[{"assessment_date":{"precision":"day","value":"2026-09-26"},"assessment_type":"tracker_evidence_synthesis","assessor":"RSI Tracker (Codex evidence synthesis)","budget":{"notes":"See study resource accounting; evaluation constraints are distinct from cumulative search cost.","status":"known"},"dimensions":{"adaptivity":{"source_refs":[{"locator":"§3.2","source_id":"src-aide2"}],"status":"measured","summary":"Seven accepted rewrites; additional runs accept two and four."},"efficiency":{"source_refs":[{"locator":"Table 2; §3.6","source_id":"src-aide2"}],"status":"qualitative","summary":"Per-benchmark constraints; no decisive outer-loop efficiency gain."},"meta_recursion":{"source_refs":[{"locator":"§3.6","source_id":"src-aide2"}],"status":"qualitative","summary":"Evolved AIDE47 invoked as outer improver; three seeds per arm are inconclusive."},"retention":{"source_refs":[],"status":"not_reported","summary":null},"stability":{"source_refs":[{"locator":"§5","source_id":"src-aide2"}],"status":"qualitative","summary":"Noisy selection can admit harmful changes."},"transfer":{"source_refs":[{"locator":"§3.3","source_id":"src-aide2"}],"status":"qualitative","summary":"Four held-out benchmarks."}},"domain":"ai_rd_frontier_model_development","effective_l5":{"comparison_budget_matched":"yes","independent_evaluation":"no","notes":"Improved harness transfer is reported; superior outer self-improvement is not established.","source_refs":[{"locator":"§§2–3.6","source_id":"src-aide2"}],"status":"partial"},"extraction_review":{"notes":"Checked against original source; not human review or independent experimental replication.","reviewed_at":"2026-09-26T05:09:03Z","reviewer":"Codex","status":"agent_checked"},"framework_id":"duan-2026","id":"rec-aide2","independence":{"notes":"Source-author report; no independent replication recorded.","status":"known"},"publication_status":"published","recursive_acceleration":{"notes":"Not established by this particular source; no universal negative claim.","source_refs":[{"locator":"§§2–3.6","source_id":"src-aide2"}],"status":"not_demonstrated"},"source_refs":[{"locator":"§§2–3.6","source_id":"src-aide2"}],"structural_l5":{"created":"yes","inherited":"yes","invoked":"yes","notes":"Mechanism and system boundary must be read with the associated study; task gains alone do not prove better future improvement.","retained":"yes","source_refs":[{"locator":"§§2–3.6","source_id":"src-aide2"}],"status":"demonstrated"},"subject_id":"aide2-study","subject_type":"study"},{"assessment_date":{"precision":"day","value":"2026-09-26"},"assessment_type":"tracker_evidence_synthesis","assessor":"RSI Tracker (Codex evidence synthesis)","budget":{"notes":"See study resource accounting; evaluation constraints are distinct from cumulative search cost.","status":"unknown"},"dimensions":{"adaptivity":{"source_refs":[{"locator":"§4; Figure 2","source_id":"src-dgm"}],"status":"qualitative","summary":"Best-of-search coding gains over 80 search iterations, not one continuous lineage."},"efficiency":{"source_refs":[],"status":"not_reported","summary":null},"meta_recursion":{"source_refs":[{"locator":"§3","source_id":"src-dgm"}],"status":"qualitative","summary":"Modified agents participate in later self-modification; archive and parent selection remain fixed."},"retention":{"source_refs":[],"status":"not_reported","summary":null},"stability":{"source_refs":[{"locator":"Limitations and safety discussion","source_id":"src-dgm"}],"status":"qualitative","summary":"Evaluation exploitation and harmful candidates remain concerns."},"transfer":{"source_refs":[{"locator":"§4.3","source_id":"src-dgm"}],"status":"qualitative","summary":"Full Polyglot and cross-model/language checks."}},"domain":"software_engineering","effective_l5":{"comparison_budget_matched":"unknown","independent_evaluation":"unknown","notes":"The revised agent produces descendants, but gains in task performance do not isolate improved search effectiveness under a matched meta-level comparison.","source_refs":[{"locator":"v1 §§3–4","source_id":"src-dgm"}],"status":"unclear"},"extraction_review":{"notes":"Checked against original source; not human review or independent experimental replication.","reviewed_at":"2026-09-26T05:09:03Z","reviewer":"Codex","status":"agent_checked"},"framework_id":"duan-2026","id":"rec-dgm","independence":{"notes":"Source-author report; no independent replication recorded.","status":"known"},"publication_status":"published","recursive_acceleration":{"notes":"Not established by this particular source; no universal negative claim.","source_refs":[{"locator":"v1 §§3–4","source_id":"src-dgm"}],"status":"not_demonstrated"},"source_refs":[{"locator":"v1 §§3–4","source_id":"src-dgm"}],"structural_l5":{"created":"yes","inherited":"yes","invoked":"yes","notes":"Mechanism and system boundary must be read with the associated study; task gains alone do not prove better future improvement.","retained":"yes","source_refs":[{"locator":"v1 §§3–4","source_id":"src-dgm"}],"status":"demonstrated"},"subject_id":"dgm-study","subject_type":"study"},{"assessment_date":{"precision":"day","value":"2026-09-26"},"assessment_type":"tracker_evidence_synthesis","assessor":"RSI Tracker (Codex evidence synthesis)","budget":{"notes":"See study resource accounting; evaluation constraints are distinct from cumulative search cost.","status":"unknown"},"dimensions":{"adaptivity":{"source_refs":[{"locator":"Enhancing AI training and inference","source_id":"src-alphaevolve"}],"status":"qualitative","summary":"Reported kernel and training-efficiency gains."},"efficiency":{"source_refs":[{"locator":"Enhancing AI training and inference","source_id":"src-alphaevolve"}],"status":"qualitative","summary":"Target-code efficiency reported; full search costs not disclosed."},"meta_recursion":{"source_refs":[],"status":"not_reported","summary":null},"retention":{"source_refs":[],"status":"not_reported","summary":null},"stability":{"source_refs":[],"status":"not_reported","summary":null},"transfer":{"source_refs":[],"status":"not_reported","summary":null}},"domain":"ai_rd_frontier_model_development","effective_l5":{"comparison_budget_matched":"unknown","independent_evaluation":"unknown","notes":"Optimized artifacts contribute to training infrastructure; this source does not test a revised improvement mechanism controlling later improvement.","source_refs":[{"locator":"Enhancing AI training and inference","source_id":"src-alphaevolve"}],"status":"unclear"},"extraction_review":{"notes":"Checked against original source; not human review or independent experimental replication.","reviewed_at":"2026-09-26T05:09:03Z","reviewer":"Codex","status":"agent_checked"},"framework_id":"duan-2026","id":"rec-alphaevolve","independence":{"notes":"Source-author report; no independent replication recorded.","status":"known"},"publication_status":"published","recursive_acceleration":{"notes":"Not established by this particular source; no universal negative claim.","source_refs":[{"locator":"Enhancing AI training and inference","source_id":"src-alphaevolve"}],"status":"not_demonstrated"},"source_refs":[{"locator":"Enhancing AI training and inference","source_id":"src-alphaevolve"}],"structural_l5":{"created":"unknown","inherited":"unknown","invoked":"unknown","notes":"Mechanism and system boundary must be read with the associated study; task gains alone do not prove better future improvement.","retained":"unknown","source_refs":[{"locator":"Enhancing AI training and inference","source_id":"src-alphaevolve"}],"status":"not_demonstrated"},"subject_id":"alphaevolve-study","subject_type":"study"}]