acceptable use & output use policy

PART 1 — Acceptable Use

1. Scope

This Policy applies to everyone using any Resolve service, and to everything you submit or do with what we produce.

We wrote it to be readable. If something here is ambiguous, ask us rather than guessing — we would much rather answer a question than close an account.

2. Don't do these things

• Break the law, or help someone else break it
• Harass, threaten, defraud, stalk, or impersonate anyone
• Submit content you don't have the right to submit
• Try to identify, surveil, or build a profile of a specific person using our tools
• Interfere with the service: attack it, overload it, probe it, or degrade it for others
• Circumvent metering, rate limits, authentication, or account restrictions
• Create accounts by automated means, or run multiple accounts to get more free usage
• Scrape, crawl, or systematically extract from the service beyond what we expressly allow
• Reverse engineer the service, or use it to build a competing product
• Resell, sublicense, or redistribute the service, our Outputs, or Licensed Data except where we expressly permit it

3. Content you must never upload

Some content is prohibited absolutely, with no exception for research, curiosity, journalism, or testing:

• Child sexual abuse material, in any form. Uploads are screened against hash sets of known material. Matches are preserved and reported to the National Center for Missing and Exploited Children, as U.S. federal law requires, and the account is restricted.
• Non-consensual intimate imagery, including material created or altered to depict a person sexually without their consent
• Content that sexualizes minors in any way, including synthetic or animated depictions
• Content depicting or facilitating serious violence, terrorism, or trafficking
• Malware, or content designed to compromise systems

If you encounter any of this, report it to us rather than uploading it again.

4. Content you should think twice about

Not prohibited, but your responsibility:

• Media of private individuals who haven't consented to your submitting it. The person in the file has interests too, and you're representing that you may lawfully submit it.
• Confidential, privileged, or regulated material — health records, legal files, anything under a duty you owe someone else.
• Anything you would not want processed by a third-party compute provider, since video analysis runs on external infrastructure.

5. Account integrity

One account per person. Registration requires a valid payment card, which lets us keep automated account farming off a service where every request has a real compute cost. Free entitlements are per person, not per account: we enforce this at both the payment-method (card fingerprint) and email level, so the same card — or the same person's email — cannot be used to collect free entitlements across multiple accounts.

6. Using our Outputs

This section governs everything the service produces — verdicts, scores, projections, briefs, analyses, and answers. Product-specific rules are in the Annexes.

What Outputs are. Machine-generated, probabilistic estimates. They are wrong some of the time, by design and by nature. Treat them as one input among several.

What Outputs are not. They are not verified facts, professional advice, certifications, authentications, or determinations about any person, file, event, or outcome.

You may: use Outputs for your own personal or internal purposes; use them to inform your own work, research, reporting, or decisions; discuss and quote them with appropriate characterization.

You may not:

1. Present an Output as a verified finding, an authentication, or a statement of fact
2. Attribute an Output to Resolve as a conclusion, verification, or determination
3. Remove or obscure any label identifying an Output as machine-generated
4. Use an Output as the sole basis for an assertion of fact about an identifiable person
5. Use Outputs as evidence in legal, disciplinary, employment, insurance, credit, housing, or law-enforcement proceedings
6. Redistribute Outputs as a data feed, or as part of a competing product

If you cite us, cite us accurately. "Resolve's detector returned a 0.87 probability of manipulation" is fine. "Resolve confirmed the video is fake" is not, and is a misrepresentation of what we do.

7. Licensed Data

Some services include data licensed from third parties. Your rights are limited to using it inside the service and to any export we expressly provide.

Exports are licensed to you for your own personal or internal use only. You may not redistribute, resell, sublicense, or publicly post exported files; collect them systematically or by automated means; aggregate them to reconstitute a dataset; or use them to build or supply a competing data product or service.

8. Enforcement

What we can do. Remove content, restrict features, suspend or terminate accounts, and report to authorities where the law requires it.

Conduct, not identity. We enforce this Policy on conduct, never identity — we do not discriminate on the basis of any protected characteristic, and we do remove people for how they treat others.

Automated action. Some enforcement is automatic — content screening and abuse detection operate without human review. Automated systems make mistakes.

What we'll tell you. If we restrict your account, we'll tell you it was for a violation of this Policy. In some cases the law limits what we can say about why, and we won't speculate about you or characterize what was detected.

Appeals. If you think we got it wrong, contact us. We review every appeal. For automated content screening, our review is limited to technical and account records — we do not review the content itself, and in some cases the law prohibits us from doing so. That limit works in your favor as often as against you, and it's not something we can waive.

Suspension is not deletion. Records we're legally required to keep are kept, regardless of account status.

9. Reporting a problem

Report abuse, prohibited content, security issues, or a mistaken enforcement action at resolveresearcher.com/contact. Copyright complaints go to our DMCA agent — see the Terms of Service, Section 12.

ANNEX A — Detector

A1. What this tool does

The detector estimates the probability that a piece of audio or video was synthetically generated or manipulated. It analyzes signal and physical properties of the file. It returns a probability and a label.

It does not identify anyone. The video pipeline locates and crops faces in order to analyze image properties. It does not compute, store, or compare facial templates, faceprints, or voiceprints, and it cannot recognize or match a person.

A2. How accurate it is

In evaluation, the audio model achieves roughly 0.76–0.81 AUC (source-dependent) and the video model roughly 0.77 AUC. Full methodology is available under NDA.

Performance on your file may be materially worse — evaluation conditions are not real-world conditions.

State plainly: the detector produces both false positives and false negatives. It will sometimes label authentic media as manipulated, and sometimes fail to detect real manipulation. Audio performance is materially weaker than video, and both degrade on sources unlike those used in evaluation.

A3. What you may not do with a verdict

In addition to Part 1, Section 6:

1. Do not publish or repeat a verdict as a factual claim about an identifiable person. "This detector scored the clip at 0.91" is a description of a tool's output. "This video of [person] is fake" is an assertion about a person, and it is yours, not ours.
2. Do not use a verdict as evidence in court, arbitration, employment or disciplinary proceedings, insurance claims, or law enforcement matters. It has not been validated for those purposes and we do not offer expert testimony.
3. Do not represent that Resolve has certified, authenticated, or verified any file. We do not do this and cannot do it.
4. Do not use the detector to investigate or make determinations about a specific private individual.

Journalistic and research use is welcome, with one rule: a verdict may inform your reporting but may not be the sole basis for an assertion of fact, and must be characterized as a probabilistic estimate wherever you cite it. [Note: this refines the blanket "no journalism use" position from earlier in the drafting — a blanket ban would be wrong for the audience this product serves.]

A4. Uploads

By uploading, you represent that you have the right to submit the file and that doing so violates no law and no one's rights.

Your uploads are not retained. Files are processed and deleted. What persists is verdict metadata — a score, a label, a version, a timestamp.

Two exceptions: content matching known illegal-material hash sets, which is preserved and reported as law requires; and material we are legally ordered to preserve.

Video analysis runs on third-party GPU infrastructure. Don't upload anything you're not willing to have processed there.

A5. Audit bundles

(Section removed — the detector does not produce audit bundles. It writes verdict metadata and discards the file. If a bundle feature is added later, this section returns.)

ANNEX B — DeepDive

B1. What this tool does

DeepDive produces AI-generated analysis of news, policy, and public affairs, using large language models operated by third parties. Output is generated on request and delivered to you. It is not published by Resolve and is not reviewed by a human before you receive it.

B2. What it is not

It is not journalism. It is not research we have verified. It is not legal, financial, investment, or voting advice. It is not a factual record.

Language models produce fluent, confident text that is sometimes wrong, including about real people and events. Fluency is not accuracy, and confident phrasing in an Output carries no information about whether it is true.

B3. Content about real people

DeepDive generates analysis about real, living, identifiable people, including political candidates and public officials.

Every factual claim in an Output about a real person is unverified. Check it against a primary source before you rely on it, and certainly before you repeat it.

You may not republish DeepDive output as a factual claim about any identifiable person. If you use it, verify it independently and publish on your own authority, not ours. Attributing an unverified claim about a person to Resolve is prohibited, and it is also the fastest way to create a legal problem for yourself.

B4. Sources

DeepDive draws on published sources. Where an Output references a source, follow the reference — the source is the authority, not the Output.

Outputs are summaries and analysis, not reproductions. We limit quotation from source material, and you should too when you use an Output.

B5. Corrections

We maintain a corrections process. When we learn an Output contained a material factual error, we record the correction and, where the error is systemic, we fix the underlying behavior.

Report an error at resolveresearcher.com/contact. Our published Corrections Policy explains what we do with reports.

B6. Election-related content

Analysis touching candidates and elections is subject to everything above, with emphasis: it is machine-generated, unverified, and not a factual account of any candidate's record or positions.

Do not use DeepDive output in campaign material, paid advertising, or voter communications. Those uses carry disclosure obligations we cannot discharge for you, and the underlying claims are unverified.

This is a termination trigger, and here is the honest reason why. DeepDive runs on language models we license from third parties, whose own policies restrict political campaigning and electioneering use. If your use of our service breaches their policy, they can revoke our access — which takes DeepDive offline for everyone, not just you. We are not willing to trade every other user's access for one account's campaign work.

So: using Outputs for electioneering, or in any other way that breaches an upstream provider's acceptable use policy, is a material breach. We will terminate the Account immediately and without notice, and unused Tokens are forfeited under Section 5 of the Terms of Service. We will refund the unused portion of any paid subscription period, because we are not in the business of keeping money for service we won't deliver. Records we are legally required to keep are kept; termination is a restriction of access, not a purge.

B7. Automation disclosure

Outputs are machine-generated and labeled as such. Don't strip the label. If you share an Output, the person receiving it is entitled to know it came from a machine.