AI and Automated Review Addendum

1. Scope and current boundary

This Addendum supplements the EvidenceGuard Terms of Service and applicable DPA. It covers automated document observations and any separately enabled external AI processing. It does not itself enable a provider, a feature or Customer’s entitlement.

Customer-document external AI is disabled in the baseline covered by this version. The public sample is pre-authored illustrative material and is not a live analysis of a visitor’s documents. Native PDF extraction, optical character recognition and deterministic checks are not the same as an external generative-model review. A later feature must satisfy this Addendum before it receives Customer Content.

2. What a permitted external processing step requires

Before sending Customer Content or a content-equivalent derivative to an external provider, we must identify the provider and relevant deployment, the specific purpose, the source version and selected content scope, applicable processing location and material data-handling terms. Scope can include extracted text, selected page images or context needed for the stated task; an image of a page is still Customer Content.

The step must be within the accepted service scope, the organization’s permitted policy, the user’s authority and the document’s eligibility. Where explicit authorization is required, it must be a separate, recorded authorization for the disclosed step or a specifically described bounded workflow. A generic acceptance of Terms, purchase of credits, earlier authorization for another document or product-statistics setting is not enough.

Required provider contracts, retention and training-use restrictions, security assessment and any transfer safeguard must be established. We will not infer a zero-retention, no-training, regional or private-deployment guarantee solely from a public marketing page, a request parameter or a configuration label. If a required permission or safeguard is unknown, the operation must not dispatch.

3. Limits of authorization and withdrawal

Authorization permits only the disclosed purpose and recipients. It does not permit independent advertising, sale, model training, a research corpus, broader document collection, unrestricted onward transfer or a change to retention. Material changes to provider, purpose or scope require the applicable new notice and authorization.

Customer can withdraw prospective authorization or request a policy change through available controls or our contact. We stop new processing dependent on it after the change is effective. Withdrawal cannot undo a completed transmission or retroactively make lawful completed processing unlawful; retained copies and outputs remain governed by the applicable deletion and provider terms. We must explain any material limitation rather than claim a completed external request was never sent.

4. Accuracy and evidence limitations

Automated systems can omit a requirement, misread a number or symbol, invent an unsupported interpretation, confuse sources or versions, mishandle a table, miss image-only text or repeat content. A citation, model agreement, deterministic label or high confidence does not eliminate those risks. Output may be nonunique, and the Service does not promise intellectual-property protection in every output.

Coverage is distinct from semantic correctness. A completed run does not prove complete visual, text, table or form coverage. Missing evidence must remain unresolved; a blank form is not complete merely because its file was located. A machine observation must not be recorded as a customer’s human approval. Old results must be considered with their source versions and freshness.

Customer must inspect material sources, resolve discrepancies and make the final decision. The Service does not provide a procurement-law opinion, legal advice, a compliance certificate or a winning-bid guarantee. These limitations do not eliminate our express duties or nonwaivable liability.

5. Human review and support access

Human Verification in the product can mean a customer user reviewing an observation. It is not automatically a staffed service. A separate staffed engagement must identify the authorized material, reviewer access, scope, deliverable and any agreed timing. Access cannot be extended to unrelated documents or used to create an independent training set.

Reviewers must distinguish observed evidence, interpretation, uncertainty and customer decisions. They must not claim an unsupported PASS, imply a professional credential they do not possess, or submit a bid on Customer’s behalf without separate authority. No professional legal engagement is created by this Addendum.

6. Prohibited consequential and deceptive uses

The standard service must not be used to make solely automated legal or similarly significant eligibility decisions about people, perform unlawful profiling, or disguise a machine output as official approval. It must not be used to evade procurement obligations or falsify source evidence. Public examples must retain their illustrative status.

7. Incidents, fees and contact

Provider-related incidents follow the DPA and applicable law. Credit consumption, service-side failures and cash refunds follow the Billing Policy; authorization is not a payment, and payment is not authorization. A provider failure cannot be turned into proof that no requirement exists.

For a processing-scope, authorization or output concern, contact kun11451454188@gmail.com with subject “EvidenceGuard AI processing” and nonsecret source/run references. Do not email confidential documents or tokens unless a suitable secure channel has been arranged. Postal contact is SourceWarden LLC, 7155 Kindred Street, Philadelphia, PA 19149-1132, United States.