Record · analyzed
R² Flow: Recursive Self-Improvement via Recursive Skill Evolution
Alternates training the policy that chains skills, checking proposed edits with an independent verifier, and committing versioned skill-library changes.
1
Listed
Added to the list. How entries were checked when they were listed is in the journal.
2
Collected
| File | Source | SHA-256 |
|---|---|---|
| paper PDF2609.33867.pdf | arxiv.org/pdf/2609.33867 | 19f89696dd7f4c43317d55ec25f0d2cf984019de53d33bc95254f6417fb92b08 |
To check a file, download it from its source and compare its SHA-256 with the one above. Collected copies are kept in the editor's RSI system; manifest: design-life/literature/rsi-workflow-improvement/sources.json.
3
Analyzed
Quoted verbatim from design-life/literature/rsi-workflow-improvement/comparison.md (2026-10-07). Parts about applying the paper to the editor's private work are left out.
From “Evidence for improving reusable agent workflows” (2026-10-07)
Inspected edition: R² Flow, v1, 27 September 2026; PDF pp. 3–9, Appendix C
Mechanism and reported evidence: Separates skill frequency from signed usefulness; admits library edits using separate verifier evidence and paired validation. Across eight phases, 33 of 37 committed edits improved the reported held-out verified score. With reward-derived labels, 25 of 44 edits raised training reward while reducing verified performance.
4
Replicated
Not done yet. Not attempted yet.
5
Audited
Not done yet. Not attempted yet.
The editor's view
None recorded yet.