Language Model Use and Transparency Disclosure

Consumer Dividends

Effective date: June 14, 2026

Last updated: June 14, 2026

This disclosure explains how our products use language models, what data is processed, and what the outputs are and are not. We publish it so that customers, their clients, and regulators can understand our posture in one document. Hand this to your compliance team, your broker, or your counsel.

1. What our products do

Neither product we sell uses a language model. Numidian runs a fixed scoring framework in ordinary code. Holdings tools are single files that run arithmetic in your browser. There is no model call anywhere in either one, at any point.

Numidian is a risk inventory and scoring tool used by HR consultants, employment attorneys, and HR managers. It scores organizational exposure across five dimensions using a fixed factor framework and a state-by-state law matrix, estimates settlement exposure from the salary supplied, Title VII damage caps (42 U.S.C. § 1981a) and state damage rules, and cites published EEOC charge-receipt data on every report with the source table and retrieval date named. Every score, every dollar figure, and every line of the report text is produced by documented rules. The same inputs always produce the same report.

Holdings tools are deterministic calculators you buy once and run on your own machine. They make no network requests of any kind, which you can confirm yourself in your browser's network panel.

2. What our products do not do

Our products do not make decisions about individuals.

Numidian does not decide whether to hire, fire, promote, discipline, or take any other employment action against any person. It produces a risk score for an advisor to read. The decision sits with the employer.

Numidian is not an "automated employment-related decision technology" under the Connecticut Artificial Intelligence Responsibility and Transparency Act (Public Act 26-15, Section 31-57aa et seq., applying to systems deployed on or after October 1, 2027). The CART Act covers technology that processes personal data and produces output that is a "substantial factor" in an employment-related decision about an individual. Numidian's output is a structured risk inventory for the firm's own analysis. It does not direct or materially influence a decision about any individual employee.

Numidian is not a "covered automated decision-making technology" under the Colorado Automated Decision-Making Technology Act (SB 26-189, effective January 1, 2027) for the same reason. The decision-maker is the employer. Numidian informs the analysis of the employer's advisors.

If a deployer of our product chooses to use the output as a substantial factor in an employment-related or consequential decision about an individual, that deployer is responsible for compliance with the laws that apply to that decision, including any disclosure, notice, bias audit, or impact assessment obligations.

3. What data is sent to a language model

None. We have no language model provider, and none is listed in our Subprocessor List. The case facts you enter into Numidian are processed on our own servers by our own code and are never transmitted to a model provider. Anything you enter into a Holdings tool never leaves your device at all.

4. Whether your data trains models

No. We do not use the content of your inputs, or any report generated for you, to train, fine-tune, or otherwise improve any model. Your case facts stay yours.

Aggregate, non-identifying usage statistics (such as how many reports were generated in a month) may inform product improvements. The content of your inputs is not used for this.

5. How outputs are produced

Numidian uses a fixed scoring framework. Each input field feeds one of five scoring dimensions through documented rules. A jurisdiction multiplier is applied based on the state entered, and the settlement range is built from statutory damage caps and published wage medians. The report text is assembled from the resulting scores by rule, not written by a model.

The practical consequence is that the tool is auditable. The same case scored twice returns identical output, and any figure in the report can be traced back to the rule and the source that produced it. That is the reason we build this way.

6. Known limitations

• A deterministic model is only as good as the framework behind it. Numidian scores documented patterns from EEOC and federal court data. It does not know your judge, your jurisdiction's recent case law, or the facts you did not enter.
• Settlement figures are statistical estimates built from general damage caps and median wage data. They are not a valuation of your specific case.
• The state law matrix covers 15 states in detail. Other states fall back to a federal baseline, which the report labels as such.
• The tool cannot replace professional judgment. Numidian outputs are not legal advice.

7. Bias and fairness

Numidian scores against documented data sources (EEOC charge data, federal court timing patterns, a 15-state law matrix). The scoring framework is fixed, so the same inputs produce the same scores every time. The protected class field is used to apply the legal framework appropriate to the situation, not to differentiate scoring against any individual.

We do not represent this as a complete or certified bias audit. We document it so that you and your counsel can evaluate it. Where regulators require formal anti-bias testing or impact assessments for the use of an automated decision tool, those obligations sit with the deployer in their specific deployment.

8. Transparency to end users

Because our products do not make decisions about individuals and do not interact directly with employees, applicants, tenants, or buyers, the disclosure obligations that apply to consumer-facing or employee-facing automated decision tools do not directly apply to us. The Connecticut CART Act notice requirements at Section 31-57aa (effective for systems deployed on or after October 1, 2027) apply to deployers using automated decision technology in employment contexts. We are not such a deployer.

We recommend that customers using our products document their use of the products in their own internal records and, where the output will be used in a way that touches an employee or applicant, consider whether their own jurisdiction requires them to disclose that an automated tool informed the work.

9. Where processing happens

Numidian runs on our own application hosting in the United States, listed in our Subprocessor List. Your inputs travel over an encrypted connection (TLS) to our servers and nowhere else. Holdings tools process nothing remotely: they run in your browser, on your machine, with no connection at all.

10. Your rights and how to exercise them

You have the right to:

• Know what inputs you have submitted (kept according to the retention rules in our Privacy Policy).
• Request a copy or deletion of your inputs and outputs.
• Object to specific processing.
• Ask questions about how a specific output was produced.

Email support@consumerdividends.com from the address associated with your account.

11. Updates

We update this disclosure when the law changes, when we add a new product, or when our practices change. The "Last updated" date at the top reflects the most recent change.

12. Contact

support@consumerdividends.com

Consumer Dividends, Hawaii, United States.