🛡️ SentinelGov: Procedural Intelligence & Enforcement Node

Documentation hardened for governance review.

SentinelGov at a Glance (Control Plane Summary)

SentinelGov is a pre-disbursement enforcement node for government procurement. It intercepts suspicious procurements before funds are released, places them on administrative hold, and requires authorized human clearance — leaving a full, immutable audit trail.

What Problem It Stops

• Payments cleared despite non-L1 awards
• Inflated bid spreads masked as valid procurement
• Over-budget tenders approved without justification
• Risks discovered only after funds are disbursed

What SentinelGov Does

• Analyzes tenders, not just invoices
• Scores procurement risk before disbursement
• Automatically places high-risk transactions in ON_HOLD
• Requires role-authorized human release
• Logs every action with cryptographic integrity

What SentinelGov Never Does

• Never cancels payments autonomously
• Never labels actions as "fraud"
• Never overrides treasury authority
• Never releases funds without clearance

Why This Is Safe for Government Use

• Enforcement logic anchored to GFR-style rules
• Statistical checks only when data is sufficient
• Deterministic, explainable scoring
• Fully reversible decisions
• Complete audit visibility

---

🚀 1. The Core Philosophy

Traditional procurement monitoring systems are reactive. SentinelGov adopts an "Administrative Hold" approach:

• Graph Reasoning: Understands the relationship between specific Tenders, Bids, Vendors, and resulting Transactions.
• Economic Integrity: Enforces strict GFR-2017 compliant rules mixed with statistical support layers.
• Secure Handling: Every action is cryptographically tracked to ensure an immutable paper trail of who suppressed or sets transaction state to ON_HOLD, preventing authorization for disbursement.

---

🛡️ 2. Operational Guarantees

• Investigators can place any high-risk transaction into ON_HOLD with a single action, introducing an administrative hold prior to release.
• Finance Officers can release funds only after investigative clearance and mandatory justification.
• Public Users can verify sanctioned procurements via immutable, hash-verified records.
• All actions are logged with actor identity, timestamp, and integrity hash.

---

🛰️ 3. Key High-Level Features

A. Tender-Centric Intelligence

Instead of monitoring single invoices, SentinelGov analyzes the Economic Root:

• Rank Violation: Automatically flags awards to Rank 2, 3, or higher bidders.
• Spread Detection: Measures the "Bid Spread" (Winning Price vs Lowest Price). Spreads >15% trigger critical reviews.
• Z-Score Mapping: Statistically compares the winning bid against the cluster of all bids to find outliers.

B. "Secured Layer" Governance

The system includes a dedicated enforcement layer:

• Transaction Hold: One-click escalation to ON_HOLD state. This sets transaction state to ON_HOLD, preventing authorization for disbursement in the backend database.
• Audit-Safe Explanations: The system translates complex math into non-accusatory, auditor-safe language (e.g., "Deviation from competitive norms").

---

📜 4. Compliance & Ethics

Built for the Hack4Delhi Governance Track, this system prioritizes the "Instrument, Not Advisor" principle. All system-generated text includes mandatory disclaimers, and human oversight is required for final resolution.

Explicit Non-Goals

• Real-time bank integration
• Autonomous sanctioning
• Criminal attribution or accusation
• Replacement of existing treasury systems

---

🔧 5. Quick Start Guide

Installation & Run

1. Backend: cd backend && python seed_tenders.py && python main.py
2. Frontend: cd frontend && npm install && npm run dev

---

Appendix: Technical & Audit Detail (For Reviewers)

---

🏛️ A1. Technical Architecture

SentinelGov operates as a state-driven enforcement system, not a notification engine. No transaction can reach disbursement without passing through this state machine.

graph TD
    User[Investigator] -->|Enforcement| UI[React Frontend]
    UI -->|API Calls| API[FastAPI Backend]
    
    subgraph "Intelligence Core"
        API -->|Ingest| Engine[AnomalyEngine (detection.py)]
        Engine -->|Rules & ML Support| Model[Scoring Logic]
        Engine -->|Write| DB[(SQLite Database)]
    end

Loading

Core Components

• Backend: FastAPI (Python), SQLAlchemy, Scikit-Learn (ML-assisted behavioral signals).
• Frontend: React + Vite + Tailwind CSS, Zustand.
• Logic: Pure scoring logic in detection.py with statistical support layer.

---

📖 A2. Detailed User Persona Flows (SOPs)

Investigator Flow (Audit & Hold)

1. Identification: Dashboard signals a spike in "Risk Exposure".
2. Deep Dive: Forensic signals show "Rank 3 Awarded".
3. Action: Click HOLD. This sets transaction state to ON_HOLD, preventing authorization for disbursement.

Finance Flow (Justified Release)

1. Review: Locate a transaction that has been "Cleared".
2. Release: Click Confirm Disbursement with mandatory justification. This resolves administrative holds.

---

📑 A3. API & Data Mapping

• GET /system/status: KPI Single Source of Truth.
• POST /alerts/{id}/hold: Introduces an administrative hold prior to release.
• POST /transactions/{id}/release: Treasury-level disbursement clearance.

SentinelGov: Watching the Wealth of the Nation.