Files
breakpilot-core/control-pipeline/services/applicability_engine.py
Benjamin Admin 38684dd903 feat(control-pipeline): add Assessment Layer to Applicability Engine
Adds confidence scoring, escalation detection, and reasoning to the
deterministic filter. All assessment is deterministic (no LLM).

Confidence scoring (0.0-1.0):
- +0.25 industry specified
- +0.15 company size specified
- +0.20-0.30 scope signals provided
- +0.15 controls found
- +0.15 no contradictions
- Capped at 0.75 for escalation cases

Escalation triggers:
- Contradictory signals (holds_client_funds without operates_payment_service)
- Ambiguous signals (provides_embedded_connectivity)
- Financial signals without explicit payment service declaration
- Incomplete profile (no industry, size, or signals)

Reasoning: template-based, includes active signals, control count,
scope-condition descriptions, and warnings.

Response now includes "assessment" field with confidence, escalation_flag,
escalation_reason, inferred_signals, reasoning, and warnings.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-23 20:36:11 +02:00

453 lines
16 KiB
Python

"""
Applicability Engine -- filters controls based on company profile + scope answers.
Two layers:
1. Deterministic Filter (Phase C2) — fast SQL + Python filtering
2. Assessment Layer — confidence scoring, escalation detection, reasoning
Filtering logic:
- Controls with NULL applicability fields are INCLUDED (apply to everyone).
- Controls with '["all"]' match all queries.
- Industry: control applies if its applicable_industries contains the requested
industry OR contains "all" OR is NULL.
- Company size: control applies if its applicable_company_size contains the
requested size OR contains "all" OR is NULL.
- Scope signals: control applies if it has NO scope_conditions, or the company
has at least one of the required signals (requires_any logic).
"""
from __future__ import annotations
import json
import logging
from dataclasses import dataclass, field, asdict
from typing import Any, Optional
from sqlalchemy import text
from db.session import SessionLocal
logger = logging.getLogger(__name__)
# Valid company sizes (ordered smallest to largest)
VALID_SIZES = ("micro", "small", "medium", "large", "enterprise")
# Signals that indicate potentially regulated financial activity
_FINANCIAL_SIGNALS = {"operates_payment_service", "holds_client_funds", "performs_kyc",
"monitors_transactions", "marketplace_model"}
# Signals that are ambiguous and may require legal review
_AMBIGUOUS_SIGNALS = {"provides_embedded_connectivity", "marketplace_model"}
# Contradictory signal pairs (if both present → escalate)
_CONTRADICTORY_PAIRS = [
("holds_client_funds", "operates_payment_service"), # holds funds but claims not a payment service
]
# Repo signals that suggest regulated activity
_REPO_SIGNAL_REGULATORY_MAP = {
"wallet_service": "financial",
"custody": "financial",
"kyc_provider": "financial",
"transaction_monitoring": "financial",
"payment_processing": "financial",
"stripe": "vendor_payment", # NOT own payment service
"paypal": "vendor_payment",
}
@dataclass
class AssessmentResult:
"""Assessment layer result — confidence, escalation, reasoning."""
confidence: float = 1.0
escalation_flag: bool = False
escalation_reason: Optional[str] = None
inferred_signals: list = field(default_factory=list)
reasoning: str = ""
warnings: list = field(default_factory=list)
def _parse_json_text(value: Any) -> Any:
"""Parse a TEXT column that stores JSON. Returns None if unparseable."""
if value is None:
return None
if isinstance(value, (list, dict)):
return value
if isinstance(value, str):
try:
return json.loads(value)
except (json.JSONDecodeError, ValueError):
return None
return None
def _matches_industry(applicable_industries_raw: Any, industry: str) -> bool:
"""Check if a control's applicable_industries matches the requested industry."""
industries = _parse_json_text(applicable_industries_raw)
if industries is None:
return True # NULL = applies to everyone
if not isinstance(industries, list):
return True # malformed = include
if "all" in industries:
return True
return industry in industries
def _matches_company_size(applicable_company_size_raw: Any, company_size: str) -> bool:
"""Check if a control's applicable_company_size matches the requested size."""
sizes = _parse_json_text(applicable_company_size_raw)
if sizes is None:
return True # NULL = applies to everyone
if not isinstance(sizes, list):
return True # malformed = include
if "all" in sizes:
return True
return company_size in sizes
def _matches_scope_signals(
scope_conditions_raw: Any, scope_signals: list[str]
) -> bool:
"""Check if a control's scope_conditions are satisfied by the given signals.
A control with scope_conditions = {"requires_any": ["uses_ai", "processes_health_data"]}
matches if the company has at least one of those signals.
A control with NULL or empty scope_conditions always matches.
"""
conditions = _parse_json_text(scope_conditions_raw)
if conditions is None:
return True # no conditions = applies to everyone
if not isinstance(conditions, dict):
return True # malformed = include
requires_any = conditions.get("requires_any", [])
if not requires_any:
return True # no required signals = applies to everyone
# Company must have at least one of the required signals
return bool(set(requires_any) & set(scope_signals))
def get_applicable_controls(
db,
industry: Optional[str] = None,
company_size: Optional[str] = None,
scope_signals: Optional[list[str]] = None,
limit: int = 100,
offset: int = 0,
) -> dict[str, Any]:
"""
Returns controls applicable to the given company profile.
Uses SQL pre-filtering with LIKE for performance, then Python post-filtering
for precise JSON matching (since columns are TEXT, not JSONB).
Args:
db: SQLAlchemy session
industry: e.g. "Telekommunikation", "Energie", "Gesundheitswesen"
company_size: e.g. "medium", "large", "enterprise"
scope_signals: e.g. ["uses_ai", "third_country_transfer"]
limit: max results to return (applied after filtering)
offset: pagination offset (applied after filtering)
Returns:
dict with total_applicable count, paginated controls, and breakdown stats
"""
if scope_signals is None:
scope_signals = []
# SQL pre-filter: broad match to reduce Python-side filtering
query = """
SELECT id, framework_id, control_id, title, objective, rationale,
scope, requirements, test_procedure, evidence,
severity, risk_score, implementation_effort,
evidence_confidence, open_anchors, release_state, tags,
license_rule, source_original_text, source_citation,
customer_visible, verification_method, category, evidence_type,
target_audience, generation_metadata, generation_strategy,
applicable_industries, applicable_company_size, scope_conditions,
parent_control_uuid, decomposition_method, pipeline_version,
created_at, updated_at
FROM canonical_controls
WHERE release_state NOT IN ('duplicate', 'deprecated', 'rejected')
"""
params: dict[str, Any] = {}
# SQL-level pre-filtering (broad, may include false positives)
if industry:
query += """ AND (applicable_industries IS NULL
OR applicable_industries LIKE '%"all"%'
OR applicable_industries LIKE '%' || :industry || '%')"""
params["industry"] = industry
if company_size:
query += """ AND (applicable_company_size IS NULL
OR applicable_company_size LIKE '%"all"%'
OR applicable_company_size LIKE '%' || :company_size || '%')"""
params["company_size"] = company_size
# For scope_signals we cannot do precise SQL filtering on requires_any,
# but we can at least exclude controls whose scope_conditions text
# does not contain any of the requested signals (if only 1 signal).
# With multiple signals we skip SQL pre-filter and do it in Python.
if scope_signals and len(scope_signals) == 1:
query += """ AND (scope_conditions IS NULL
OR scope_conditions LIKE '%' || :scope_sig || '%')"""
params["scope_sig"] = scope_signals[0]
query += " ORDER BY control_id"
rows = db.execute(text(query), params).fetchall()
# Python-level precise filtering
applicable = []
for r in rows:
if industry and not _matches_industry(r.applicable_industries, industry):
continue
if company_size and not _matches_company_size(
r.applicable_company_size, company_size
):
continue
if scope_signals and not _matches_scope_signals(
r.scope_conditions, scope_signals
):
continue
applicable.append(r)
total_applicable = len(applicable)
# Apply pagination
paginated = applicable[offset : offset + limit]
# Build domain breakdown
domain_counts: dict[str, int] = {}
for r in applicable:
domain = r.control_id.split("-")[0].upper() if r.control_id else "UNKNOWN"
domain_counts[domain] = domain_counts.get(domain, 0) + 1
# Build severity breakdown
severity_counts: dict[str, int] = {}
for r in applicable:
sev = r.severity or "unknown"
severity_counts[sev] = severity_counts.get(sev, 0) + 1
# Build industry breakdown (from matched controls)
industry_counts: dict[str, int] = {}
for r in applicable:
industries = _parse_json_text(r.applicable_industries)
if isinstance(industries, list):
for ind in industries:
industry_counts[ind] = industry_counts.get(ind, 0) + 1
else:
industry_counts["unclassified"] = (
industry_counts.get("unclassified", 0) + 1
)
# Assessment layer
assessment = _assess(
industry=industry,
company_size=company_size,
scope_signals=scope_signals,
total_applicable=total_applicable,
applicable_controls=applicable,
)
return {
"total_applicable": total_applicable,
"limit": limit,
"offset": offset,
"controls": [_row_to_control(r) for r in paginated],
"breakdown": {
"by_domain": domain_counts,
"by_severity": severity_counts,
"by_industry": industry_counts,
},
"assessment": asdict(assessment),
}
def _row_to_control(r) -> dict[str, Any]:
"""Convert a DB row to a control dict for API response."""
return {
"id": str(r.id),
"framework_id": str(r.framework_id),
"control_id": r.control_id,
"title": r.title,
"objective": r.objective,
"rationale": r.rationale,
"severity": r.severity,
"category": r.category,
"verification_method": r.verification_method,
"evidence_type": getattr(r, "evidence_type", None),
"target_audience": r.target_audience,
"applicable_industries": r.applicable_industries,
"applicable_company_size": r.applicable_company_size,
"scope_conditions": r.scope_conditions,
"release_state": r.release_state,
"control_id_domain": (
r.control_id.split("-")[0].upper() if r.control_id else None
),
"created_at": r.created_at.isoformat() if r.created_at else None,
"updated_at": r.updated_at.isoformat() if r.updated_at else None,
}
# =============================================================================
# Assessment Layer — Confidence, Escalation, Reasoning
# =============================================================================
def _assess(
industry: Optional[str],
company_size: Optional[str],
scope_signals: Optional[list[str]],
total_applicable: int,
applicable_controls: list,
) -> AssessmentResult:
"""Compute assessment result from filter inputs and outputs.
Deterministic scoring — no LLM needed.
"""
signals = scope_signals or []
result = AssessmentResult(inferred_signals=list(signals))
warnings = []
# --- Confidence scoring ---
score = 0.0
# Industry specified? (+0.25)
if industry:
score += 0.25
else:
warnings.append("Keine Branche angegeben — alle Controls werden angezeigt")
# Company size specified? (+0.15)
if company_size:
score += 0.15
else:
warnings.append("Keine Unternehmensgroesse angegeben")
# Scope signals provided? (+0.20 if any, +0.30 if >=3)
if len(signals) >= 3:
score += 0.30
elif len(signals) >= 1:
score += 0.20
else:
warnings.append("Keine Scope-Signale angegeben — Filterung nur nach Branche/Groesse")
# Controls found? (+0.15 if >5, +0.05 if 1-5)
if total_applicable > 5:
score += 0.15
elif total_applicable > 0:
score += 0.05
# No contradictions? (+0.15)
contradictions = _detect_contradictions(signals)
if not contradictions:
score += 0.15
else:
for c in contradictions:
warnings.append(f"Widerspruch: {c}")
result.confidence = round(min(score, 1.0), 2)
# --- Escalation detection ---
escalation_reasons = []
# Rule 1: Contradictory signals
if contradictions:
escalation_reasons.append(
f"Widersprüchliche Angaben: {'; '.join(contradictions)}"
)
# Rule 2: Ambiguous signals present
active_ambiguous = set(signals) & _AMBIGUOUS_SIGNALS
if active_ambiguous:
escalation_reasons.append(
f"Mehrdeutige Signale erfordern vertiefte Prüfung: {', '.join(sorted(active_ambiguous))}"
)
# Rule 3: Financial signals without explicit payment service declaration
active_financial = set(signals) & _FINANCIAL_SIGNALS
if active_financial and "operates_payment_service" not in signals:
if any(s in signals for s in ("holds_client_funds", "performs_kyc", "monitors_transactions")):
escalation_reasons.append(
"Finanznahe Signale ohne explizite Angabe zu Zahlungsdienst-Status — "
"regulatorische Einordnung (PSD2/ZAG) vertieft prüfen"
)
# Rule 4: Very few inputs → low confidence
if not industry and not company_size and not signals:
escalation_reasons.append(
"Unvollständiges Profil — keine Branche, Größe oder Scope-Signale angegeben"
)
if escalation_reasons:
result.escalation_flag = True
result.escalation_reason = " | ".join(escalation_reasons)
# Cap confidence for escalation cases
result.confidence = min(result.confidence, 0.75)
# --- Reasoning ---
reasoning_parts = []
if industry:
reasoning_parts.append(f"Branche: {industry}")
if company_size:
reasoning_parts.append(f"Unternehmensgröße: {company_size}")
if signals:
reasoning_parts.append(f"Aktive Scope-Signale: {', '.join(sorted(signals))}")
reasoning_parts.append(f"{total_applicable} Controls zugewiesen")
if total_applicable > 0:
# Collect unique source regulations from controls
sources = set()
for r in applicable_controls[:500]:
sc = _parse_json_text(getattr(r, "scope_conditions", None))
if isinstance(sc, dict) and sc.get("requires_any"):
for sig in sc["requires_any"]:
if sig in signals:
desc = sc.get("description", "")
if desc:
sources.add(desc)
if sources:
reasoning_parts.append(
f"Scope-bedingte Controls: {'; '.join(sorted(sources)[:5])}"
)
if warnings:
reasoning_parts.append(f"Hinweise: {'; '.join(warnings)}")
if result.escalation_flag:
reasoning_parts.append(f"ESKALATION: {result.escalation_reason}")
result.reasoning = ". ".join(reasoning_parts) + "."
result.warnings = warnings
return result
def _detect_contradictions(signals: list[str]) -> list[str]:
"""Detect contradictory signal pairs."""
contradictions = []
signal_set = set(signals)
# holds_client_funds but NOT operates_payment_service
if "holds_client_funds" in signal_set and "operates_payment_service" not in signal_set:
contradictions.append(
"holds_client_funds=true aber operates_payment_service nicht gesetzt — "
"unklar ob regulierter Zahlungsdienst"
)
# performs_kyc but NOT operates_payment_service and NOT marketplace_model
if ("performs_kyc" in signal_set
and "operates_payment_service" not in signal_set
and "marketplace_model" not in signal_set):
contradictions.append(
"performs_kyc=true ohne Payment- oder Marktplatz-Kontext — "
"regulatorische Grundlage für KYC unklar"
)
return contradictions