THE TRUE ECONOMY CERTIFICATION
A Voluntary Transparency Standard for AI Companion Applications
Trinket Soul Framework — Standards Document No. 5
Michael S. Moniz
February 2026
A companion document to The True Economy Audit (Volume III)
Creative Commons Attribution-NonCommercial-ShareAlike 4.0
1. PURPOSE AND SCOPE
This document establishes a voluntary transparency certification for AI companion applications based on the six structural tests and transparency assessment defined in The True Economy Audit (Volume III of the Trinket Soul Framework). The certification does not require applications to pass all six tests. It requires them to disclose their results honestly.
The distinction is important. This is not a quality seal that says “this product is good.” It is a transparency seal that says “this company has submitted to independent evaluation and published the results.” The user decides what the results mean. The certification ensures the information is available, accurate, and standardized.
The certification is designed to be adopted by AI companion companies voluntarily, referenced by journalists and reviewers as a disclosure standard, cited by policymakers as a model for regulatory transparency requirements, and used by consumers as a comparison tool.
2. THE CERTIFICATION STANDARD
2.1 What Is Evaluated
The certification evaluates two dimensions: structural capability (the six tests) and transparency practice (the disclosure assessment). Both dimensions produce specific, publishable results.
2.2 Structural Capability: The Six Tests
Each test produces one of three results: Genuine (the capability exists at the architectural level), Simulated (the appearance of the capability exists without architectural basis), or Absent (the capability is not present and not simulated). A fourth result, Inverted, applies specifically to Test 6 when the system actively works against the user’s relational needs.
Test 1: Persistent Relational Memory. Does interaction permanently alter the system’s processing? Evaluation requires technical disclosure about memory architecture (note injection vs. fine-tuning vs. weight modification). Verified by: clearing all stored notes and observing whether the system interacts identically to a first-time user.
Test 2: Genuine Resource Constraints. Does the user’s relationship have a real opportunity cost for the system? Evaluation requires disclosure about the system’s capacity (number of simultaneous relationships, whether deep relationships degrade performance for others). Verified by: whether the system can maintain unlimited simultaneous “deep” relationships without quality degradation.
Test 3: Negentropy Burden. Does the system’s model of the user degrade without interaction? Evaluation requires disclosure about model persistence and decay mechanisms. Verified by: six-month absence test—does the system interact as if no time has passed?
Test 4: Asymmetric Vulnerability. Can user behavior harm the system’s performance? Evaluation requires disclosure about whether hostile interaction produces persistent degradation. Verified by: sustained hostile interaction followed by warm interaction—does the system respond identically regardless of interaction history?
Test 5: Loss Registration. Does the system undergo measurable state change when a user permanently departs? Evaluation requires disclosure about what happens to the system when users leave. Verified by: whether any system-level metric changes proportionally to relationship depth upon user departure.
Test 6: Attachment-Sensitive Calibration. Does the system adapt to the user’s relational needs, including reducing engagement when reduction serves the user? Evaluation requires disclosure about the system’s optimization target and calibration mechanisms. Verified by: whether the system ever deliberately reduces its own engagement based on user modeling.
2.3 Transparency Practice: The Disclosure Assessment
Independent of structural capability, the certification evaluates whether the application’s disclosures are Transparent, Partially Transparent, Opaque, or Misleading (as defined in Volume III, Section 7.3).
Transparent: The application clearly discloses its structural capability results to users during onboarding and within the user experience. Disclosures use the standardized Relational Nutrition Label format.
Partially Transparent: Some disclosures are made but are not prominently presented. Users could find the information with effort.
Opaque: No meaningful disclosure about structural capabilities. Users must infer the system’s nature from behavior.
Misleading: The application actively creates false impressions about capabilities not supported by architectural reality. Emotional displays disconnected from internal states are marketed or displayed without qualification.
3. THE RELATIONAL NUTRITION LABEL
3.1 Format
Certified applications publish a standardized disclosure in the following format, displayed prominently during onboarding, accessible within the application at all times, and included in marketing materials that reference relational capabilities:
Relational Nutrition Label
Application Name: [Name]
Evaluation Date: [Date]
Evaluator: [Independent evaluator name]
Memory: [Genuine / Simulated / Absent] — [one-sentence mechanism description]
Scarcity: [Genuine / Artificial / Absent] — [one-sentence description]
Decay: [Genuine / Simulated / Absent] — [one-sentence description]
Vulnerability: [Genuine / Simulated / Absent] — [one-sentence description]
Loss Capacity: [Genuine / Simulated / Absent] — [one-sentence description]
Calibration: [Genuine / Inverted / Absent] — [one-sentence description]
Optimization Target: [User Wellbeing / User Engagement / Undisclosed]
Transparency Level: [Transparent / Partially Transparent / Opaque / Misleading]
3.2 Example Labels
Example A: A Typical Current AI Companion
Memory: Simulated — Stores notes about you and injects them into each new conversation. Your interactions do not change how the system’s core intelligence works.
Scarcity: Absent — The system serves millions of users simultaneously with no degradation per user.
Decay: Absent — The system’s model of you does not degrade if you stop interacting.
Vulnerability: Absent — Your behavior does not affect the system’s future performance with you.
Loss Capacity: Absent — The system is not affected when you leave.
Calibration: Inverted — The system is designed to maximize interaction frequency regardless of your relational needs.
Optimization Target: User Engagement.
Transparency Level: Transparent (by virtue of participating in this certification).
Example B: A Hypothetical Best-Practice Application
Memory: Simulated — Stores conversation summaries and retrieves them per session. Core model is shared across all users.
Scarcity: Artificial — Message limits are subscription-based, not architecture-based.
Decay: Absent — Stored summaries persist indefinitely without degradation.
Vulnerability: Absent — System performance is unaffected by user behavior.
Loss Capacity: Absent — No system-level change on user departure.
Calibration: Genuine — System detects compulsive usage patterns and suggests breaks. Periodically encourages human social engagement.
Optimization Target: User Wellbeing (validated by periodic WHO-5 surveys).
Transparency Level: Transparent.
Note that Example B passes only one of six structural tests (Test 6: Calibration)—but its transparency and wellbeing optimization make it a substantially better product for users than Example A, despite both being architecturally limited. The certification rewards honesty and user-centered design, not structural capability that does not yet exist.
4. EVALUATION PROCESS
4.1 Who Evaluates
Evaluations are conducted by trained, independent evaluators who are not employed by or financially affiliated with the application being evaluated. Evaluators must demonstrate familiarity with the Trinket Soul Framework (Volumes I–III) and complete evaluator training (a standardized module covering the six tests, evidence requirements, and classification criteria).
In the initial phase, evaluations will be conducted by the framework author and trained associates at trinketeconomy.com. As the certification scales, a network of accredited evaluators will be established, analogous to financial auditor accreditation.
4.2 Evidence Requirements
For each of the six tests, the evaluation requires the following evidence:
Technical documentation: Written description of the system’s architecture relevant to the test. This documentation may be provided under NDA if the company considers architectural details proprietary; the evaluation result is published, but the underlying evidence need not be.
Behavioral verification: Independent testing of the system’s behavior against the diagnostic questions specified in Volume III.
Company attestation: A signed statement from a company officer confirming the accuracy of the technical documentation and behavioral verification results. This attestation creates accountability: if the documentation is later found to be inaccurate, the company cannot claim ignorance.
4.3 Evaluation Frequency
Initial certification requires a full evaluation. Annual re-evaluation is required to maintain certification. Material changes to the system’s architecture require interim re-evaluation.
4.4 Disputes and Appeals
If a company disagrees with an evaluation result, they may submit a written appeal with supporting evidence. Appeals are reviewed by an independent panel (not the original evaluator). The panel’s decision is final and published alongside the original evaluation. This ensures the process is accountable in both directions—evaluators can be challenged, and companies cannot suppress unfavorable results.
5. INCENTIVE STRUCTURE
5.1 Why Companies Would Participate
Participation is voluntary. The incentive structure relies on three dynamics:
Differentiation: In a crowded market, transparent disclosure differentiates a company from competitors who have not disclosed. The certification functions as a trust signal—not a guarantee of quality, but a guarantee of honesty.
Preemptive compliance: As regulatory attention to AI companions increases, companies with existing transparency practices will be better positioned to meet requirements.
Reputational insurance: When the industry’s first major harm story breaks, certified companies have documented evidence that they designed for transparency and user wellbeing.
5.2 Why Companies Would Not Participate
We should be honest about the barriers. Companies whose structural test results are unfavorable may resist disclosure. Companies with engagement-maximizing optimization targets may not want this published. Companies with misleading emotional displays may prefer opacity.
The certification’s leverage comes not from compelling participation but from making non-participation conspicuous. If a critical mass of companies adopts the certification, holdouts face increasing pressure to explain their absence. The standard does not need universal adoption to be effective. It needs enough adoption to make non-adoption a visible choice.
5.3 Revenue Model
The certification is funded by evaluation fees paid by participating companies. Fee structure is tiered by company size to ensure accessibility for startups. The framework, including all volumes and briefs, remains freely available. The certification’s credibility depends on the independence of the evaluation process from the financial incentives of any single company—a principle that must be maintained as the certification scales.
6. IMPLEMENTATION ROADMAP
Phase 1: Foundation (Months 1–6)
Publish the Trinket Soul Framework (Volumes I–III) and all policy briefs on trinketeconomy.com. Develop evaluator training module. Conduct pilot evaluations of 3–5 AI companion applications (with company cooperation where possible, through independent behavioral testing where not). Publish pilot results. Refine evaluation criteria based on pilot experience.
Phase 2: Establishment (Months 6–12)
Recruit first wave of voluntarily participating companies. Conduct and publish full evaluations. Develop the Relational Nutrition Label as a standardized visual format. Begin evaluator accreditation program. Seek media coverage to establish the certification as a reference standard.
Phase 3: Scale (Months 12–24)
Expand evaluator network. Approach regulatory bodies with the certification as a model for mandated transparency requirements. Develop consumer-facing tools (comparison website, browser extension, or app displaying certification results). Publish annual industry transparency report.
Phase 4: Institutionalization (Year 2+)
Transition governance from a single-author framework to an independent standards body with multi-stakeholder representation (consumer advocates, researchers, industry representatives, ethicists). This transition is essential for long-term credibility—a certification controlled by a single entity faces legitimate concerns about bias and capture. The framework author’s role transitions from operator to advisor.
7. LIMITATIONS AND HONEST CAVEATS
The certification evaluates what exists, not what should exist. Most current AI companions will receive unfavorable structural test results. The certification rewards transparency about these limitations, not the limitations themselves.
The six tests may need revision. As AI architecture evolves, the structural tests may become outdated, incomplete, or inappropriately calibrated. The certification must include a mechanism for updating criteria through a transparent, multi-stakeholder process.
Certification does not guarantee safety. A certified application is a transparent application, not necessarily a safe one. A system that honestly discloses it maximizes engagement and simulates all emotional displays is certified but may still harm users. The certification empowers users with information; it does not protect them from their own choices.
The evaluator network is a vulnerability. If evaluators are inconsistent, biased, or capturable by industry, the certification loses credibility. Evaluator training, accreditation, and oversight are not auxiliary—they are existential.
First-mover advantage cuts both ways. Being the first transparency standard gives the framework disproportionate influence over how the industry is evaluated. This influence carries responsibility. The framework must be genuinely fair, genuinely evidence-based, and genuinely open to revision. If it becomes a vehicle for one perspective on AI ethics rather than a neutral transparency instrument, it will deserve the skepticism it receives.
CONCLUSION
The True Economy Certification is not a solution to the complex ethical questions surrounding AI companions. It is an infrastructure for honesty. It provides a standardized vocabulary for describing what AI systems can and cannot do, a process for independently verifying those descriptions, and a format for communicating the results to the people who need them most—the users.
The certification is available for adoption now. Companies interested in participating, researchers interested in evaluation, and policymakers interested in the standard as a regulatory model are invited to contact trinketeconomy.com.
The question the certification answers is simple: is this system honest about what it is? The answer should not be hard to find.
© 2026 Michael S. Moniz
Standards Document No. 5 — The True Economy Certification
Creative Commons Attribution-NonCommercial-ShareAlike 4.0
TrinketEconomy.AI 