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.
Danger Model and UAI Memory Expansion v1.15.0
Version 1.15.0 adds a deeper danger-model section, new Risk Lab worksheets, new schematic concepts, new glossary terms, and a refreshed .uai source-report memory ledger.
What changed
- Added
/danger-modeland supporting pages for lifecycle, action-layer risk, observability, synthetic feedback, promotion rules, governed diversityUseful variety under rules. Open glossary definition, retirement failure, warning signals, evidence boundaries, and implementation review. - Added browser-side worksheet panels for transition-graph review, action-boundary review, observability coverage, synthetic feedback, promotion pressure, and retirement completeness.
- Preserved newly uploaded reports under
/docs/source-reports/raw-markdown. - Added public-safe source summaries under
/docs/source-report-summaries. - Regenerated
/docs/report-coverage-inventory.json. - Regenerated
.uai/long-term-memory.uai,.uai/report-pointer-index.uai, and per-report.uai/memory/source-reports/*.uairecords.
Memory boundary
Hot .uai files remain concise. Full report bodies remain in /docs/source-reports. Long-term memory points to durable report paths with stable IDs, summaries, authority, review status, proof of use, and checksums.
Safety boundary
The new content remains defensive and non-operational. It does not provide exploit instructions, self-replication recipes, evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition-bypass steps, credential-harvesting workflows, or backdoor construction guidance.