Legal / Terms of use

Terms of use.

1. Operator and scope

ML Referee is operated by Sandra Carrico as an individual, not by a corporation. These terms govern access to and use of the ML Referee website, ChatGPT connection (MCP connection), hosted evaluation service, beta-access process, and related support. OpenAI and Auth0 provide separate services; they do not operate ML Referee, and their names do not imply sponsorship, endorsement, or partnership.

2. Accounts and authority

You may use ML Referee only if you are at least 18 years old and use the service for business or professional research. If you act for an organization, you must have authority to bind that organization. You are responsible for your ChatGPT and Auth0 accounts, the security of your own credentials, and activity submitted through your authenticated connection. Do not share passwords, OAuth secrets, access tokens, or recovery codes with ML Referee support.

3. Submitted data and permission to process it

You keep whatever rights you have in data and evaluation materials you submit. You authorize Sandra Carrico and the service providers used to operate ML Referee to receive, host, encrypt, inspect by automated processes, analyze, produce results from, secure, troubleshoot, and delete submitted material as needed to provide and protect the service, comply with these terms, and comply with applicable law. This is a limited permission to operate the service, not a transfer of ownership.

Evaluation submissions may contain synthetic data, non-personal data, or genuinely anonymized data. They must not contain regulated data or identifiable personal data, including pseudonymized person-level records. Ordinary account and support contact details may be provided only through the designated account and support channels; they must not be included in evaluation datasets.

You must have the rights, permissions, and lawful basis needed to submit the data and direct the requested evaluation. Do not submit confidential data you are not authorized to use, executable code, credentials, or data whose use would violate law or another person’s rights. ML Referee does not use customer evaluation data to train models.

4. Acceptable use

You may use ML Referee for lawful feature evaluation, bounded model-repair research, and related analysis. You may not use it to break the law; harm, deceive, surveil, or discriminate unlawfully against people; violate intellectual-property or privacy rights; probe or disrupt the service without written authorization; bypass access, credit, rate, or security controls; impersonate another person; submit malicious content; or resell or provide access to the service without permission.

You remain responsible for deciding whether a proposed use, dataset, feature, model, or deployment is appropriate for its domain, including uses involving employment, housing, credit, education, health, insurance, public benefits, finance, safety, or other consequential decisions.

The Acceptable Use Policy supplies additional rules and is incorporated into these terms. If it conflicts with these terms, these terms control.

5. Invite-only beta

The current service is an invite-only beta. Connecting or signing in, submitting a beta request, or receiving a support reply does not automatically grant access or credits. Approval is manual and may be accepted, declined, limited, suspended, or withdrawn. Beta features, limits, interfaces, and availability may change while the service is tested.

6. Credits

Credits are service-use units, not currency. A quote states the credits required before an evaluation is submitted. Credits may not be transferred, sold, or used outside the authorized account. Current beta credits expire on January 1, 2027 at 12:00 a.m. Pacific. Signup does not automatically grant credits.

No paid purchase, subscription, renewal, refund, or payment policy applies because paid access is not currently offered through the ChatGPT connection (MCP connection). Any future paid offer will include separately disclosed terms before purchase.

7. Service availability

ML Referee is provided as an experimental beta on an “as is” and “as available” basis. To the extent permitted by law, no guarantee is made that the service or its results will be accurate, complete, uninterrupted, error-free, or fit for deployment or any particular purpose. No service-level commitment or guaranteed response time is offered. The service may be unavailable, delayed, interrupted, changed, or discontinued, and the operator may limit or suspend access to protect the service, comply with law, address misuse, or administer the beta.

8. Evaluation results and user responsibility

ML Referee results are evidence within the objective, baseline, candidate family, data, timing, assumptions, thresholds, and other conditions declared for that evaluation. They are not guarantees of future performance, causal effects, model quality, safety, legal compliance, fairness, fitness for a particular purpose, or deployment readiness.

Results can narrow a research search and identify supported, unsupported, or unresolved candidates. They do not build, deploy, monitor, or approve a final model. You are responsible for reviewing the result, checking data and assumptions, conducting appropriate independent and prospective validation, and deciding whether or how to use any feature or model.

9. Rights in reports and ML Referee

As between you and ML Referee, you retain your rights in submitted inputs. You may use, modify, and share reports produced for your account without attribution to ML Referee, subject to any rights or restrictions that apply to the underlying inputs. Sandra Carrico retains all rights in the ML Referee service, software, interfaces, documentation, and methodology. Receiving a report does not transfer those rights or authorize access to nonpublic implementation details.

10. Privacy, retention, and deletion

The Privacy Notice describes the current identity, submission, result, log, hosting, retention, and deletion practices. Use the support form for privacy or deletion requests. These terms do not create a different retention schedule.

11. Third-party services

ChatGPT, Auth0, website hosting, and form-delivery services are operated by separate providers under their own terms and privacy practices. ML Referee is not responsible for those providers’ independent services. Use of their names on this site identifies technical providers only.

12. Ending use and changes

You may stop using ML Referee at any time. The operator may limit, suspend, or end access to administer the beta, protect the service or other users, address misuse, or comply with law. Sections that by their nature should survive will continue after use ends.

The operator may update these terms and will post a revised effective date. Material changes apply prospectively and will not alter a dispute for which arbitration or litigation has already begun. Continued use after updated terms take effect constitutes acceptance.

13. Warranty disclaimer

To the fullest extent permitted by law, ML Referee disclaims all express and implied warranties, including implied warranties of merchantability, fitness for a particular purpose, title, and non-infringement. No statement, advice, or information outside these terms creates a warranty. ML Referee does not warrant that the service, an evaluation, or a report will meet your requirements; be accurate, complete, timely, secure, uninterrupted, or error-free; interoperate with another system; preserve data without loss; or produce a result suitable for deployment. Some jurisdictions do not allow particular warranty exclusions, so those exclusions apply only to the extent the law permits.

14. Limitation of liability

To the fullest extent permitted by law, Sandra Carrico will not be liable for indirect, incidental, special, consequential, exemplary, or punitive damages arising from or relating to ML Referee, including loss of profits, revenue, use, data, goodwill, business opportunity, or business interruption, regardless of the legal theory and even if advised that such damages were possible.

To the fullest extent permitted by law, Sandra Carrico’s total aggregate liability arising from or relating to ML Referee will not exceed the greater of (a) one hundred U.S. dollars ($100) or (b) the amount you actually paid for the ML Referee service giving rise to the claim during the six months before the event that gave rise to liability. This limitation applies even if a stated remedy does not fully compensate for a loss or fails of its essential purpose. It does not apply to liability that applicable law does not allow the parties to exclude or limit.

These terms do not impose a broad customer indemnity.

15. Dispute resolution and class-action waiver

Please read this section carefully. It is intended to require most disputes to be resolved through binding individual arbitration rather than in court, waive a jury trial for arbitrated claims, and prevent class or representative proceedings, subject to the exceptions and rights that applicable law requires.

Informal resolution first

Before starting arbitration or a court proceeding, either party must send a written description of the dispute and requested resolution through the support form and allow 30 days for a good-faith effort to resolve it. This requirement does not prevent a party from seeking urgent relief when waiting would cause legally cognizable harm.

Individual arbitration

After the informal process, any remaining claim arising from or relating to ML Referee or these terms is intended to be resolved by final, binding arbitration in English before one arbitrator, administered by JAMS under the rules applicable to the claim. The Federal Arbitration Act governs the arbitration agreement, and the arbitrator applies applicable substantive law and limitations periods. Judgment on an award may be entered by a court with jurisdiction.

Court exceptions

Either party may bring an eligible individual claim in small-claims court. A court with jurisdiction may hear a request for emergency or injunctive relief, a claim for public injunctive relief when applicable law preserves that right, or an intellectual-property dispute. Courts may also decide matters that applicable law does not permit an arbitrator to decide.

Fees and opt-out

Arbitration fees will be allocated under the applicable JAMS rules and governing law. If those costs would make arbitration prohibitively more expensive than court, ML Referee will pay the amount the arbitrator determines is necessary to preserve access to the process. You may opt out of arbitration and the class-action waiver by submitting a written notice through the support form within 30 days after you first accept these terms. The notice must identify the account and clearly state the decision to opt out.

Individual proceedings only

To the fullest extent permitted by law, each party may bring claims only in its individual capacity, not as a plaintiff or class member in a class, consolidated, collective, or representative proceeding. An arbitrator may not combine different users’ claims unless all affected parties agree or applicable law requires otherwise. If this waiver cannot lawfully be enforced for a particular claim or requested relief, that portion will proceed in a court with jurisdiction while the enforceable portions remain in arbitration where permitted.

California law and court venue

California law governs these terms without regard to conflict-of-law rules. Any court proceeding that is not subject to arbitration must be brought in the state or federal courts located in San Francisco, California, and each party consents to those courts’ jurisdiction and venue.

16. Additional terms

Force majeure. Neither party is liable for delay or failure caused by events beyond its reasonable control, including failures of utilities, networks, hosting, identity, or other third-party infrastructure, except that this does not excuse payment obligations already due.

Severability and no waiver. If a provision is unenforceable, it will be limited or severed to the minimum extent necessary, and the remaining provisions will continue. A failure to enforce a provision is not a waiver.

Assignment. You may not assign these terms without written consent. Sandra Carrico may assign them in connection with a transfer of the ML Referee service or business, subject to applicable law.

Entire agreement and survival. These terms, the incorporated Acceptable Use Policy, and the Privacy Notice are the entire agreement concerning the service unless a signed written agreement says otherwise. Provisions that by their nature should survive termination—including ownership, disclaimers, liability limits, and dispute terms—will survive.