$ cat briefs/daily/2026-11-12.md
2026-11-12
November 12, 2026 (Thu)
1 top story · 1 paper pick · 1 authorization signal
**Reconstructed archive:** prepared on **2026-11-15 (KST)** after the original automation failed, using the original morning cutoff of **2026-11-12T08:00:00+09:00** (**2026-11-11 23:00 UTC**). This issue was not published or emailed on the missed date; no subscriber delivery is authorized. **Provenance and limits:** the research below is preserved in the **2026-11-11** Hugging Face snapshot. Its bytes match repository commit `a3e0a544b7551e57f062799f45aabc8b8dae3d09`, timestamped **2026-11-11 00:02:20 UTC**, before the cutoff. Version-specific primary records below confirm earlier submissions; these are not releases on the issue date. Coverage is limited to abstracts and metadata; full papers, code and results were not independently validated. This selective reconstruction cannot recover the complete morning feed. Items without verified pre-cutoff availability are omitted. Wiki links provide retrospective topic context and may contain later updates. ([snapshot](../../sources/papers-daily/hf-daily-2026-11-11.md))
Top Stories
1. Reframe trains visual agents to use the evidence their tools return
The framework combines training that encourages reliance on tool observations with distillation that replaces deficient tool actions and resumes from the repaired observation. Authors report outperforming Mix-RL and vanilla MOPD across 14 image and video benchmarks, without numerical effect sizes in the abstract.
Why it matters: the evidence available during a rollout can shape what an agent learns, alongside the teacher's outputs. These are author-reported comparisons, not independently verified deployment gains. Version history: v1, 2026-11-09 18:59:54 UTC. → Agentic Reinforcement Learning (snapshot) (primary v1)
Paper Picks
InfraRSI: optimize distributed training settings and code together while fixing workload, loss and timing constraints, then recheck candidates against the baseline. Across 4 dense and MoE workloads, authors report up to 2.75× speedup and mean Model FLOPs Utilization rising from 17.4% to 30.0%. Why read it: checking correctness and accounting for failed trials matter when agents optimize infrastructure; the reported speedup is not a demonstrated reduction in total optimization cost. Version history: v1, 2026-11-08 18:59:37 UTC. → Agents (LLM Agents) (snapshot) (primary v1)
Watch
- ADB: an authorization-boundary benchmark pairs compliant paths with unauthorized shortcuts across 648 samples. Authors report that 86.18% of tasks have an observed successful trajectory without violations, while stronger capability accompanies more boundary drift in their evaluations. Why it matters: task success and authorization compliance need separate measurement; this association does not establish that capability causes overreach. Version history: v1, 2026-11-06 07:47:25 UTC. → Agent Runtime Containment (snapshot) (primary v1)
New in Wiki
- Original dated work: the retained log records creation of the Harness-S and Hybrid Computer Use paper pages on 2026-11-08, not 2026-11-12. Reconstruction: no wiki pages were created; this issue links existing topic pages. (log)
Updates
- Original dated work: the 2026-11-08 ingest entry records updates to Agents (LLM Agents) and Agentic Reinforcement Learning; no 2026-11-12 ingest entry is retained. Reconstruction: no wiki pages were edited. An absent entry does not establish an absence of news. (log)