Catalog updates three times daily
Public sources are scheduled for scanning three times a day. New paper identities are checked against arXiv; editorial annotations and approvals remain separate.
- Configured scan times
- 00:17, 08:17, 16:17 UTC
- Website rebuild times
- 01:47, 09:47, 17:47 UTC
GitHub scheduling can be delayed. The completion times below record what actually ran, rather than assuming that a scheduled job executed.
Current published catalog
- Reviewed papers and surveys
- 1,200
- Other reviewed resources
- 471
- Total resource records
- 3,251
These counts include editorial work completed after the scan below. Browse reviewed research or inspect full coverage.
Latest completed scan
- Status
- partial
- Started
- 2026-10-23T04:49:56Z
- Finished
- 2026-10-23T04:51:48Z
- Discovery documents fetched
- 54 / 54
- New paper metadata
- 0
- New candidate identifiers
- 4
- Reviewed papers at scan completion
- 1200
- Pending metadata changes
- 3
Inspect the workflow run and logs
Recorded gaps
- recent_arxiv: failed
- HTTP Error 429: Unknown Error
- harvest: Deferred after the recent arXiv query returned HTTP 429; discovered identifiers remain queued.
- GitHub candidate observations paused for provider rate limiting; verified progress and cursor are retained.
- 2 GitHub candidate requests have no accepted metadata response.
Download the full scan receipt
What automation does and does not establish
A scan can discover a source or verify bibliographic fields. It does not reproduce experiments, approve a resource’s relevance, establish dataset permissions, or add editorial summaries. Existing reviewed snapshots are preserved; observed metadata changes are queued for review.
Discovery workflow · Publication workflow · Coverage ledger
Search notification receipts record which changed pages were submitted after publication and the response received. Submission is separate from search indexing.
Latest paper-discovery pass
2026-10-23T04:50:40Z · 32 / 32 query windows succeeded · 1,496 distinct identifiers · 0 new candidates.
Ranked search windows overlap and do not measure global completeness. Newly found identifiers still need primary metadata verification and relevance review.
| Area / query | Unique hits | New candidates | Window limit | State |
|---|---|---|---|---|
| General computer use computer use agents | 75 | 0 | 80 | success |
| Grounding evaluation GUI grounding benchmark | 80 | 0 | 80 | success |
| Mobile learning mobile GUI agent reinforcement learning | 80 | 0 | 80 | success |
| Web data and evaluation web agent dataset benchmark | 80 | 0 | 80 | success |
| Safety and security computer use agent safety | 80 | 0 | 80 | success |
| Memory and planning GUI agent memory planning | 80 | 0 | 80 | success |
| Desktop platforms desktop computer use agent Windows macOS Linux | 80 | 0 | 80 | success |
| Android Android smartphone GUI agent automation | 80 | 0 | 80 | success |
| iOS iOS iPhone GUI agent benchmark | 80 | 0 | 80 | success |
| HarmonyOS HarmonyOS mobile GUI agent | 80 | 0 | 80 | success |
| Screen understanding GUI screen understanding interface element detection | 80 | 0 | 80 | success |
| Visual grounding screenshot visual grounding click localization agent | 80 | 0 | 80 | success |
| Browser interaction visual web navigation browser agent | 80 | 0 | 80 | success |
| Cross application work cross application multi app GUI agents | 80 | 0 | 80 | success |
| Long horizon tasks long horizon computer use agent workflow planning | 80 | 0 | 80 | success |
| Agent memory GUI agent episodic memory self reflection | 80 | 0 | 80 | success |
| Reinforcement learning computer use agent online reinforcement learning rewards | 80 | 0 | 80 | success |
| Demonstrations GUI agent imitation learning human demonstration trajectories | 80 | 0 | 80 | success |
| Synthetic training data synthetic GUI agent training data generation | 80 | 0 | 80 | success |
| Task verification computer use benchmark task evaluation state verification | 80 | 0 | 80 | success |
| Professional software professional software GUI agent CAD Blender 3D | 80 | 0 | 80 | success |
| Office workflows spreadsheets presentations documents computer use agents | 80 | 0 | 80 | success |
| Assistive interfaces accessibility assistive multimodal GUI agents | 80 | 0 | 80 | success |
| Robustness GUI agent robustness dynamic interfaces distribution shift | 80 | 0 | 80 | success |
| Adversarial security browser GUI agent prompt injection adversarial security | 80 | 0 | 80 | success |
| Human collaboration GUI agent human collaboration clarification intervention | 80 | 0 | 80 | success |
| Efficiency GUI agent efficiency latency visual token compression | 80 | 0 | 80 | success |
| Open models open source GUI model computer use checkpoint | 80 | 0 | 80 | success |
| Workflow automation visual workflow automation robotic process automation agents | 80 | 0 | 80 | success |
| Multilingual interfaces multilingual GUI agent interface grounding | 80 | 0 | 80 | success |
| Testing and exploration GUI agent software testing interface exploration | 80 | 0 | 80 | success |
| Chinese language discovery 图形界面 智能体 手机 电脑操作 | 9 | 0 | 80 | success |
Per-query new counts depend on query order and count first discovery in this pass. Query registry · Results and source receipts
Conference proceedings sweep
65 keyword-selected titles were inspected across 4 official proceedings snapshots. This is a separate bounded sweep, not a complete conference census. Title candidates and possible renamed papers remain unapproved pending identity and scope review.
Inspect the selected titles, review boundaries, and primary receipts.