Conference Tracking

For each conference with new results:

1. Navigate to the official accepted papers page — e.g.:
   • OpenReview: openreview.net/group?id=ICLR.cc/YEAR/Conference
   • Virtual site: conference.cc/virtual/YEAR/papers.html
   • Proceedings: proceedings.mlr.press/vXXX/ (ICML), papers.nips.cc/paper_files/paper/YEAR (NeurIPS)
2. Use web_fetch on the official URL to retrieve the full paper list.
3. Do NOT fall back to arxiv keyword search as a substitute for the official list. The goal is the complete, official acceptance record — not a keyword-filtered subset.
4. If the page is too large, fetch the proceedings index or category listing pages.

Step 4 — Analyze Overall Trends from the Full List

From the complete paper list, identify:

• Top directions by paper count (cluster similar titles by topic)
• Emerging topics (strong presence this year vs. previous years)
• Methodology signals (theoretical vs empirical, large-scale vs efficient, benchmark types)
• Acceptance rate context (if available: total submissions vs acceptances)

Step 5 — Project-Relevant Subset Analysis

Using the project's research topic (extracted in Phase 1/2 of the radar protocol):

• Filter the full list for papers in the same or adjacent research area
• Identify papers directly competitive with or complementary to the project
• Note what properties the accepted papers in this area share (what seems to be rewarded)

Step 6 — Actionable Recommendations for the Project

Provide specific, evidence-based advice:

• Target venue: which conference best fits, based on acceptance trends
• Strengthen: what to add or emphasize based on what reviewers seem to reward
• Mitigate risks: what weaknesses to address based on apparent rejection signals
• Competitive positioning: how to differentiate from accepted papers in the same area

Step 7 — Record Execution in Memory

Call memory_write at the end:

• kind: "job_run"
• scope: "job:radar.conference.track"
• intent: "job_progress"
• ttl: "90d"
• Content must include: which conferences were checked, which had new results, brief trend summary, date of this execution