Terms of Use

1. Scope

Evidence Layer is a software project for selecting, ranking, and tracing supplied evidence for use by retrieval-augmented generation systems. The public website includes an authenticated workspace explorer and research documentation. Workspace records shown there are returned by the server after account verification; they are not a synthetic sample. The explorer is not a hosted customer-document retrieval or answer-generation service. These terms must be revised if the scope, operator, or available features change.

2. Acceptable use

You may use the public website and materials only in a lawful and authorised way. You must not:

• attempt unauthorised access, disrupt the site, evade access controls, or probe systems without written authorisation;
• submit or process material unless you have the necessary rights, notices, permissions, and authority;
• use evidence selection or automated output as the sole basis for a decision where a human review or legally required process is needed;
• use the service to violate privacy, intellectual-property, confidentiality, export-control, or other applicable rules;
• misrepresent benchmark findings, source provenance, or the status of an evaluation.

3. Customer responsibilities and data rights

Before any future source connection or hosted retrieval feature is offered, the customer must have authority to access and process the supplied sources and to instruct the service to process them. The customer remains responsible for source permissions, notices to data subjects, the lawfulness and accuracy of supplied data, and reviewing evidence before use. Detailed processor terms and deletion/retention instructions must be agreed separately where the service acts as a processor.

4. Third-party providers and external services

The repository contains authentication/workspace integrations and separate evaluation tooling. Historical Phase 7 and Phase 7B experiments used OpenAI and DeepSeek for test answer generation, and TypeSafe/JEV for evidence scoring. Those experiments used the fixed public Real-World v1 benchmark data. They do not constitute an active answer-generation integration for the public Evidence Explorer. Any future provider connection, data transfer, provider terms, processing location, or subprocessor must be disclosed and authorised before customer data is sent.

External providers operate under their own terms and availability. The operator must publish a verified list of providers and applicable terms before offering an integrated customer service.

5. Evidence selection and AI output

Evidence Layer selects and ranks evidence; it is not itself the answer-generating language model. Selection can omit relevant context, include irrelevant material, or reflect limits in the query, source corpus, scorer, and policy. If a customer or third-party model generates an answer, that answer may be inaccurate, incomplete, outdated, biased, or unsupported. Users must check sources, citations, and material claims before relying on an output.

The product makes no guarantee of a particular percentage of token savings, monetary savings, latency improvement, retrieval quality, answer correctness, completeness, or citation quality. Evaluation results are specific to the disclosed benchmark and do not guarantee production performance.

6. Usage and estimated costs

A future service may measure token counts, request duration, retrieval volume, evidence selection, retries, and estimated provider costs. Cost estimates may use returned token counts and published rate cards. They are not invoices or guaranteed charges: provider billing may vary with model, region, cache, service tier, pricing changes, retries, taxes, or account-specific terms. Any subscription, quota, overage, refund, or payment term must be presented and agreed before that feature is offered. No such commercial terms are confirmed by this repository.

7. Suspension and termination

For a future hosted service, the operator may need to restrict or suspend access to address a security threat, unlawful use, material breach, provider outage, or legal requirement. The final contract must define notice, cure periods, data export/deletion, and effects of termination. The current repository does not establish those commercial processes.

Account deletion requests. A signed-in person may request deletion of the account data this service manages. Accepting a request blocks access immediately and is not reversible in the current implementation. The implemented slice removes source grants, member workspace memberships, and evidence and relationships in workspaces where the account is the sole active owner with no other active members. Workspaces, owner memberships, shared-workspace content, the source catalogue, snapshots, stored objects, and the authentication identity are retained. The interface therefore does not represent complete erasure, and stored files and published snapshots are not deleted.

8. Warranties and liability

The final terms must describe service availability, warranties, liability allocation, and any permitted limitation in a manner reviewed under the chosen governing law. Nothing in the final terms may exclude or limit liability where applicable law does not permit that exclusion or limitation. No service level or guaranteed availability is currently defined.

9. Governing law and contact

Governing law and venue: [insert after confirming the contracting entity and obtaining legal review]. Operator contact: [insert verified legal contact and address]. These details are publication blockers and have deliberately not been inferred from repository or account information.