In plain English
This page is part of the technical reference. It keeps the expert detail but starts with a plain-language summary for first-time readers.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
Evidence and Provenance
Cognivirus.com uses six evidence labels, local source summaries, /docs/manifest.json, /docs/integrity-manifest.json, and UTC review dates to keep research-dependent claims inspectable.
Direct answer
Every major CognivirusA behavior pattern that can survive, move, or reappear across a changing AI system. Open glossary definition claim should show what kind of evidence supports it, where the supporting source summary lives, what the evidence does not show, and when the item was reviewed. The site must distinguish primary findings from site interpretation.
Evidence labels used by the site
- Demonstrated
- Experimentally observed
- Emerging evidence
- Architectural inferenceA conclusion or output produced from data. Open glossary definition
- Open research question
- Speculative scenario
Provenance surfaces
| Surface | Purpose |
|---|---|
/docs/manifest.json | Lists source metadata, evidence level, publication date, and local summary path. |
/docs/integrity-manifest.json | Records SHA-256 hashes for preserved local docs and summaries. |
/evidence | Provides readable claim cards and source summaries. |
.uai/long-term-memory.uai | Points future maintainers to durable reports, summaries, and handoff records. |
/contact | Gives a human correction path. |
Limitation
A hash proves that a local file did not silently change after hashing. It does not prove that an external source is true, complete, current, or universally applicable.