)
    assert not any(s["level"] == "state" and s["state"] != "Karnataka" for s in ka)
    # search matches name/summary/category
    found = subsidies.list_schemes(search="insurance")
    assert found and all("insurance" in (s["name"] + s["summary"] + s["category"]).lower()
                         for s in found)
    # full detail
    detail = subsidies.get_scheme("pm-kisan")
    assert detail and detail["application_process"] and detail["documents"] and detail["portal"]
    assert subsidies.get_scheme("does-not-exist") is None
    # recent_updates aggregates across schemes, newest first, capped
    ups = subsidies.recent_updates(limit=5)
    assert len(ups) <= 5 and all("scheme" in u and "date" in u for u in ups)
    assert ups == sorted(ups, key=lambda u: u["date"], reverse=True)


def test_subsidies_add_update_persists_on_tempfile():
    import json
    import tempfile
    from pathlib import Path
    from agrosense import subsidies
    data = [{"id": "demo", "name": "Demo Scheme", "level": "central", "state": "",
             "category": "Test", "summary": "x", "updates": []}]
    with tempfile.TemporaryDirectory() as d:
        p = Path(d) / "subsidies.json"
        p.write_text(json.dumps(data), encoding="utf-8")
        s = subsidies.add_update("demo", "2026-05-30", "New guidelines issued.", path=p)
        assert s["updates"][-1]["text"] == "New guidelines issued."
        on_disk = json.loads(p.read_text(encoding="utf-8"))
        assert on_disk[0]["updates"][-1]["date"] == "2026-05-30"          # persisted
        try:
            subsidies.add_update("nope", "2026-05-30", "x", path=p)
            assert False, "expected KeyError"
        except KeyError:
            pass
    subsidies._CACHE = None      # add_update invalidated cache; ensure real data reloads


def test_engine_subsidies_flow():
    e = _engine()
    schemes = e.list_subsidies(level="central")
    assert schemes and all(s["level"] == "central" for s in schemes)
    one = e.get_subsidy(schemes[0]["id"])
    assert one and "application_process" in one
    assert isinstance(e.subsidy_updates(limit=3), list)


def test_engine_subsidies_resolves_city_to_state():
    # The UI sends the sidebar Location (a CITY, e.g. "Belagavi") as `state`.
    # The engine must geocode it to admin1 ("Karnataka") so state schemes show.
    import agrosense.weather as weather
    e = _engine()
    orig = weather.geocode
    weather.geocode = lambda q: {"admin1": "Karnataka"} if "belagavi" in q.lower() else None
    try:
        ka = e.list_subsidies(state="Belagavi")            # city input
        assert any(s["level"] == "state" and s["state"] == "Karnataka" for s in ka), \
            "city location must surface its state's schemes"
        assert not any(s["level"] == "state" and s["state"] != "Karnataka" for s in ka)
        # a state name passed directly still works (no geocode needed)
        ka2 = e.list_subsidies(state="Karnataka")
        assert any(s["state"] == "Karnataka" for s in ka2)
        # unresolved place -> graceful fallback: show all (don't hide state schemes)
        allp = e.list_subsidies(state="Nowhereville")
        states = {s["state"] for s in allp if s["level"] == "state"}
        assert len(states) >= 3
    finally:
        weather.geocode = orig


def test_finance_catalog_filter_and_get():
    from agrosense import finance
    finance._CACHE = None
    products = finance.list_products()
    assert len(products) >= 10
    cats = finance.categories()
    assert cats and "Crop loan / working capital" in cats
    # category + search filters
    crop = finance.list_products(category="Crop loan")
    assert crop and all("crop loan" in p["category"].lower() for p in crop)
    found = finance.list_products(search="tractor")
    assert found and any("Machinery" in p["name"] or "machinery" in p["summary"].lower()
                         for p in found)
    # full detail
    kcc = finance.get_product("kcc-credit")
    assert kcc and kcc["application_process"] and kcc["documents"] and kcc["interest"]
    assert finance.get_product("nope") is None


def test_finance_apply_validation_and_persist():
    import json
    from pathlib import Path
    from agrosense import finance
    assert finance.validate_application("", "x") and "name" in finance.validate_application("", "x").lower()
    assert "contact" in finance.validate_application("Ravi", "").lower()
    p = Path(finance.APPLICATIONS_PATH)
    original = p.read_text(encoding="utf-8") if p.exists() else None
    try:
        # unknown product -> KeyError
        try:
            finance.apply("nope", "Ravi", "999")
            assert False, "expected KeyError"
        except KeyError:
            pass
        # missing contact -> ValueError
        try:
            finance.apply("kcc-credit", "Ravi", "")
            assert False, "expected ValueError"
        except ValueError:
            pass
        app = finance.apply("kcc-credit", "Ravi", "98765", amount="50000",
                            location="Belagavi", message="crop loan", at="10:00")
        assert app["id"] == "FA0001" and app["product"] and app["amount"] == 50000.0
        app2 = finance.apply("agri-term-loan", "Asha", "asha@x.in")
        assert app2["id"] == "FA0002" and app2["amount"] is None
        apps = finance.list_applications()
        assert apps[0]["id"] == "FA0002"                       # newest first
        on_disk = json.loads(p.read_text(encoding="utf-8"))
        assert len(on_disk) == 2 and on_disk[0]["name"] == "Ravi"   # persisted, insertion order
    finally:
        if original is None:
            p.unlink(missing_ok=True)
        else:
            p.write_text(original, encoding="utf-8")


def test_engine_finance_flow():
    from pathlib import Path
    from agrosense import finance
    e = _engine()
    crop = e.list_finance(category="Crop loan")
    assert crop and all("crop loan" in p["category"].lower() for p in crop)
    assert "Farm mechanization" in e.finance_categories()
    one = e.get_finance(crop[0]["id"])
    assert one and "application_process" in one
    p = Path(finance.APPLICATIONS_PATH)
    original = p.read_text(encoding="utf-8") if p.exists() else None
    try:
        app = e.apply_finance("agri-gold-loan", "Murugan", "90000", amount=20000)
        assert app["product_id"] == "agri-gold-loan" and app["status"] == "received"
        assert any(a["id"] == app["id"] for a in e.list_finance_applications())
        try:
            e.apply_finance("agri-gold-loan", "X", "")          # missing contact
            assert False, "expected ValueError"
        except ValueError:
            pass
    finally:
        if original is None:
            p.unlink(missing_ok=True)
        else:
            p.write_text(original, encoding="utf-8")


def test_land_records_directory_and_fallback():
    from agrosense import land_records
    land_records._CACHE = None
    states = land_records.list_states()
    assert "Karnataka" in states and "Maharashtra" in states and "_default" not in states
    ka = land_records.get_for_state("karnataka")          # case-insensitive
    assert ka["system"] == "Bhoomi" and "Pahani" in ka["record_name"]
    assert ka["portal"].startswith("http") and ka["steps"] and ka["search_by"]
    mh = land_records.get_for_state("Maharashtra")
    assert "7/12" in mh["record_name"]
    # unknown state -> generic _default with the requested name + covered flag
    fallback = land_records.get_for_state("Atlantis")
    assert fallback["state"] == "Atlantis" and fallback.get("covered") is False
    assert fallback["portal"].startswith("http")          # still points somewhere useful
    # no state at all -> default without covered flag set False on a real name
    none = land_records.get_for_state(None)
    assert none["steps"] and "covered" not in none


def test_land_records_guide_bytes():
    from agrosense import land_records
    entry = land_records.get_for_state("Karnataka")
    try:
        pdf = land_records.build_guide(entry, "pdf")
        assert pdf[:4] == b"%PDF" and len(pdf) > 500
        docx = land_records.build_guide(entry, "docx")
        assert docx[:2] == b"PK" and len(docx) > 500      # docx is a zip
    except ImportError:
        print("  (skipped guide-bytes: python-docx/fpdf2 not installed)")
    try:
        land_records.build_guide(entry, "rtf")
        assert False, "expected ValueError for bad format"
    except ValueError:
        pass


def test_engine_land_record_resolves_city_and_builds_guide():
    import agrosense.weather as weather
    e = _engine()
    orig = weather.geocode
    weather.geocode = lambda q: {"admin1": "Karnataka"} if "belagavi" in q.lower() else None
    try:
        info = e.land_record_info(location="Belagavi")    # city -> Karnataka
        assert info["system"] == "Bhoomi" and "disclaimer" in info
        assert "Karnataka" in info["covered_states"]
        info2 = e.land_record_info(state="Maharashtra")   # explicit state
        assert "7/12" in info2["record_name"]
        # unresolved city -> generic guide, not a crash
        info3 = e.land_record_info(location="Nowhereville")
        assert info3["steps"]
        try:
            fname, data = e.land_record_guide(state="Karnataka", fmt="pdf")
            assert fname.endswith(".pdf") and "Karnataka" in fname and data[:4] == b"%PDF"
        except ImportError:
            print("  (skipped guide build: doc deps not installed)")
    finally:
        weather.geocode = orig


if __name__ == "__main__":
    failures = 0
    for name, fn in sorted(globals().items()):
        if name.startswith("test_") and callable(fn):
            try:
                fn()
                print(f"PASS {name}")
            except AssertionError as exc:
                failures += 1
                print(f"FAIL {name}: {exc}")
    print(f"\n{'OK' if not failures else f'{failures} FAILED'}")
    raise SystemExit(1 if failures else 0)