dennisthiessen
02b8df58f0
The early-warning score showed n/a because it required an exact date match between the live benchmark (Alpaca, may have today's bar) and the stored universe breadth (DB, often a day behind), which blanked the newest snapshot — the one the UI displays. - Look up the divergence as-of the snapshot date (newest value within a 7-day lag) instead of requiring an exact match. - Backfill early_warning + combined onto recent existing snapshots (the index history predates this signal) so the 7/30-day trends populate on the first run rather than only filling in over the coming weeks. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
167 lines
6.3 KiB
Python
167 lines
6.3 KiB
Python
"""Unit tests for the regime-monitor pure functions and aggregation."""
|
|
|
|
from __future__ import annotations
|
|
|
|
from datetime import date, timedelta
|
|
|
|
from app.services.regime_monitor_service import (
|
|
DEFAULT_CONFIG,
|
|
_attach_early_warning,
|
|
band_for,
|
|
compute_regime_score,
|
|
f2_credit_spreads,
|
|
p1_trend_break,
|
|
p2_death_cross,
|
|
p3_drawdown,
|
|
p4_relative_strength,
|
|
p5_volatility,
|
|
p6_canary,
|
|
_compute_index,
|
|
)
|
|
|
|
|
|
def _dated(values: list[float], end: date = date(2026, 6, 26)) -> list[tuple[date, float]]:
|
|
n = len(values)
|
|
return [(end - timedelta(days=(n - 1 - i)), v) for i, v in enumerate(values)]
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Bands
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def test_band_for():
|
|
assert band_for(10) == "stable"
|
|
assert band_for(45) == "watch"
|
|
assert band_for(70) == "elevated"
|
|
assert band_for(90) == "breaking"
|
|
|
|
|
|
def test_attach_early_warning_blends():
|
|
result = {"total_score": 80.0}
|
|
_attach_early_warning(result, 40.0, {"coincident": 0.6, "early_warning": 0.4})
|
|
assert result["early_warning"]["score"] == 40.0
|
|
assert result["early_warning"]["band"] == "watch"
|
|
# combined = (80*0.6 + 40*0.4) / 1.0 = 64
|
|
assert result["combined"]["score"] == 64.0
|
|
assert result["combined"]["band"] == "elevated"
|
|
|
|
|
|
def test_attach_early_warning_none_falls_back_to_index():
|
|
result = {"total_score": 80.0}
|
|
_attach_early_warning(result, None, {"coincident": 0.6, "early_warning": 0.4})
|
|
assert result["early_warning"]["score"] is None
|
|
assert result["combined"]["score"] == 80.0 # no early warning -> just the index
|
|
|
|
|
|
def test_divergence_asof_tolerates_small_lag():
|
|
from app.services.regime_monitor_service import _divergence_asof
|
|
items = [(date(2026, 6, 1), 55.0), (date(2026, 6, 3), 60.0)]
|
|
assert _divergence_asof(items, date(2026, 6, 3)) == 60.0 # exact date
|
|
assert _divergence_asof(items, date(2026, 6, 4)) == 60.0 # 1-day lag -> newest
|
|
assert _divergence_asof(items, date(2026, 6, 20)) is None # too stale
|
|
assert _divergence_asof([], date(2026, 6, 3)) is None
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Price sub-scores
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def test_p1_blends_leader_double():
|
|
smh_under = [100.0] * 199 + [50.0] # last below its 200-DMA
|
|
qqq_above = [100.0] * 200 # last at/above its 200-DMA -> healthy
|
|
score = p1_trend_break(smh_under, qqq_above, leader_weight=2.0)
|
|
# leader(100) weighted 2, confirm(0) weighted 1 -> 66.7
|
|
assert round(score, 1) == 66.7
|
|
|
|
|
|
def test_p1_none_without_history():
|
|
assert p1_trend_break([100.0] * 50, [100.0] * 50, 2.0) is None
|
|
|
|
|
|
def test_p2_death_cross_bearish_vs_healthy():
|
|
bearish = [300.0 - i for i in range(260)] # falling: 50 < 200, slope down
|
|
healthy = [100.0 + i * 0.5 for i in range(260)] # rising: 50 > 200
|
|
assert p2_death_cross(bearish, bearish, 2.0) > 0
|
|
assert p2_death_cross(healthy, healthy, 2.0) == 0
|
|
|
|
|
|
def test_p3_drawdown_linear():
|
|
closes = [100.0] * 252 + [80.0] # 20% below the 52w high -> 100
|
|
assert p3_drawdown(closes, [100.0] * 253) == 100.0
|
|
|
|
|
|
def test_p4_relative_strength_direction():
|
|
falling = [100.0 - i * 0.5 for i in range(70)] # SMH underperforms flat SPY
|
|
rising = [100.0 + i * 0.5 for i in range(70)]
|
|
spy = [100.0] * 70
|
|
assert p4_relative_strength(falling, spy, 60) > 50
|
|
assert p4_relative_strength(rising, spy, 60) < 50
|
|
|
|
|
|
def test_p5_volatility_linear():
|
|
assert p5_volatility(15) == 0
|
|
assert p5_volatility(30) == 100
|
|
assert p5_volatility(22.5) == 50
|
|
assert p5_volatility(None) is None
|
|
|
|
|
|
def test_f2_credit_percentile():
|
|
rising = [float(i) for i in range(1, 31)] # latest is the max -> ~100th pct
|
|
assert f2_credit_spreads(rising) == 100.0
|
|
falling = [float(i) for i in range(30, 0, -1)] # latest is the min
|
|
assert f2_credit_spreads(falling) < 10
|
|
assert f2_credit_spreads([1.0] * 5) is None # too short
|
|
|
|
|
|
def test_p6_canary_divergence():
|
|
nvda_weak = [100.0] * 49 + [80.0] # below its 50-DMA
|
|
smh_intact = [100.0] * 199 + [120.0] # above its 200-DMA
|
|
assert p6_canary(nvda_weak, smh_intact) == 100.0
|
|
assert p6_canary([100.0] * 50, smh_intact) == 0.0
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Aggregation
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def test_compute_regime_score_excludes_na_and_zero_weight():
|
|
weights = {"P1": 10, "P2": 0, "F2": 5}
|
|
subs = {"P1": 80.0, "P2": 50.0, "F2": None}
|
|
result = compute_regime_score(subs, weights)
|
|
# Only P1 counts: P2 weight 0, F2 unavailable.
|
|
assert result["total_score"] == 80.0
|
|
ids = {row["id"]: row for row in result["breakdown"]}
|
|
assert "P2" not in ids # zero-weight signals are hidden
|
|
assert ids["F2"]["available"] is False
|
|
assert ids["P1"]["contribution"] == 80.0
|
|
|
|
|
|
def test_compute_regime_score_contributions_sum_to_total():
|
|
weights = {"P1": 10, "F2": 10}
|
|
subs = {"P1": 80.0, "F2": 40.0}
|
|
result = compute_regime_score(subs, weights)
|
|
assert result["total_score"] == 60.0
|
|
total = sum(row["contribution"] for row in result["breakdown"])
|
|
assert round(total, 1) == 60.0
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# As-of index replay (backfill mechanics)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
def test_compute_index_as_of_truncates_history():
|
|
rising = [100.0 + i * 0.2 for i in range(260)]
|
|
prices = {sym: _dated(rising) for sym in ("SMH", "QQQ", "SPY", "RSP", "NVDA")}
|
|
overrides = {"f1_score": 50.0, "f3_score": 50.0}
|
|
|
|
full = _compute_index(prices, None, None, overrides, DEFAULT_CONFIG, date(2026, 6, 26))
|
|
by_id = {r["id"]: r for r in full["breakdown"]}
|
|
assert by_id["P1"]["available"] is True # 200-DMA computable on full history
|
|
assert 0 <= full["total_score"] <= 100
|
|
assert full["band"] in {"stable", "watch", "elevated", "breaking"}
|
|
|
|
# As-of 250 days earlier: only ~10 bars are in scope -> long-lookback signals n/a.
|
|
early = _compute_index(prices, None, None, overrides, DEFAULT_CONFIG, date(2026, 6, 26) - timedelta(days=250))
|
|
early_by_id = {r["id"]: r for r in early["breakdown"]}
|
|
assert early_by_id["P1"]["available"] is False
|