rrt-advocate 0.1.1

PROTOTYPE / not medical advice — Crisis Detection Engine (CDE): Python port of the RRT Advocate 3-layer, local-first crisis detection & assessment pipeline. NeuroLift HAIEF Solidarity Framework.

⚠️ PROTOTYPE — NOT A SAFETY SYSTEM

This is an experimental crisis-detection library with stubbed/placeholder intervention layers. It is NOT medical advice, NOT a crisis service, and performs no real-time monitoring. It can miss real crisis signals (known detection/recall gaps) — do not rely on it as a safety net or as the sole safety mechanism.

If you or someone else needs help now: in the US, call or text 988 (Suicide & Crisis Lifeline) or chat 988lifeline.org; in an emergency call 911. Outside the US: findahelpline.com.

Provided AS-IS, without warranty.

rrt-advocate (Python)

Python port of the RRT Advocate Crisis Detection Engine (CDE) — a 3-layer, local-first crisis detection and assessment pipeline from the NeuroLift HAIEF Solidarity Framework.

This is a faithful port of the published npm package @neurolift-technologies/rrt-advocate (the TypeScript CDE, which is the source of truth for this port). The public API mirrors the npm package's surface in snake_case.

Safety-critical. This port preserves every layer weight, threshold, and confidence formula from the TypeScript source. The one intentional behavioral divergence (apostrophe fail-open in Layer 1) is documented in KNOWN_LIMITATIONS.md. This package performs detection and assessment only — it does not generate responses, interventions, or persona-blended output.

Install

pip install rrt-advocate

vaderSentiment is an optional dependency. If installed, Layer 2 uses it for polarity scoring; otherwise it falls back to a deterministic built-in heuristic.

Usage

from rrt_advocate import CrisisEngine, CrisisLevel

engine = CrisisEngine("user-123")

assessment = engine.assess("I can't cope, everything is too much")

assessment.crisis_level             # CrisisLevel.GREEN | YELLOW | ORANGE | RED | BLACK
assessment.user_safety_score        # 1.0 (safe) → 0.05 (immediate danger)
assessment.confidence_score         # aggregate crisis confidence, 0.0–1.0
assessment.recommended_interventions
assessment.primary_indicators

if assessment.crisis_level is not CrisisLevel.GREEN:
    ...  # route to appropriate support

The pipeline

Layer	Module	Weight	Signal
1	KeywordLayer	0.45	Semantic-field keyword matching (self-harm forces 1.0)
2	SentimentLayer	0.35	Polarity & declining-trend detection over a sliding window
3	BehavioralLayer	0.20	Latency, message complexity, and looping (Jaccard)

CrisisDetector aggregates the three layers into CrisisIndicators; CrisisAssessor maps those to a CrisisAssessment using the bundled config/crisis_thresholds.yaml. CrisisEngine wires the two together.

Lower-level building blocks are exported too:

from rrt_advocate import (
    CrisisDetector,
    CrisisAssessor,
    KeywordLayer,
    SentimentLayer,
    BehavioralLayer,
)

Privacy

Everything runs locally. The behavioral layer stores only message metadata (timing, length, and HMAC-hashed word tokens) — never raw message content.

License

Apache-2.0 © NeuroLift Technologies, LLC