EN한
← archive

$ cat briefs/daily/2026-11-13.md

2026-11-13

November 13, 2026 (Fri)

1 top story · 1 paper pick · 1 robot-memory signal

**Reconstructed archive:** prepared on **2026-11-15 (KST)** after the original automation failed, using the original morning cutoff of **2026-11-13T08:00:00+09:00** (**2026-11-12 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 **2026-11-12** Hugging Face snapshot. Its bytes match repository commit `b02b96a765d014e31daa5fba8dd3a09c6bc37d1e`, timestamped **2026-11-12 01:32:13 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-2026-11-12.md))

[01]

Top Stories

1. WikiTrace links agent memory to the evidence behind each lesson

The system connects each stored insight to a trajectory summary and the original trajectory, letting an agent inspect missing details or conflicting lessons before acting. Authors report gains across 5 agentic benchmarks over existing memory approaches and an insights-only baseline, with less memory-induced degradation; the abstract supplies no numerical effect sizes.

Why it matters: access to supporting evidence gives agents a way to check remembered advice. These are author-reported comparisons, not independently verified reliability guarantees. Version history: v1, 2026-11-10 16:46:59 UTC. → Agents (LLM Agents) (snapshot) (primary v1)

[02]

Paper Picks

RoboScript: a frozen model rehearses manipulation tasks in simulation and stores feedback in a task memory for deployment in new scenes. Authors report 83.4% mean success across clean and randomized settings on 50 RoboTwin 2.0 tasks, using 1 demonstration per task without gradient updates. Why read it: it separates learning through context from weight updates; the benchmark average does not establish general physical-robot reliability. Version history: v1, 2026-11-09 18:59:45 UTC. → Embodied Agents (snapshot) (primary v1)

[03]

Watch

  • Temporal shortcuts in robot memory: authors find that pauses or wrist fidgets can be mistaken for task progress in a memory-augmented VLA. On RoboMME counting tasks, event-indexed memory recovers half or more of the resulting loss without retraining. Why it matters: memory evaluations should distinguish elapsed time from completed actions; this is a result in the studied designs, not a universal diagnosis. Version history: v1, 2026-11-10 18:47:57 UTC. → Embodied Agents (snapshot) (primary v1)
[04]

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 2026-11-13. Reconstruction: no wiki pages were created; this issue links existing topic pages. (log)
[05]

Updates

  • Original dated work: the 2026-11-08 ingest entry records updates to Agents (LLM Agents) and Agentic Reinforcement Learning; no 2026-11-13 ingest entry is retained. Reconstruction: no wiki pages were edited. An absent entry does not establish an absence of news. (log)