Single Paper Review

Produce a single-paper note that is:

• Obsidian-friendly Markdown (YAML frontmatter + headings)
• Method/result focused (not reviewer-style acceptance/rejection)
• Metric-literate (each metric defined mathematically + interpreted)
• Reusable (key figures/tables embedded as Attachments so they can be referenced from other notes)

Use the bundled template

• Preferred template: assets/review_template.md
• If the vault already contains Templates/Single Paper Review Template.md, it is OK to use that instead (keep the structure identical).

Workflow

1. Identify the target note path

   • If the user provides a target .md file, update that file.
   • Otherwise create a new note named after the paper title (short, descriptive, ASCII).

2. Fill frontmatter minimally but correctly

   • author, venue, year, url (and doi/arxiv/code if available)
   • Keep categories: [[Papers]] unless user requests otherwise.

3. Summarize the paper (single-paper "review" = 정리)

   • TL;DR: 2-5 sentences focused on the mechanism and the headline result.
   • Contributions: 3-5 bullet points, phrased as claims (what changed vs prior work).
   • Problem setup: clearly state task, inputs/outputs, assumptions, and what breaks without the method.

4. Write the method with enough detail to re-implement

   • Start with a high-level pipeline, then expand into key components.
   • Include notation (shapes and symbols) where it reduces ambiguity.
   • If the paper includes an algorithm box, rewrite it as concise pseudo-steps (do not copy verbatim).

5. Add a "Metrics (math + meaning)" section (mandatory)
   For every metric used in Methods/Results:

   • Provide a mathematical definition in LaTeX (use $$ ... $$).
   • When using LaTeX, just use single \, instead of double \\.
   • When writing text, if you want to write < or >, prepend \ before the characters. (i.e., \<BOS\>)
   • Specify how the paper computes it (averaging, thresholds, matching rules, postprocessing).
   • Interpret what it means:
     • What higher/lower implies about behavior
     • What it is sensitive to (imbalance, calibration, scale, label noise, etc.)
     • Common failure modes / when the metric can mislead

6. Summarize experiments with "figure/table anchors"

   • State each important claim in plain language.
   • Support it by referencing the exact paper table/figure number and the reported numbers (absolute + delta).
   • Prefer "what matters" over exhaustive enumeration.

7. Make results reusable (figures/tables)

   • Create an attachment folder: Attachments/<paper-slug>/
   • For each key result figure/table:
     • Save a cropped image (PNG preferred) into the attachment folder.
     • Embed it in the note under "Figures and tables (reusable)" with:
       • Source: Figure/Table number + page
       • Why it matters: what conclusion it supports
       • How to read: axes/legend/what to focus on
   • Tables:
     • If simple: transcribe to Markdown table for copy/paste reuse.
     • If complex: embed the table image and optionally add a short "takeaway row" list below it.

Constraints

• Do NOT add steps about generating PDFs of the note.
• Do NOT add steps about creating new schematic/diagram figures.
• Keep text ASCII (use " not curly quotes).