Evidence
Every product answer should point back to evidence.
Video intelligence fails commercially when a system sounds confident but cannot show what supported the answer. Ayneye treats evidence bundles as first-class product outputs.
Evidence operating model
| Evidence part | Why it exists | How users inspect it |
|---|---|---|
| evidence_ref | Stable pointer from answer to video span | Displayed beside answer and artifact record |
| time span | Shows when the observation happened | Timeline UI and evidence viewer |
| supporting artifact | Connects answer to scene graph or detector report | Artifact download or API response |
| confidence and review note | Prevents unsupported certainty | Dashboard badge and answer trace |
| blocked action | Keeps risky workflows under review | Policy result and review_required state |
Evidence-first product behavior
Evidence is not only a compliance feature. It improves the user experience because teams can trust what they can inspect. A security reviewer can open a span. A developer can debug an API response. A finance reviewer can see whether repeated questions reuse state. An agent can cite evidence instead of making unsupported claims.
Evidence quality test
- Ask a factual question with a clear visible event.
- Verify that the answer includes evidence_refs.
- Open the referenced time span.
- Check whether the evidence really supports the answer.
- Ask an ambiguous question and expect review_required instead of a confident guess.
Product boundaries are part of the product, not footnotes. Ayneye is not presented as a replacement for every GPU video foundation model, not a free-infinite-query engine, and not an autonomous surveillance decision system. The current beta path is controlled signup, dashboard, REST API, bounded processing, evidence artifacts, visible limits, and read-only agent/MCP-style evaluation. Hard scenes, identity-sensitive workflows, emergency response, physical access, discipline, and destructive actions require human review or remain blocked.