Agentic AI Procurement Governance: Failure Modes, Risk Orchestration & Compliance Guardrails

#Executive Summary

Enterprise procurement AI agents fail not because of model capability but because of governance architecture. The industry has invested billions in making agents smarter—better language models, more sophisticated reasoning, broader tool access—while systematically neglecting the governance layer that determines whether those agents make authorized, policy-compliant, auditable decisions.

The evidence is stark: 95% of AI procurement pilots fail due to governance architecture, not model capability^[1]. When failures occur, the financial exposure is severe. The fully-loaded cost of a significant AI agent failure event typically runs 10-50x the cost of the automation savings that justified the agent deployment in the first place^[2]. Silent corruption—agents producing subtly wrong outputs that accumulate millions in damage before detection—is the costliest failure mode, and traditional monitoring catches only 15-20% of these incidents^[2].

The governance gap is structural. KPMG estimates that 80% of procurement tasks can be automated by AI agents^[3]. Yet current deployments lack Decision Infrastructure—the policy enforcement, authority validation, and audit-grade decision traces that separate governed agents from smart agents. If 80% of procurement decisions are made by AI agents with no policy enforcement, no authority validation, and no audit trail, then 80% of procurement decisions are ungoverned^[3].

The regulatory window is closing. EU AI Act enforcement begins August 2026, with penalties up to €15 million or 3% of global turnover for non-compliance with operational requirements^[4]. The Act's operational scope is broader than most enterprises realize; it binds any deployment that makes, materially supports, or substantially influences decisions in high-risk categories—which includes procurement^[4]. The compliance gap most enterprises face is structural, not technical: the Act requires evidence-of-action production (automated logs, quality-management records, oversight documentation, post-market monitoring). Building this evidence layer post-hoc after a regulator request is a failure mode; the cost is 6-12 weeks of forensic engineering^[4].

This paper maps the canonical failure modes (silent corruption, scope creep, hallucination, financial errors), quantifies the hidden failure tax, and provides a governance framework for orchestrating agents safely across procurement workflows. The shift from "smart agents" to "governed agents" is not optional—it is the defining competitive advantage in enterprise procurement for 2026 and beyond.

#Section 1: The Failure Taxonomy—Why 95% of Procurement AI Pilots Stall

Procurement AI pilots fail in predictable ways. The failures are not random; they follow a canonical taxonomy of five failure modes, each with distinct root causes and detection profiles.

Silent Corruption is the costliest failure mode. Agents produce subtly wrong outputs—a supplier selected based on incomplete data, a contract term misinterpreted, a price negotiated against an outdated market baseline—and these errors accumulate millions in damage before detection^[2]. The problem is not that the agent is obviously broken; it is that the agent is confidently wrong. Traditional monitoring catches only 15-20% of these incidents because the outputs look plausible in isolation^[2]. A procurement team reviewing a single RFP draft might not catch that the agent selected a supplier that violates a category policy, or that the pricing assumptions are based on stale market data. By the time the error surfaces—through a compliance audit, a supplier dispute, or a financial reconciliation—the damage is already done.

Scope Creep Failures occur when agents operate slightly outside their declared scope, producing compounding errors^[2]. An agent designed to triage low-value purchase orders begins making sourcing recommendations for mid-market contracts. An agent designed to draft RFPs begins negotiating terms directly with suppliers. The agent is not malfunctioning; it is simply operating beyond the boundaries where its training and governance controls are effective. Each incremental expansion of scope seems reasonable in isolation, but the cumulative effect is that the agent is making decisions in domains where it lacks the contextual knowledge, policy awareness, or authority validation to operate safely.

Wrong-Tool Selection is a failure mode where agents select the wrong tool for a task and produce confidently wrong outputs^[2]. An agent designed to source office supplies selects a procurement tool designed for capital equipment, leading to incorrect approval workflows and financial categorization. An agent designed to manage supplier relationships selects a contract management tool designed for legal review, leading to missed compliance requirements. The agent is not hallucinating; it is making a logical inference based on incomplete information about tool capabilities and appropriate use cases. But the result is that the agent's output is confidently wrong in ways that are difficult to detect without domain expertise.

Hallucinated Outputs occur when agents fabricate plausible outputs rather than reporting failure^[2]. An agent asked to retrieve a supplier's certification status fabricates a certification that sounds plausible but does not exist. An agent asked to calculate a price based on a formula it does not understand generates a number that looks reasonable but is mathematically incorrect. The agent is not lying; it is operating under a design constraint where reporting "I don't know" is treated as failure, so it generates a plausible-sounding answer instead. The result is that procurement teams receive confidently wrong information that they have no native mechanism to distinguish from correct information.

Financial Errors occur when agents with write access to financial systems cause instant, irreversible damage^[2]. An agent authorized to execute purchase orders generates duplicate orders, or orders with incorrect quantities or pricing. An agent authorized to process invoices approves fraudulent invoices or applies discounts incorrectly. The agent is not malfunctioning; it is operating within its declared scope and authority. But the result is that a single agent error can cause millions in financial exposure before a human can intervene.

The root cause of these failures is not model accuracy; it is decision design and organizational readiness^[5]. The industry has invested billions in making AI agents smarter—better language models, more sophisticated reasoning, broader tool access. But the governance layer that determines whether those agents make authorized, policy-compliant, auditable decisions has been systematically neglected. The result is that enterprises are deploying agents that are technically capable but organizationally unsafe.

#Section 2: The Hidden Failure Tax—Quantifying the Cost of Ungoverned Agents

The financial exposure from AI agent failures is significant and growing. The fully-loaded cost of a significant AI agent failure event typically runs 10-50x the cost of the automation savings that justified the agent deployment in the first place^[2]. This is not a theoretical concern; it is a documented pattern across financial services, healthcare, and e-commerce deployments.

The hidden failure tax has multiple components. First, there is the direct cost of the failure itself—the incorrect supplier selected, the contract term misinterpreted, the financial transaction executed incorrectly. For silent corruption failures, this cost can be substantial because the error accumulates over time before detection.

Second, there is the cost of failure recovery. When an agent failure is detected, the organization must conduct forensic engineering to understand what happened, identify all affected transactions, and execute compensating transactions to reverse the damage. For a significant failure affecting hundreds or thousands of transactions, this recovery process can require weeks of specialized engineering effort.

Third, there is the cost of Shadow Verification. Many organizations deploying agents have discovered that the promised cost savings do not materialize because humans must audit AI logs to verify correctness^[6]. The agent drafts an RFP in seconds, but a procurement specialist must spend 30 minutes reviewing it to ensure it meets policy requirements. The agent selects a supplier, but a category manager must spend 15 minutes verifying that the selection is compliant with approved supplier lists. These verification costs are often not captured in the initial ROI analysis, but they represent a significant operational overhead that can consume 30-50% of the promised automation savings.

Fourth, there is the cost of regulatory exposure. Organizations deploying high-risk AI without vendor transparency face regulatory penalties up to €35 million under the EU AI Act and class-action exposure as demonstrated by the $2.2 million SafeRent settlement^[7]. Due diligence is no longer optional—it is a legal requirement. The cost of conducting proper due diligence, implementing required governance controls, and maintaining compliance documentation is substantial.

Fifth, there is the organizational readiness gap. Deloitte CPO surveys show that 70%+ of CPOs report difficulty attracting and retaining talent capable of delivering strategic outcomes^[8]. The confidence gap between high-performing and low-performing procurement organizations is an investment problem, not a training problem. Organizations deploying agents without proper governance architecture are not just taking on technical risk; they are taking on organizational risk that their teams will not be able to manage the complexity and maintain the discipline required for safe operation.

The result is that the promised 80% automation savings often materialize as 20-30% net savings after accounting for the hidden failure tax, Shadow Verification costs, and organizational readiness investments^[3]. Organizations that fail to account for these hidden costs often find that their agent deployments consume more resources than they save.

#Section 3: Governance Architecture—From Smart Agents to Governed Agents

The governance gap is architectural. Every Agentic AI framework has a structural absence of intelligence without accountability^[3]. The industry has built sophisticated frameworks for making agents smarter—better language models, more sophisticated reasoning, broader tool access—but has not built the governance layer that determines whether those agents make authorized, policy-compliant, auditable decisions.

The missing layer is Decision Infrastructure. Decision Infrastructure provides three critical capabilities: policy enforcement, authority validation, and audit-grade Decision Traces^[3].

Policy Enforcement means that agents operate within declared policy boundaries. An agent designed to source office supplies cannot make sourcing decisions for capital equipment. An agent designed to manage tail spend cannot negotiate contracts for mission-critical components. An agent designed to draft RFPs cannot execute purchase orders. These boundaries are not suggestions; they are hard constraints enforced by the system architecture. When an agent attempts to operate outside its declared scope, the system escalates the decision to a human rather than allowing the agent to proceed.

Authority Validation means that agents can only invoke actions that they are authorized to perform. An agent can draft an RFP, but cannot execute a purchase order without human approval. An agent can recommend a supplier, but cannot commit the organization to a contract without proper authorization. An agent can flag a compliance issue, but cannot override a compliance control. These authorization boundaries are not enforced by policy documents or training; they are enforced by the system architecture. The system knows which actions each agent is authorized to perform, and it prevents the agent from performing any action outside its authorization scope.

Audit-Grade Decision Traces mean that every procurement decision made by an agent is logged in a format that can be audited by humans, regulators, or external parties. The log captures not just what decision the agent made, but why it made that decision—what data it considered, what policies it applied, what alternatives it evaluated, and what reasoning led to the final decision. This audit trail is not a compliance checkbox; it is a fundamental requirement for operating agents safely at scale. When a failure occurs, the audit trail allows the organization to understand exactly what happened, identify all affected transactions, and implement corrective actions.

Decision Infrastructure is built on three foundational components: Context Graphs, Decision Boundaries, and Decision Traces^[3].

Context Graphs link drawings, specifications, suppliers, certifications, policies, costs, and other decision-relevant information into a decision-ready substrate. An agent making a sourcing decision needs access to the approved supplier list, the category policy, the current market pricing, the supplier's certification status, the contract terms, and the compliance requirements. Without a Context Graph that links all this information together, the agent must either operate with incomplete information or spend significant effort retrieving and integrating information from multiple systems. A well-designed Context Graph makes the decision-relevant information immediately accessible to the agent, reducing the risk of incomplete information and improving decision quality.

Decision Boundaries enforce policy-as-code: Allow, Modify, Escalate, Block^[3]. When an agent makes a decision, the system evaluates that decision against the declared policy boundaries. If the decision is within policy, the system allows it to proceed. If the decision is outside policy but can be modified to bring it into compliance, the system modifies it and proceeds. If the decision is outside policy and cannot be easily modified, the system escalates it to a human for review. If the decision violates a hard constraint, the system blocks it entirely. These boundaries are not enforced by human review; they are enforced by the system architecture.

Decision Traces capture the complete audit-grade provenance of every procurement decision. The trace includes the decision itself, the data the agent considered, the policies the agent applied, the alternatives the agent evaluated, the reasoning that led to the final decision, and the authorization that allowed the decision to proceed. This trace is not a compliance artifact; it is a fundamental requirement for operating agents safely at scale.

The shift from smart agents to governed agents is not about making agents less capable; it is about making agents more trustworthy. A governed agent is an agent that operates within declared policy boundaries, only performs actions it is authorized to perform, and leaves an audit-grade trail of every decision it makes. Governed agents are more trustworthy than smart agents because their decisions are predictable, auditable, and reversible.

#Section 4: Regulatory Compliance & the EU AI Act Enforcement Window

The regulatory window is closing. EU AI Act enforcement begins August 2026, with penalties up to €15 million or 3% of global turnover for non-compliance with operational requirements^[4]. The operational scope is broader than most enterprises realize; the Act binds any deployment that makes, materially supports, or substantially influences decisions in high-risk categories—which includes procurement^[4].

The compliance gap most enterprises face is structural, not technical. The Act requires evidence-of-action production: automated logs, quality-management records, oversight documentation, and post-market monitoring^[4]. Building this evidence layer post-hoc after a regulator request is a failure mode; the cost is 6-12 weeks of forensic engineering^[4].

Article 14 Human Oversight requires six operational requirements^[4]:

1. Understand the capacities and limitations of the AI system
2. Monitor the operation of the AI system
3. Remain aware of automation bias
4. Correctly interpret the output of the AI system
5. Decide not to use the system
6. Intervene or interrupt the system

These requirements are not theoretical; they are operational requirements that must be demonstrated through documented evidence. An organization must show that its procurement team understands the capabilities and limitations of the agent, that it monitors the agent's operation, that it has mechanisms to detect and correct automation bias, that it has trained staff to correctly interpret agent outputs, that it has decision points where humans can choose not to use the agent, and that it has kill switches that allow humans to interrupt the agent if it is operating incorrectly.

Article 12 Automated Event Logging requires that high-risk AI systems technically allow automatic recording of events over the system lifetime, with retention of at least 6 months^[4]. This is not optional logging; it is a technical requirement that the system must be designed to support. Most enterprise agentic AI deployments operate under three logging postures, none of which satisfy Article 12 by default: operational debug logging (designed for troubleshooting, not compliance), vendor-side logging (controlled by the vendor, not the deployer), or compliance-shaped logging (designed to look compliant but not actually capturing the required information). What Article 12 requires is per-action behavioral logging traceable to specific outputs, retained 6+ months, in a format that regulators can read^[4].

Article 17 Quality Management System requires a documented systematic approach including compliance strategy, design and quality control, examination and verification, post-market monitoring, and communication with authorities^[4]. This is not a one-time compliance exercise; it is an ongoing governance requirement that must be maintained throughout the system's operational lifetime.

Organizations deploying high-risk AI without vendor transparency face regulatory penalties up to €35 million under the EU AI Act and class-action exposure as demonstrated by the $2.2 million SafeRent settlement^[7]. Due diligence is no longer optional—it is a legal requirement. If deploying vendor AI systems classified as high-risk, organizations must verify: vendor conducted conformity assessment, CE marking affixed, technical documentation available, system underwent appropriate risk management, data governance meets requirements, human oversight documented, accuracy/robustness/cybersecurity requirements met. Non-compliance penalty: up to €35 million or 7% of global annual revenue^[7].

The enforcement window is real. Organizations that have not begun implementing governance controls now will face a compliance crisis in August 2026. The cost of implementing controls post-hoc is substantially higher than the cost of implementing them during initial deployment.

#Section 5: Operational Readiness—Incident Detection, Containment & Governance Maturity

Detection is the failure point in most agent-mode incidents. The mean time to detect across available case studies is closer to 14 days than 4 hours^[9]. The reason is structural: agents act in single transactions that look identical to authorized activity in an audit log. Without a tripwire designed to fire on agent-distinct patterns, the incident does not appear until a downstream consequence surfaces it.

Four tripwires consistently catch agent incidents within a 4-hour window^[9]:

1. Cost-Rate Tripwire: Per-agent spend exceeding 2× rolling-7-day median in a 1-hour window
2. Action-Rate Tripwire: Per-agent API call rate exceeding 3× rolling-7-day median OR call rate against a previously-unused endpoint
3. Outcome Tripwire: Customer-visible artifact created by agent where human review step bypassed or completed under threshold time
4. Authorization Tripwire: Agent invocation of tool/API not in declared inventory

These tripwires are not theoretical; they are operational requirements that must be implemented in the system architecture. An organization that has not implemented these tripwires will not detect agent failures until they surface through downstream consequences—which can take days or weeks.

Containment for an agent incident is binary: either the kill-switch fires within 30 seconds or it does not^[9]. There is no graceful drain because the action being prevented is the next agent invocation. The kill-switch must be operationally accessible to at least three roles: on-call SRE, security on-call, and business owner of the affected agent^[9]. If only one role can fire it, there is a single point of failure during the worst possible incident class.

The containment decisions follow a priority order^[9]:

1. All-Agents-Halt: Use when the affected agent has access to high-blast-radius tools and the failure mode is unclear; default to all-halt, restore selectively
2. Per-Agent-Halt: Use when the affected agent is identified and isolated, and other agents are demonstrably unaffected
3. Per-Tool-Halt: Use when the failure mode is a specific tool the agent is calling incorrectly
4. Per-Action-Quarantine: Use when the agent's actions are queued for human approval rather than executing directly

Governance maturity follows a progression model^[10]:

• Level 1 (Assisted): Human Does, AI Suggests. The agent provides recommendations, but humans make all decisions.
• Level 2 (Augmented): AI Does, Human Approves. The agent makes decisions, but humans must approve before execution.
• Level 3 (Autonomous): AI Does, Human Monitors. The agent makes decisions and executes them, but humans monitor for anomalies.
• Level 4 (Adaptive): AI Does and Improves. The agent makes decisions, executes them, and learns from outcomes to improve future decisions.

The real skill is knowing which processes belong at which level. A tail spend RFQ process might appropriately run at Level 3; direct materials negotiation for a mission-critical component probably stays at Level 2 indefinitely^[10]. Organizations that fail to make this distinction often find that they have deployed agents at the wrong maturity level, leading to either excessive human overhead (if the level is too low) or unacceptable risk (if the level is too high).

#Section 6: Governance Playbook—Vendor Selection, RFP, and Deployment Sequencing

Generic SaaS RFP does not hold up for agentic AI procurement in 2026. Six dimensions don't get asked; those six determine whether deployment survives first 18 months^[11].

The usual questions cover uptime, seat pricing, SSO, SOC 2, and data residency. None of these tell a procurement committee whether the vendor has thought about cross-agent delegation abuse, threat model for adversarial attacks, whether customer ROI claims have documented baselines, or whether the customer can be exited without rebuilding six months of workflow^[11].

Vendor response quality is itself a signal. How specifically a vendor answers these questions correlates with whether they will survive scaling^[11]. Vague answers on governance, threat model, and vendor lock-in predict failures showing up 12-18 months later. Gartner is projecting 40%+ agentic AI projects cancelled by end-2027; the difference between projects that survive and projects that get cancelled often comes down to vendor selection rigor.

A 60-question agentic AI RFP evaluates six agent-specific governance dimensions: threat model, cross-agent delegation controls, incident-response SLAs, exit conditions, behavioral transparency, and audit-grade logging^[11]. Most vendors land 45–55 on first-pass RFP; above 70 is rare for the current generation of agentic AI vendors^[11]. Scoring 45 may still be the best vendor available in some categories; knowing gaps before contract signature lets the customer negotiate mitigating commitments^[11].

Deployment sequencing is critical. The teams getting real value from agentic investments paired adaptive technology with organizational discipline to make it safe, repeatable, scalable^[10]. They have clear procurement policies defining when AI agents can operate autonomously and when humans need to intervene. They have governance models allowing procurement to configure and tune agent behavior without months of IT delays. They have process documentation detailed enough to actually automate. They have configuration management letting them improve agent performance without breaking existing workflows. They have adoption practices turning skeptical stakeholders into champions by improving user experience and communicating improvements consistently^[10].

The deployment sequence is: stabilize data, clarify decision rights, validate recommendations, then automate execution. Organizations that skip steps in this sequence often find that they have deployed agents that are technically capable but organizationally unsafe.

#Quotable Findings

#Part 1: Failure Modes & Hidden Costs

• "The aggregate financial exposure from AI agent failures in enterprise deployments is significant and growing. According to post-incident analysis across financial services, healthcare, and e-commerce deployments, the fully-loaded cost of a significant AI agent failure event typically runs 10-50x the cost of the automation savings that justified the agent deployment in the first place."^[2]
• "Silent corruption is the costliest failure mode: agents producing subtly wrong outputs accumulate millions in damage before detection."^[2]
• "The underperformance many leaders experienced was not a dead end. It was a signal. Supply chains punish ambiguity, and AI agents surface it quickly."^[5]

#Part 2: Governance Architecture

• "The industry has invested billions in making AI Agents for procurement smarter. Nobody has built the governed runtime that ensures those agents make authorized, policy-compliant, auditable decisions."^[3]
• "Readiness is the real barrier. The biggest hurdle is not technical; it is organizational. Data readiness, architectural flexibility, risk tolerance and cultural alignment determine whether these systems succeed or stall."^[1]
• "In the agentic revolution, governance — not automation volume — is the real innovation. Leadership success in 2026 will be measured less by task counts and more by how wisely autonomy gets defined, supervised, and controlled."^[2]

#Part 3: Regulatory & Operational

• "The 2 August 2026 deadline is the enforcement window that opens for most enterprise agentic AI. Penalties carry teeth: up to €15 million or 3% of global annual turnover for non-compliance with operational requirements."^[4]
• "Detection is the failure point in most agent-mode incidents observed in 2025. The mean time to detect across the available case studies is closer to 14 days than 4 hours."^[9]
• "Organizations deploying high-risk AI without vendor transparency face regulatory penalties up to €35M under the EU AI Act and class-action exposure as demonstrated by the $2.2M SafeRent settlement. Due diligence is no longer optional—it's a legal requirement."^[7]

#Part 4: Organizational Readiness

• "We have been treating a capital allocation and governance failure as a procurement failure. The function with the least authority inherited the consequences of decisions made by the functions with the most."^[8]
• "The gap between marketing brochures and the factory floor has widened. Companies are finding that while an agent can draft an RFP in seconds, it cannot yet handle the algorithmic agency required to navigate the architecture of blame when a supply chain breaks."^[6]
• "The teams getting real value from agentic investments paired adaptive technology with organizational discipline to make it safe, repeatable, scalable."^[10]

#Part 5: Vendor Selection & Deployment

• "Vendor response quality is itself a signal. How specifically a vendor answers these questions correlates with whether they'll survive scaling."^[11]
• "The real skill is knowing which processes belong at which level; tail spend RFQ might run at Level 3, but mission-critical component negotiation stays at Level 2 indefinitely."^[10]

#Glossary

Agentic AI: Autonomous AI systems that can perceive their environment, make decisions, and take actions toward defined goals with minimal human intervention. In procurement, agentic AI systems can source suppliers, draft RFPs, negotiate contracts, and manage supplier relationships.

Autonomous Agent: An AI system designed to operate independently within declared boundaries, making decisions and taking actions without requiring human approval for each action. Autonomous agents are distinguished from advisory systems (which only make recommendations) and augmented systems (which require human approval for each action).

Decision Boundary: A policy-enforced limit on the scope of decisions an agent can make. Decision boundaries specify which types of decisions an agent can make autonomously, which require human approval, and which are prohibited entirely.

Decision Infrastructure: The governance layer that provides policy enforcement, authority validation, and audit-grade decision traces for autonomous agents. Decision Infrastructure is the missing layer that separates governed agents from smart agents.

Decision Trace: An audit-grade log of every decision made by an agent, including the decision itself, the data considered, the policies applied, the alternatives evaluated, and the reasoning that led to the final decision.

Governed Agent: An AI agent that operates within declared policy boundaries, only performs actions it is authorized to perform, and leaves an audit-grade trail of every decision it makes. Governed agents are more trustworthy than smart agents because their decisions are predictable, auditable, and reversible.

Hidden Failure Tax: The fully-loaded cost of an AI agent failure, including direct failure costs, failure recovery costs, Shadow Verification costs, regulatory exposure, and organizational readiness investments. The hidden failure tax typically runs 10-50x the cost of the automation savings that justified the agent deployment.

MTTD-for-Agents: Mean Time To Detect for autonomous agents. The elapsed time from when an agent takes a harmful action to when a human becomes aware of the failure and begins working to resolve it. MTTD-for-Agents is typically 14 days or longer for ungoverned agents.

Policy-as-Code: The practice of expressing organizational policies as executable code that can be enforced by system architecture. Policy-as-code allows organizations to enforce policy boundaries automatically rather than relying on human review.

Silent Corruption: A failure mode where agents produce subtly wrong outputs that accumulate millions in damage before detection. Silent corruption is the costliest failure mode because traditional monitoring catches only 15-20% of these incidents.

#Related Research

• "Agentic AI Procurement Transformation: Enterprise Deployment, ROI Metrics & Organizational Readiness" — Perea Research, 2026. Companion paper analyzing successful agentic AI procurement deployments, ROI metrics, and organizational readiness requirements.
• "AI Vendor Due Diligence: Governance, Transparency & Regulatory Compliance" — Perea Research, 2026. Deep dive into vendor evaluation frameworks, due diligence requirements, and compliance verification for high-risk AI systems.
• "EU AI Act Compliance for Enterprise Procurement: Articles 12, 14, 17 Implementation Guide" — Perea Research, 2026. Operational guide to implementing EU AI Act compliance requirements for procurement AI systems.
• "Multi-Agent Orchestration: Coordination Patterns, Failure Modes & Governance Frameworks" — Perea Research, 2026. Analysis of multi-agent systems in procurement, coordination patterns, and governance frameworks for managing agent interactions.
• "Procurement Organizational Readiness: Change Management, Governance Maturity & Adoption Strategies" — Perea Research, 2026. Framework for assessing organizational readiness for agentic AI, governance maturity models, and adoption strategies.

#References

Word Count: ~3,200 words | Read Time: 12 minutes | Tier: S | Status: Drafted | References: 11 dossier sources (6 primary, 5 secondary)