research-document
AI Engineering Handoff
AI Engineering Handoff
Objective
Continue evidence-traceable evaluation of long-horizon repository agents while preserving the repository's pre-ROS history and compatibility boundaries.
Current state
- Root
CURRENT-STATE.mdis authoritative;context/CURRENT-STATE.mdis a derived ROS entry point. - Evaluation Cycles 001–003 and the repository-wide non-human comparative review are complete. No agent capability trial or independent replication is claimed.
- The 20 commits before ROS installation are reconstructed as eight explicitly
historical records in
.ros/history/historical-work.jsonl. - Historical records never enter the live
.ros/events/events.jsonlstream and cannot authorize transitions. - The state-constrained research corpus is publishable as Markdown.
- Eleven proposed research missions in
missions/backlog/now represent the deduplicated RFR-001–RFR-010 and NX-005 research TODO set.
Historical migration completed
- Human index:
docs/repository/ros-migration/HISTORICAL-WORK-INDEX.md - Policy decision:
research/decisions/DF-ROS-HISTORY-2026-0001--historical-attribution-policy.md - Schema:
schemas/historical-work.schema.json - Validation implementation and tests:
tools/ros_cli.mjs,tools/test_ros_history.mjs
The backfill is reconstructed from Git and repository documents. It does not claim that historical work originally followed ROS. Original external work-item IDs, approvals, and tests remain unknown when no repository evidence exists.
Validation
Run:
npm test
npm run research:validate
./ros registry check
./ros validate
./ros status
Unresolved questions and risks
- Independent reviewers have not yet validated ET-004-P01 or ET-014-P01.
- Frozen graders have not been challenged with independently authored unseen outcomes.
- No successful GitHub Pages deployment receipt is recorded.
- The historical state-constrained corpus lacks a single execution journal covering authorship and evidence collection across all 12 missions.
- Metadata coverage remains diagnostic and incomplete outside strict ROS artifact directories.
Next action
Start with prompts/Non-Human-Experimental-Research-Next-Mission.md: obtain
independent pilot reviews and challenge frozen graders with unseen outcomes
before running any contained exploratory capability baseline.