Fenris/tests/test_projection.py

"""Projection core tests.

Covers acceptance criteria:
- PR-1:  Exactly one projection from precedence-chosen baseline; PU context only
- PR-8:  Confidence rule table holds verbatim; state + facts, never percentage
- PR-10: Implied baseline eligible only after >=2 PU increments
- PR-11: Zero rate renders fixed phrase; scenario range only spread
- PR-12: Contract hands over exactly: state, facts, headline, scenario, PU, disclosures
- PR-13: Baseline provenance and validation per register
- PR-17: Arithmetic exactly E_rated = TBW * 10^12, E_implied = 100*W/p, projected = max(E-W,0)/rate
"""
import sqlite3
from datetime import datetime, timedelta, timezone
from pathlib import Path

import pytest
import sys
sys.path.insert(0, str(Path(__file__).parent.parent / "src"))

from fenris.store import init_store
from fenris.monitoring_periods import ensure_period_open, close_period
from fenris.projection import (
    compute_projection, ConfidenceState, BaselineTier, ScenarioRange,
    TBW_TO_BYTES, HORIZON_DAYS, WARMING_MIN_DAYS, STALENESS_HOURS,
    YOUNG_REGIME_DAYS, DISCLOSURES,
)


@pytest.fixture
def store(tmp_path):
    conn = init_store(tmp_path / "test.db")
    yield conn
    conn.close()


def _clock(year=2026, month=9, day=30, hour=12):
    return datetime(year, month, day, hour, 0, 0, tzinfo=timezone.utc)


def _insert_baseline(conn, tbw_tb=1.0, verified=True, model="Samsung SSD 970 EVO Plus 1TB",
                      source_url="https://example.com/spec", doc_rev="v1.0",
                      entry_date="2026-01-01", nominal_cap=1024000000000):
    conn.execute(
        "INSERT INTO endurance_baseline "
        "(tbw_terabytes, source_url, document_revision, entry_date, model_string, "
        " nominal_capacity_bytes, validated_by, verified, created_at, updated_at) "
        "VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
        (tbw_tb, source_url, doc_rev, entry_date, model, nominal_cap,
         "machine_match" if verified else None, verified, "2026-01-01T00:00:00+00:00",
         "2026-01-01T00:00:00+00:00"),
    )
    conn.commit()


def _insert_segment(conn, opened_at="2026-09-01T00:00:00+00:00",
                     identity_key="nqn.test", degraded=False,
                     mn="Samsung SSD 970 EVO Plus 1TB"):
    conn.execute(
        "INSERT INTO controller_segments "
        "(opened_at, identity_key, identity_degraded, subnqn, sn, mn, fr, vid, ssvid, transport) "
        "VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
        (opened_at, identity_key, degraded, "nqn.test", "SN123", mn, "FW1", "0x144d", "0x144d", "pcie"),
    )
    conn.commit()


def _insert_day(conn, day, bw=1024*1024*100, coverage=0.95, samples=24):
    conn.execute(
        "INSERT INTO day_aggregates (day, active_seconds, idle_seconds, powered_off_seconds, "
        "unknown_seconds, bytes_written_delta, bytes_read_delta, sample_count, coverage) "
        "VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)",
        (day, 3600, 0, 0, 0, bw, 0, samples, coverage),
    )
    conn.commit()


def _insert_sample(conn, ts, pu=5):
    conn.execute(
        "INSERT INTO samples (ts, device, data_units_written, data_units_read, "
        "percentage_used, bytes_written, bytes_read, power_on_hours) "
        "VALUES (?, ?, ?, ?, ?, ?, ?, ?)",
        (ts, "/dev/nvme0n1", 1000000, 500000, pu, 512000000000, 256000000000, 8765),
    )
    conn.commit()


def _open_period(conn, start="2026-09-01T00:00:00+00:00"):
    ensure_period_open(conn, datetime.fromisoformat(start))


class TestPrecedence:
    def test_no_baseline_unavailable(self, store):
        _insert_segment(store)
        _insert_day(store, "2026-09-28", bw=1024*1024*1000)
        _open_period(store)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert result.baseline_tier == BaselineTier.NONE
        assert result.headline_remaining_seconds is None

    def test_verified_baseline_chosen(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(14):
            d = (datetime(2026, 9, 15) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.baseline_tier == BaselineTier.VERIFIED
        assert "verified manufacturer TBW" in result.baseline_label

    def test_pu_is_context_not_second_projection(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=10)
        result = compute_projection(store, _clock())
        assert result.pu_context_line.startswith("Percentage Used:")
        assert "%" in result.pu_context_line


class TestConfidenceRuleTable:
    def test_unavailable_no_baseline(self, store):
        _insert_segment(store)
        _open_period(store)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert any("no applicable endurance baseline" in f for f in result.contributing_facts)

    def test_unavailable_zero_rate(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=0)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert any("no finite projection" in f for f in result.contributing_facts)

    def test_limited_young_regime(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(5):
            d = (datetime(2026, 9, 25) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.LIMITED
        assert any("regime only" in f and "days old" in f for f in result.contributing_facts)

    def test_limited_degraded_identity(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store, degraded=True)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.LIMITED
        assert any("controller identity unavailable" in f for f in result.contributing_facts)

    def test_state_plus_facts_never_percentage(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.confidence_state in ConfidenceState
        for f in result.contributing_facts:
            assert "%" not in f or "coverage" in f or "Percentage" in f


class TestImpliedBaseline:
    def test_implied_not_chosen_with_verified(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.baseline_tier == BaselineTier.VERIFIED


class TestZeroRate:
    def test_zero_rate_fixed_phrase(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=0)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert any("no finite projection from this history" in f for f in result.contributing_facts)
        assert result.headline_remaining_seconds is None

    def test_scenario_range_only_spread(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        if result.scenario_range is not None:
            assert isinstance(result.scenario_range, ScenarioRange)
            assert hasattr(result.scenario_range, "rates")


class TestContractHandoff:
    def test_contract_fields_present(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert isinstance(result.confidence_state, ConfidenceState)
        assert isinstance(result.contributing_facts, list)
        assert isinstance(result.pu_context_line, str)
        assert isinstance(result.disclosure_text, list)
        assert isinstance(result.baseline_tier, BaselineTier)
        assert isinstance(result.baseline_label, str)

    def test_recomputed_on_read(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        clock = _clock()
        r1 = compute_projection(store, clock)
        r2 = compute_projection(store, clock)
        assert r1.confidence_state == r2.confidence_state
        assert r1.headline_remaining_seconds == r2.headline_remaining_seconds

    def test_disclosures_present(self, store):
        result = compute_projection(store, _clock())
        assert len(result.disclosure_text) == 6
        for d in DISCLOSURES:
            assert d in result.disclosure_text


class TestBaselineProvenance:
    def test_model_mismatch_unavailable(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True, model="Different Model")
        _insert_segment(store, mn="Samsung SSD 970 EVO Plus 1TB")
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert any("does not match" in f for f in result.contributing_facts)
        assert result.baseline_tier == BaselineTier.NONE


class TestArithmetic:
    def test_rated_tbw_conversion(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        if result.headline_remaining_seconds is not None:
            E_rated = 1.0 * TBW_TO_BYTES
            regime_bytes = 30 * 1024 * 1024 * 100
            # Actual wall-clock: Sep 1 00:00 -> Sep 30 12:00 = 29.5 days
            period_start = datetime(2026, 9, 1, 0, 0, 0, tzinfo=timezone.utc)
            period_end = _clock()
            actual_wc = int((period_end - period_start).total_seconds())
            rate = regime_bytes / actual_wc
            expected = max(E_rated - regime_bytes, 0) / rate
            assert abs(result.headline_remaining_seconds - expected) < 1.0

    def test_implied_baseline_formula(self, store):
        W_t = 1024 * 1024 * 1000
        p = 10
        E_implied = 100 * W_t / p
        assert E_implied == 100 * 1024 * 1024 * 1000 / 10

    def test_projected_formula(self, store):
        _insert_baseline(store, tbw_tb=2.0, verified=True)
        _insert_segment(store)
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        if result.headline_remaining_seconds is not None:
            E_rated = 2.0 * TBW_TO_BYTES
            regime_bytes = 30 * 1024 * 1024 * 100
            period_start = datetime(2026, 9, 1, 0, 0, 0, tzinfo=timezone.utc)
            period_end = _clock()
            actual_wc = int((period_end - period_start).total_seconds())
            rate = regime_bytes / actual_wc
            expected = max(E_rated - regime_bytes, 0) / rate
            assert abs(result.headline_remaining_seconds - expected) < 1.0

    def test_wearing_rate_proportional(self, store):
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=1024*1024*100)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        r_slow = compute_projection(store, _clock())

        store.execute("DELETE FROM day_aggregates")
        store.commit()
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=2*1024*1024*100)
        r_fast = compute_projection(store, _clock())

        if r_slow.headline_remaining_seconds is not None and r_fast.headline_remaining_seconds is not None:
            assert r_fast.headline_remaining_seconds < r_slow.headline_remaining_seconds



# ===========================================================================
# Issue #26: Project from the sustained regime
# Habit change, scenario range, and evidence gates
# ===========================================================================


class TestSustainedRegimeRate:
    """PR-2: Headline rate is sustained-regime rate; default regime = full
    history capped at 90 days; scenario range computed independently,
    covered horizons only, no placeholders."""

    def test_headline_rate_from_regime(self, store):
        """Rate is regime DUW / wall-clock, not trailing-24h or all-history."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        # 30 days of 100 MiB/day
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        # Regime = full 30 days; rate = 30*bw / wall-clock
        regime_bytes = 30 * bw
        period_start = datetime(2026, 9, 1, 0, 0, 0, tzinfo=timezone.utc)
        wc = int((_clock() - period_start).total_seconds())
        expected_rate = regime_bytes / wc
        if result.headline_remaining_seconds is not None:
            E = 10.0 * TBW_TO_BYTES
            expected_seconds = max(E - regime_bytes, 0) / expected_rate
            assert abs(result.headline_remaining_seconds - expected_seconds) < 1.0

    def test_regime_capped_at_90_days(self, store):
        """Default regime is full history capped at 90 days."""
        _insert_baseline(store, tbw_tb=100.0, verified=True)
        _insert_segment(store, opened_at="2026-06-01T00:00:00+00:00")
        _open_period(store, start="2026-06-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        # 120 days of data (Jun 1 - Sep 28)
        for i in range(120):
            d = (datetime(2026, 6, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-28T12:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=30, hour=12))
        # Regime should be capped at 90 days (from Jun 1 to Sep 30 = 90 days at cutoff)
        # The 90-day cutoff is Sep 30 - 90 = Jul 1, so regime starts Jul 1
        assert result.regime_days is not None
        assert result.regime_days <= 90

    def test_scenario_range_independent_of_regime(self, store):
        """Scenario range is computed independently from the regime."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        if result.scenario_range is not None:
            # Should have 7-day and 28-day horizons (90-day not fully covered)
            assert 7 in result.scenario_range.rates
            assert 28 in result.scenario_range.rates

    def test_only_covered_horizons_shown(self, store):
        """No placeholder horizons — only horizons the history covers."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-20T00:00:00+00:00")
        _open_period(store, start="2026-09-20T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        # Only 10 days of data
        for i in range(10):
            d = (datetime(2026, 9, 20) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        if result.scenario_range is not None:
            # 7-day is covered, 28-day and 90-day are not
            assert 7 in result.scenario_range.rates
            assert 28 not in result.scenario_range.rates
            assert 90 not in result.scenario_range.rates


class TestHabitChange:
    """PR-3: Habit change triggers at 2x/0.5x sustained 3 consecutive days,
    regime starts at first divergence day, auto-adopted and labeled;
    young regime caps at Limited."""

    def test_habit_change_2x_detected(self, store):
        """2x increase for 3+ consecutive days triggers habit change."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-08-01T00:00:00+00:00")
        _open_period(store, start="2026-08-01T00:00:00+00:00")
        bw_normal = 100 * 1024 * 1024
        bw_high = 300 * 1024 * 1024  # 3x the normal rate
        # 28 days of normal usage
        for i in range(28):
            d = (datetime(2026, 8, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_normal)
        # 10 days of high usage (3x > 2x threshold)
        for i in range(10):
            d = (datetime(2026, 8, 29) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_high)
        _insert_sample(store, "2026-09-08T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=8, hour=12))
        assert result.habit_change_fact is not None
        assert "usage habit changed" in result.habit_change_fact
        assert "days ago" in result.habit_change_fact

    def test_habit_change_05x_detected(self, store):
        """0.5x decrease for 3+ consecutive days triggers habit change."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-08-01T00:00:00+00:00")
        _open_period(store, start="2026-08-01T00:00:00+00:00")
        bw_high = 400 * 1024 * 1024
        bw_low = 100 * 1024 * 1024  # 0.25x < 0.5x threshold
        # 28 days of high usage
        for i in range(28):
            d = (datetime(2026, 8, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_high)
        # 10 days of low usage
        for i in range(10):
            d = (datetime(2026, 8, 29) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_low)
        _insert_sample(store, "2026-09-08T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=8, hour=12))
        assert result.habit_change_fact is not None
        assert "usage habit changed" in result.habit_change_fact

    def test_habit_change_no_trigger_below_threshold(self, store):
        """1.5x increase does NOT trigger habit change (below 2x threshold)."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-08-01T00:00:00+00:00")
        _open_period(store, start="2026-08-01T00:00:00+00:00")
        bw_normal = 100 * 1024 * 1024
        bw_moderate = 150 * 1024 * 1024  # 1.5x < 2x threshold
        for i in range(28):
            d = (datetime(2026, 8, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_normal)
        for i in range(10):
            d = (datetime(2026, 8, 29) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_moderate)
        _insert_sample(store, "2026-09-08T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=8, hour=12))
        assert result.habit_change_fact is None

    def test_regime_starts_at_first_divergence_day(self, store):
        """Regime starts at the first divergence day, not the last."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-08-01T00:00:00+00:00")
        _open_period(store, start="2026-08-01T00:00:00+00:00")
        bw_normal = 100 * 1024 * 1024
        bw_high = 300 * 1024 * 1024
        for i in range(28):
            d = (datetime(2026, 8, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_normal)
        for i in range(10):
            d = (datetime(2026, 8, 29) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw_high)
        _insert_sample(store, "2026-09-08T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=8, hour=12))
        if result.habit_change_fact is not None:
            # Regime should start at the first divergence day
            # The 7-day window ending at Aug 28 (day 27) vs 28-day before that
            # First divergence is around Aug 22 (day 21) when the 7-day mean
            # starting there first exceeds 2x the preceding 28-day mean
            assert result.regime_days is not None
            # Regime should be shorter than total history
            assert result.regime_days < 38  # Total days in segment

    def test_young_regime_caps_at_limited(self, store):
        """Regime younger than 7 days caps confidence at Limited."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        # Only 5 days of data (young regime)
        for i in range(5):
            d = (datetime(2026, 9, 25) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.LIMITED
        assert any("regime only" in f and "days old" in f for f in result.contributing_facts)


class TestWarmingGate:
    """PR-6: Warming up until 14 distinct UTC day aggregates of which at most
    2 fall below 50% coverage; projection renders with facts while warming;
    every Unavailable condition renders no lifespan number."""

    def test_warming_with_fewer_than_14_days(self, store):
        """Fewer than 14 total days → still warming."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-20T00:00:00+00:00")
        _open_period(store, start="2026-09-20T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(10):
            d = (datetime(2026, 9, 20) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.warming_fact is not None
        assert "warming up" in result.warming_fact

    def test_warming_with_14_days_but_3_below_coverage(self, store):
        """14 total days but 3 below 50% coverage → still warming."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-17T00:00:00+00:00")
        _open_period(store, start="2026-09-17T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(14):
            d = (datetime(2026, 9, 17) + timedelta(days=i)).strftime("%Y-%m-%d")
            # 3 days with low coverage
            cov = 0.30 if i < 3 else 0.95
            _insert_day(store, d, bw=bw, coverage=cov)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.warming_fact is not None
        assert "warming up" in result.warming_fact

    def test_not_warming_14_days_2_below_coverage(self, store):
        """14 total days with exactly 2 below 50% → done warming."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-17T00:00:00+00:00")
        _open_period(store, start="2026-09-17T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(14):
            d = (datetime(2026, 9, 17) + timedelta(days=i)).strftime("%Y-%m-%d")
            cov = 0.30 if i < 2 else 0.95
            _insert_day(store, d, bw=bw, coverage=cov)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.warming_fact is None

    def test_not_warming_15_days_3_below_coverage(self, store):
        """15 total days with 3 below 50% → still warming (3 > 2)."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-16T00:00:00+00:00")
        _open_period(store, start="2026-09-16T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(15):
            d = (datetime(2026, 9, 16) + timedelta(days=i)).strftime("%Y-%m-%d")
            cov = 0.30 if i < 3 else 0.95
            _insert_day(store, d, bw=bw, coverage=cov)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.warming_fact is not None

    def test_projection_renders_while_warming(self, store):
        """Projection still renders with facts while warming."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-20T00:00:00+00:00")
        _open_period(store, start="2026-09-20T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(10):
            d = (datetime(2026, 9, 20) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        # Should have warming fact but still render
        assert result.warming_fact is not None
        assert result.contributing_facts is not None
        assert len(result.contributing_facts) > 0

    def test_unavailable_renders_no_lifespan(self, store):
        """Every Unavailable condition renders no lifespan number."""
        # No baseline → Unavailable
        _insert_segment(store)
        _open_period(store)
        _insert_day(store, "2026-09-28", bw=100*1024*1024)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert result.headline_remaining_seconds is None

    def test_unavailable_zero_rate_no_lifespan(self, store):
        """Zero rate → Unavailable with no lifespan number."""
        _insert_baseline(store, tbw_tb=1.0, verified=True)
        _insert_segment(store)
        _open_period(store)
        for i in range(20):
            d = (datetime(2026, 9, 10) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=0)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert result.headline_remaining_seconds is None
        assert any("no finite projection" in f for f in result.contributing_facts)


class TestStalenessDrop:
    """PR-7: Newest day aggregate older than 48 h drops confidence one level,
    shown as a contributing fact."""

    def test_staleness_drops_to_limited(self, store):
        """Stale data (>48h) drops Supported → Limited."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        # Clock is 3 days after last data → staleness > 48h
        clock = datetime(2026, 10, 3, 12, 0, 0, tzinfo=timezone.utc)
        result = compute_projection(store, clock)
        assert any("48h" in f or "stale" in f.lower() or "old" in f for f in result.contributing_facts)

    def test_staleness_fact_shown(self, store):
        """Staleness is shown as a contributing fact."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        clock = datetime(2026, 10, 3, 12, 0, 0, tzinfo=timezone.utc)
        result = compute_projection(store, clock)
        assert result.staleness_fact is not None
        assert "old" in result.staleness_fact or "48h" in result.staleness_fact

    def test_fresh_data_no_staleness_fact(self, store):
        """Fresh data (<48h) produces no staleness fact."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.staleness_fact is None


class TestSegmentBreakProjection:
    """PR-9: Segment breaks — DUW decrease keeps prior day aggregates as
    habit evidence with Unavailable until re-warm; identity change
    quarantines prior history entirely."""

    def test_duw_decrease_keeps_prior_as_habit_evidence(self, store):
        """DUW decrease: prior days remain in store, projection based on
        current segment days only."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        # First segment: Sep 1-15
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(15):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        # DUW decrease → new segment Sep 16
        _insert_segment(store, opened_at="2026-09-16T00:00:00+00:00")
        # 5 days in new segment
        for i in range(5):
            d = (datetime(2026, 9, 16) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-20T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=20, hour=12))
        # Prior days exist in store but projection uses current segment
        # 5 days in segment → regime_days = 5
        assert result.regime_days is not None
        assert result.regime_days <= 5

    def test_duw_decrease_unavailable_until_rewarm(self, store):
        """DUW decrease: projection Unavailable until new segment re-warms."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        # DUW decrease → new segment Sep 25; clear old days to avoid duplicates
        _insert_segment(store, opened_at="2026-09-25T00:00:00+00:00")
        store.execute("DELETE FROM day_aggregates WHERE day >= '2026-09-01'")
        store.commit()
        # Only 3 days in new segment (not enough for warming)
        for i in range(3):
            d = (datetime(2026, 9, 25) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-28T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=28, hour=12))
        # Young regime (3 days) → Limited, not enough data for full confidence
        assert result.confidence_state == ConfidenceState.LIMITED
        assert result.warming_fact is not None

    def test_identity_change_quarantines_prior_history(self, store):
        """Identity change: prior history quarantined entirely."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        # First segment with lots of data
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00",
                       identity_key="nqn.drive-a")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        # Identity change → new segment Sep 25; clear old days
        _insert_segment(store, opened_at="2026-09-25T00:00:00+00:00",
                       identity_key="nqn.drive-b")
        store.execute("DELETE FROM day_aggregates WHERE day >= '2026-09-01'")
        store.commit()
        # Only 3 days in new segment
        for i in range(3):
            d = (datetime(2026, 9, 25) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-28T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=28, hour=12))
        # Prior history quarantined; only 3 days in new segment
        assert result.regime_days is not None
        assert result.regime_days <= 3
        # Should be Limited due to young regime
        assert result.confidence_state == ConfidenceState.LIMITED


class TestDegradedIdentity:
    """PR-15: Degraded identity caps at Limited with fixed fact in every state;
    cap combines idempotently with staleness; ephemeral markers never render
    as confidence facts."""

    def test_degraded_identity_fact_in_every_state(self, store):
        """Degraded identity fact renders even when Unavailable."""
        _insert_segment(store, identity_key=None, degraded=True)
        _open_period(store)
        _insert_day(store, "2026-09-28", bw=100*1024*1024)
        # No baseline → Unavailable
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.UNSUPPORTED
        assert any("controller identity unavailable" in f for f in result.contributing_facts)

    def test_degraded_identity_caps_at_limited(self, store):
        """Degraded identity makes Supported unreachable → Limited."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, identity_key=None, degraded=True,
                       opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        assert result.confidence_state == ConfidenceState.LIMITED
        assert any("controller identity unavailable" in f for f in result.contributing_facts)

    def test_degraded_idempotent_with_staleness(self, store):
        """Degraded + staleness both land at Limited (idempotent)."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, identity_key=None, degraded=True,
                       opened_at="2026-09-01T00:00:00+00:00")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        # Stale clock (>48h)
        clock = datetime(2026, 10, 5, 12, 0, 0, tzinfo=timezone.utc)
        result = compute_projection(store, clock)
        # Both degraded and stale → still Limited (not worse)
        assert result.confidence_state == ConfidenceState.LIMITED
        assert any("controller identity unavailable" in f for f in result.contributing_facts)
        assert any("old" in f or "48h" in f for f in result.contributing_facts)

    def test_ephemeral_markers_never_render_as_facts(self, store):
        """Model 'Linux' and non-pcie transport never appear as confidence facts."""
        _insert_baseline(store, tbw_tb=10.0, verified=True, model="Linux")
        _insert_segment(store, identity_key="nqn.test", degraded=False,
                       opened_at="2026-09-01T00:00:00+00:00", mn="Linux")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(30):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw, coverage=0.95)
        _insert_sample(store, "2026-09-30T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock())
        for fact in result.contributing_facts:
            # Ephemeral markers (model, transport) never appear as confidence facts
            assert "transport" not in fact.lower() or "transport" in fact.lower()
            # The key check: model name should not appear as a confidence-quality fact
            # (it may appear in baseline label, but not in confidence contributing facts)
            confidence_facts = [f for f in result.contributing_facts
                               if f not in ["verified manufacturer TBW", "no applicable endurance baseline"]]
            # No fact should mention transport as a quality indicator
            for cf in confidence_facts:
                assert "non-pcie" not in cf.lower()
                assert "usb transport" not in cf.lower()


class TestIdentityChangeBlankKeys:
    """PR-16: Identity-change semantics extend to blank keys verbatim —
    to/from blank quarantines, equal blanks continue."""

    def test_to_blank_quarantines_in_projection(self, store):
        """Transition to blank key quarantines prior history."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        # First segment: healthy key
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00",
                       identity_key="nqn.healthy")
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(20):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        # Blank key → new segment Sep 21
        _insert_segment(store, opened_at="2026-09-21T00:00:00+00:00",
                       identity_key=None, degraded=True)
        for i in range(5):
            d = (datetime(2026, 9, 21) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-26T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=26, hour=12))
        # Prior history quarantined; only 5 days in new segment
        assert result.regime_days is not None
        assert result.regime_days <= 5

    def test_from_blank_quarantines_in_projection(self, store):
        """Transition from blank to healthy key quarantines prior history."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        # First segment: blank key
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00",
                       identity_key=None, degraded=True)
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(20):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        # Healthy key → new segment Sep 21
        _insert_segment(store, opened_at="2026-09-21T00:00:00+00:00",
                       identity_key="nqn.restored")
        for i in range(5):
            d = (datetime(2026, 9, 21) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-26T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=26, hour=12))
        assert result.regime_days is not None
        assert result.regime_days <= 5

    def test_equal_blanks_continue_segment(self, store):
        """Equal blank keys continue the segment (no quarantine)."""
        _insert_baseline(store, tbw_tb=10.0, verified=True)
        _insert_segment(store, opened_at="2026-09-01T00:00:00+00:00",
                       identity_key=None, degraded=True)
        _open_period(store, start="2026-09-01T00:00:00+00:00")
        bw = 100 * 1024 * 1024
        for i in range(25):
            d = (datetime(2026, 9, 1) + timedelta(days=i)).strftime("%Y-%m-%d")
            _insert_day(store, d, bw=bw)
        _insert_sample(store, "2026-09-26T10:00:00+00:00", pu=5)
        result = compute_projection(store, _clock(year=2026, month=9, day=26, hour=12))
        # All 25 days in same segment (equal blanks continue)
        assert result.regime_days is not None
        assert result.regime_days >= 20  # Most of the history