Files
glpi-neural-brain/CHANGELOG-AUTONOMOUS-RESEARCH-ROUTING-V7.md
jbergner 440423c5b6
release-tag / release-image (push) Successful in 2m43s
RC-3
2026-08-09 11:29:13 +02:00

30 lines
1.5 KiB
Markdown

# Changelog: Autonomous Research Routing v7
## Changed
- Added entity-facet-aware bounded primary-source exploration for compound research questions.
- Allow up to two distinct primary/authoritative entity facets per query, one probe per facet.
- Keep the strict full-text evidence assessment after a facet probe.
- Preserve autonomous opportunity production nodes as required article seeds through topic filtering and planner source selection.
- Added `article.sources.autonomous_seeds` observability event.
- Added initial-bootstrap Compute-Agent registration grace using the existing `BRAIN_VECTOR_GRAPH_AGENT_WAIT` value.
- Added `vector.graph.agent.waiting` observability event.
## Explicitly unchanged
- `BRAIN_AUTONOMOUS_RESEARCH_INTERVAL` is not changed or reinterpreted.
- No autonomous cooldown/dedupe-key behavior was changed in this release.
- No vector graph similarity, affinity or orphan-pass thresholds were changed.
- No article/research quality gate was lowered.
## Validation focus
Regression tests cover:
- two primary facets (`OWASP SAMM`, `MITRE ATT&CK`) may each receive one bounded probe;
- a second source for the same facet is deferred;
- a primary full-text source may provide valid partial evidence for one side of a compound comparison question;
- an unrelated authoritative source still cannot bypass the topic/entity guard;
- autonomous opportunity seeds survive topic filtering and are prepended to planner source IDs;
- startup Agent grace is active only during bootstrap and uses the configured vector-agent wait.