Jaemin's Research Notes

Curation Queue

2 min read

Curation Queue

A curation queue holds items that are promising but not yet ready to become polished wiki notes, paper notes, project notes, logs, or posts. It keeps the public site from filling with raw fragments while preserving follow-up work.

Each queued item should have:

Current Queue (2026-06-26)

#PaperDestinationStatusBlockerNext Action
1The Structural Origin of Attention Sinkpapers/ (TBD)inboxpure-AI, low domain fitVerify venue status; check if super-neuron analysis connects to Attention or Transformer notes
2RelayCachingpapers/ (TBD)inboxpure-AI, low current fitVerify speedup claims; skip unless multi-agent LLM systems become an active track

Queue Fields

  • Source: where the item came from, marked to verify if uncertain.
  • Paper candidate intake: source, metadata, why collected, route, axis, claim, evidence pointer, risk, next action.
  • Destination: paper, concept, project, research, log, or post.
  • Promotion workflow: target destination and reason, following Content promotion workflow.
  • Status: inbox, stub, reading, draft, ready, or drop.
  • Blocker: missing metadata, weak relevance, private detail, missing verification, or duplicate.
  • Next action: the smallest useful action.

Promotion Rules

  • Promote to Papers when the item is paper-specific, metadata is verified, and it passes Paper triage.
  • Use Paper candidate intake before promoting agent-collected paper candidates.
  • Before reproduction planning, record paper-specific public artifacts with Artifact availability.
  • Promote to Concepts when the idea is reusable across papers or projects.
  • Promote to Projects when it describes a public artifact or design decision.
  • Promote to Public logs when it is a cleaned work record.
  • Promote to Posts only when there is a reader-facing story or synthesis.

Drop Rules

  • Drop duplicates when an existing note only needs an update.
  • Drop unverifiable items that cannot be sourced.
  • Drop or keep private anything that cannot be sanitized.
  • Drop items that are merely interesting but do not fit the public scope.

Checks

  • Is the next destination explicit?
  • Is the smallest next action concrete?
  • Has sensitive information been removed before public promotion?
  • Is to verify used for missing metadata instead of invented details?
  • Is the queue shrinking through decisions rather than growing forever?

Graph View

Backlinks