Skip to content

160-tool MCP server for real-time global intelligence from free public APIs: markets, liquidity, prediction markets, SEC filings, conflict, military, aircraft-cohort early warning, migration and border flows, cyber, climate, news and more. Dated observation history, 29 research prompts, cited briefs, geofences, SSE dashboard. MIT, no paid API keys.

7.2 KB
Raw Blame
"""Tests for analysis/classifier.py — keyword threat classification.

Pure-function tests with realistic headlines. Each positive case is paired
with a negative (a text that must NOT trigger the same outcome) so a broken
classifier cannot pass by over- or under-matching.
"""

from world_intel_mcp.analysis.classifier import (
    CATEGORIES,
    classify_event,
    fetch_classify_event,
)

# Three military keywords ("troops", "artillery", "frontline"), no severity
# modifiers, and no keywords from any other category.
_MILITARY_TEXT = "Russian troops shelled the frontline with artillery near the border"

# No category keyword or severity modifier appears anywhere in this text.
_BENIGN_TEXT = "The bakery unveiled a fresh sourdough recipe on Sunday"


def test_military_headline_classified() -> None:
    result = classify_event(_MILITARY_TEXT)
    assert result["primary_category"] == "military"
    assert (
        result["severity"] == CATEGORIES["military"]["severity_base"]
    )  # 8, no modifiers
    # 3 keyword hits -> confidence min(1.0, 3*0.2 + 0.3) = 0.9
    assert result["confidence"] == 0.9
    assert result["category_count"] == 1
    assert result["severity_modifiers"] == []
    matched = result["all_categories"][0]["keywords"]
    assert "troops" in matched
    assert "frontline" in matched
    assert result["source"] == "keyword-classifier"


def test_benign_text_is_unclassified() -> None:
    """The load-bearing negative: a classifier that fires on everything
    proves nothing when it fires on a real threat."""
    result = classify_event(_BENIGN_TEXT)
    assert result["primary_category"] == "unclassified"
    assert result["severity"] == 0
    assert result["confidence"] == 0.0
    assert result["all_categories"] == []
    assert result["category_count"] == 0
    assert result["severity_modifiers"] == []


def test_high_severity_modifier_and_cap_at_ten() -> None:
    # 4 nuclear keywords; "nuclear" is also a high-severity modifier (+2),
    # so 9 + 2 must cap at 10.
    result = classify_event("IAEA warns of uranium enrichment at nuclear facility")
    assert result["primary_category"] == "nuclear"
    assert result["severity"] == 10
    assert result["confidence"] == 1.0  # 4 hits saturates the scale
    assert "nuclear" in result["severity_modifiers"]


def test_moderate_severity_modifier_adds_one() -> None:
    # Maritime base 5, "attack" is a moderate modifier -> 6.
    result = classify_event("Piracy attack on a cargo ship near the strait")
    assert result["primary_category"] == "maritime"
    assert result["severity"] == 6
    assert result["severity_modifiers"] == ["attack"]


def test_high_modifier_wins_over_moderate_no_stacking() -> None:
    # Both "killed" (high, +2) and "attack" (moderate, +1) appear; the
    # bumps must not stack: maritime base 5 + 2 = 7, not 8.
    result = classify_event("Piracy attack near the strait killed three seafarers")
    assert result["primary_category"] == "maritime"
    assert result["severity"] == 7
    assert "killed" in result["severity_modifiers"]
    assert "attack" in result["severity_modifiers"]


def test_primary_category_is_most_keyword_hits() -> None:
    # cyber matches 2 keywords ("ransomware", "hack"), maritime 1 ("port").
    result = classify_event("Ransomware hack hits port operators")
    assert result["primary_category"] == "cyber"
    assert result["severity"] == CATEGORIES["cyber"]["severity_base"]
    cats = [c["category"] for c in result["all_categories"]]
    assert cats[0] == "cyber"
    assert "maritime" in cats
    assert result["category_count"] == 2


def test_single_keyword_confidence_floor() -> None:
    # 1 hit -> min(1.0, 0.2 + 0.3) = 0.5
    result = classify_event("The satellite mission was a success")
    assert result["primary_category"] == "space"
    assert result["confidence"] == 0.5
    assert result["severity"] == CATEGORIES["space"]["severity_base"]


def test_classification_is_case_insensitive() -> None:
    upper = classify_event(_MILITARY_TEXT.upper())
    assert upper["primary_category"] == "military"
    assert upper["confidence"] == 0.9


async def test_fetch_classify_event_wraps_sync_classifier() -> None:
    result = await fetch_classify_event(None, _MILITARY_TEXT)
    direct = classify_event(_MILITARY_TEXT)
    assert result["primary_category"] == direct["primary_category"]
    assert result["severity"] == direct["severity"]
    assert result["confidence"] == direct["confidence"]


# ---------------------------------------------------------------------------
# Word-boundary regressions (Phase 24.5): substring matching fired categories
# and severity bumps from the inside of unrelated words — "port" in "report",
# "coup" in "couple". Verb stems must still match their inflections.
# ---------------------------------------------------------------------------


def test_keywords_inside_unrelated_words_do_not_fire() -> None:
    # Pre-fix this fired maritime ("port" in "report"), political ("coup" in
    # "couple"), social_unrest ("riot" in "patriotic"), and cyber ("apt" in
    # "captured").
    result = classify_event(
        "The report said a couple of patriotic films were captured on camera"
    )
    assert result["primary_category"] == "unclassified"
    assert result["all_categories"] == []
    assert result["severity"] == 0


def test_airstrike_is_military_not_labor_strike() -> None:
    result = classify_event("Airstrike hits depot near the frontline")
    assert result["primary_category"] == "military"
    cats = {c["category"] for c in result["all_categories"]}
    assert "social_unrest" not in cats
    # "strike" inside "airstrike" must not bump severity either.
    assert result["severity_modifiers"] == []
    assert result["severity"] == CATEGORIES["military"]["severity_base"]


def test_labor_strike_still_fires_social_unrest() -> None:
    result = classify_event("General strike and labor dispute paralyze the capital")
    assert result["primary_category"] == "social_unrest"


def test_verb_stems_still_match_inflected_forms() -> None:
    result = classify_event("SpaceX launched a rocket into orbit")
    assert result["primary_category"] == "space"
    matched = result["all_categories"][0]["keywords"]
    assert {"launch", "rocket", "orbit"} <= set(matched)


def test_protest_stem_matches_protesters() -> None:
    result = classify_event("A thousand protesters marched on parliament")
    assert result["primary_category"] == "political"


def test_deadline_does_not_bump_severity() -> None:
    # "deadline" contains "dead" (a high-severity modifier).
    result = classify_event("Sanctions deadline nears for the regime")
    assert result["primary_category"] == "political"
    assert result["severity_modifiers"] == []
    assert result["severity"] == CATEGORIES["political"]["severity_base"]


def test_deadly_still_bumps_severity() -> None:
    result = classify_event("Deadly explosion reported at the port")
    assert result["primary_category"] == "maritime"
    assert "dead" in result["severity_modifiers"]
    assert result["severity"] == CATEGORIES["maritime"]["severity_base"] + 2


def test_apt_matches_group_ids_as_words() -> None:
    result = classify_event("APT29 exploited a zero-day in the wild")
    assert result["primary_category"] == "cyber"
    assert "apt" in result["all_categories"][0]["keywords"]