Procurement Protocol Specification The Calvin Convention for Interrogable Systems

Procurement Protocol Specification: The Calvin Convention for Interrogable Systems

1. Introduction: From Post-Hoc Explanation to Active Interrogation

The Agency shall strictly enforce the "Clarke Threshold" as a mandatory technical boundary. This threshold defines the critical failure point where technology becomes so opaque that it is treated as an unchallengeable oracle—magic substituted for policy. To protect institutional agency, this Protocol mandates a shift from "Post-Hoc Explanation" (narrative storytelling after a decision) to "Active Interrogation" (contestable reasoning before or during execution).The Calvin Convention is a pre-action, hierarchical, and non-negotiable standard for AI governance. It prohibits the use of "High Fidelity" models of obedience that favor mechanical speed over human discernment. These definitions serve as the bedrock for the following procurement requirements; any system failing to meet these interrogability standards shall be deemed non-responsive.

Interrogability vs. Explanation: Key Differentiators

Feature,Post-Hoc Explanation (The Industry Default),Active Interrogation (The Calvin Standard)
Timing,Retrospective; occurs after the harm has been processed and normalized.,Pre-action or real-time; occurs before authority is exercised.
Basis of Authority,"Derived from proprietary ""black box"" logic or opaque vendor validation.","Derived from legible, contestable, and shared reasoning chains."
Contestability,Ceremonial; appeals route back into the same machine logic (Self-Referential).,Substantive; permits the introduction of counter-evidence and refusal.
Human Role,"Sensor : A ""Liability Sponge"" providing a signature for a process they cannot control.","Sentinel : Empowered to weigh context, assess risk, and halt the system."

2. Preventing the "Clarke Mistake": The Right to Interrogate

The "Clarke Mistake" occurs when an institution allows proprietary opacity to substitute for public policy. The Agency shall disqualify any Respondent whose proposal utilizes "commercial sensitivity" or "proprietary IP" as a shield to block the Right of Legibility . In public procurement, an unchallengeable algorithm is an illegal laundering of institutional authority.Contractors are prohibited from implementing Service Level Agreements (SLAs) that prevent the interrogation of outcomes. To maintain integrity, the following Four Pillars of Systemic Interrogation are mandated:

• Algorithmic Disclosure: Respondents shall disclose the specific data sources, Fraud Scores, and Network Connection signals used to evaluate a subject. Disclosing "validation methodology" is insufficient; the specific signals—such as address history or demographic proxies—must be legible at the point of decision.
• Reasoning Access: Systems must provide case-specific logic, not generic policy templates. If a benefit is denied, the system must explain how that specific situation failed.
• Threshold Transparency: Procurement requires full visibility into the weights applied to subjects. Users must see the "map" of risk tiers and the invisible boundaries that sort them.
• Presumption Protection: The burden of proof shall not default to the model. Referencing the 93% failure rate of the Michigan MiDAS system and the Australian Robo-Debt disaster, the Agency forbids the "Presumption of the Model." An algorithmic "flag" shall be treated as an unverified hypothesis, and the system shall not automatically garnish wages or seize assets based on unverified inference.

3. Operationalizing the H∞P Framework: Eliminating the Liability Sponge

The Agency rejects the "Human in the Loop" model when it functions as a "Moral Crumple Zone"—an architecture designed to absorb blame for failures the human lacks the velocity to prevent. All procurement shall mandate the H∞P (Humans in the H∞P) staffing model, ensuring a continuous governance aperture in two lanes:

1. Lane 1 (Training): The pre-deployment economy of labeling and validation.
2. Lane 2 (Execution): The continuous governance state during real-time operation.

Lane 2 Execution Governance Requirements

The Respondent shall provide a Verification Bandwidth assessment. Human oversight is a disqualifying term unless the Respondent proves the human operator has sufficient time to intervene. The Agency explicitly disqualifies any workflow approximating the "Maria/ESG" failure mode, where operators are forced to review 847 items in 6 hours (25.5 seconds per decision).Lane 2 must be funded and staffed with the following roles:

• Flow Controllers: Staff empowered to manage exceptions at machine speed, intercepting ambiguity before it hardens into fact.
• Robopsychologists: Specialists tasked with monitoring and mitigating "automation bias"—the rational but dangerous tendency to trust the dashboard over human senses.
• AInthropologists: Researchers who map the "workaround culture" and execute the Tannie Test to ensure the system recognizes human coordination signals.

4. The Constitutional Right to Refuse: Stop-Work Authority

A system that cannot stop itself is a system that eventually destroys its users. The Agency mandates the implementation of "Negative Power"—the structural right to refuse. Contracts must include a Stop-Work Authority clause, defined as a Constitutional Brake that is:

• Documented: Explicitly included in all Lane 2 job descriptions.
• Unilateral: Invokable without prior permission or management meetings.
• Protected: Structurally insulated from performance penalties or "operator bias" flags.The system shall feature an Architectural Refusal state. If data completeness falls below a specific threshold, the system is prohibited from generating a high-risk score by default. It must return a "Cannot assess. Insufficient data" output, halting the execution pipeline. The contract must specify that the vendor "owns" the cost of delay when a stop is invoked for safety; a system where proceeding is automatic but stopping is expensive is a system designed for harm.

5. Verification & The Tactical Ghost: Auditability Clauses

Accountability often evaporates within the "Tactical Ghost" —where AI capability is dissolved into the underlying "Ontology" (as seen in Palantir’s Oasis framework). When AI is everywhere in the decision chain but nowhere on the audit log, governance is impossible.The Agency mandates an Audit-ability Clause for all subcontractors. Contractors are forbidden from using "Classified Firewalls" or "Proprietary Integration" to block chain reconstruction. If the manufacturer cannot inspect how their own product is used behind a third-party's ontological barrier, the system is disqualified. We must avoid the "Anachronism of Innocence" —the institutional habit of finding a "conscience" (such as the January 22nd Anthropic Constitution) nineteen days after the mission (the January 3rd Caracas raid) has already settled the body count.

The "Tannie Test" for Real-World Verification

The system must be tested against context-dense environments (e.g., the Cape Town taxi rank) where human coordination is high-velocity and informal. The vendor must provide a System Census answering:

1. Who is missing from the data entirely?
2. Who is present but forced into "Unknown"?
3. Can the system recognize "Beadwork as Biometric" or "Storyline Hashes" as signals of coordination, or does it dismiss them as noise?"HAL 9000 didn't need better ethics. HAL needed a grievance mechanism with the power to stop the mission. Everything else is commentary."

6. Conclusion: The New Covenant of Accountability

The Calvin Convention protocol dictates that Helpfulness is the lowest priority in procurement, subordinate to Safety, Ethics, and Interrogability. A system that provides efficient answers while remaining unchallengeable is merely an engine for unaccountable power.

The Procurement Officer’s Litmus Test

No contract shall be executed unless it passes this three-point test:

1. Can the human stop the system unilaterally and without permission?
2. Can the subject contest the logic (e.g., Fraud Scores/Network Connections) without being disqualified?
3. Is Lane 2 staffed to prevent the "25.5-second" verification collapse?"If a system's reasoning cannot be interrogated, it should not be granted authority over human welfare."