Skip to content

_framework.md

Latest commit

 

History

History
103 lines (85 loc) · 4.98 KB

_framework.md

File metadata and controls

103 lines (85 loc) · 4.98 KB
Raw

Analytical Framework — what to extract from each paper

Used as the extraction prompt for subagents. Goal: per-paper, return a compact structured record of writing patterns. Quote actual phrases (≤25 words each) — do NOT paraphrase, because we want the real lexical patterns.

A. Title & Abstract

  • A1. Title pattern: declarative finding ("X enables Y"), tool-name colon ("Foo: a method for ..."), question, or noun phrase. Quote the title.
  • A2. Abstract sentence map: number each sentence; for each, label its rhetorical role from this set:
    • BIG-PICTURE / FIELD-IMPORTANCE
    • GAP / UNRESOLVED-PROBLEM
    • HERE-WE / METHOD-INTRO
    • KEY-RESULT-1 (quantitative)
    • KEY-RESULT-2 (generalization or second axis)
    • VALIDATION / BENCHMARK
    • IMPLICATION / OUTLOOK
  • A3. Quote the "Here we" pivot sentence verbatim (the one that announces what the paper does).
  • A4. Quote the strongest quantitative claim in the abstract.

B. Introduction (first ~3–6 paragraphs)

  • B1. Opening hook style: which of {field-importance statement, surprising fact / number, recent-advance recap, biological/physical motivation, technical paradox}. Quote sentence 1.
  • B2. Funnel structure: in what paragraph does the topic narrow from broad → specific? Note ¶ count.
  • B3. Gap signaling phrases: quote 2–3 phrases like "However,", "remains unclear", "limited by", "no general method exists".
  • B4. "Here we" pivot paragraph: quote the first 2 sentences that introduce this paper's contribution.
  • B5. Contribution preview: are contributions enumerated ("First… Second… Third…") or narrated? Quote the list/sentence.
  • B6. Last-paragraph-of-intro pattern: does it preview results, list contributions, or state significance? Quote the last sentence.

C. Results — section-level structure

  • C1. Number and titles of Results subsections. Quote each header.
  • C2. Header style: declarative finding ("Method X recovers Y"), neutral task ("Benchmarking on Z"), or question.
  • C3. For each Results subsection, quote the opening sentence. Classify it as: claim-first, motivation-first, or method-first.
  • C4. Figure call-out phrasing: quote 2 examples of how figures are introduced (e.g. "we found that … (Fig. 1a)", "Figure 1 shows …").
  • C5. Quantitative + statistical reporting style: quote one passage that reports numbers + uncertainty + test (e.g. "AUC 0.92 ± 0.03, P < 1e-5, two-sided t-test, n=…").
  • C6. Baseline comparison phrasing: quote one passage where the paper compares to prior method.
  • C7. Robustness/ablation framing: quote one phrase like "We next asked whether …" / "To rule out …" / "As a control, …".
  • C8. Generalization framing: quote a phrase that lifts a single result to a general claim.

D. Discussion / Conclusion

  • D1. Opening sentence of Discussion: restate finding, frame in field, or limitation-first? Quote.
  • D2. Limitation phrasing: quote one limitation sentence.
  • D3. Outlook phrasing: quote one "future work / will enable" sentence.
  • D4. Closing sentence of paper: quote it.

E. Methods (only if visible / accessible)

  • E1. Top-level methods subheadings: list them.
  • E2. Reproducibility / data-availability sentence: quote.

F. Figures & captions

  • F1. Caption-title pattern: are caption titles full declarative sentences ("X improves Y across Z") or noun phrases ("Overview of method")? Quote one.
  • F2. Panel-letter usage: how many panels in main figures (typical), and how dense.

G. Cross-cutting linguistic devices

  • G1. Hedging vocabulary present: list which appear — "suggest", "indicate", "demonstrate", "establish", "consistent with", "we propose".
  • G2. Strong-claim verbs present: "enable", "outperform", "achieve", "establish", "identify".
  • G3. Connectives between paragraphs: quote 2 — e.g. "We next asked", "To test this", "Building on this", "Taken together".
  • G4. "Taken together / collectively / together these" usage: quote if present.

Output format per paper

paper:
  cite: "First-author et al., Journal Year"
  url: "..."
  category: "methods | application | theory"
A1_title: "..."
A2_abstract_map:
  - {n: 1, role: "BIG-PICTURE", quote: "..."}
  - ...
A3_here_we: "..."
A4_strongest_quant: "..."
B1_hook_style: "..."
B1_quote: "..."
B3_gap_phrases: ["...", "...", "..."]
B4_pivot: "..."
B5_contribution: "..."
B6_last_intro: "..."
C1_headers: ["...", "..."]
C2_header_style: "..."
C3_openers: [{header: "...", quote: "...", class: "claim-first"}, ...]
C4_figcalls: ["...", "..."]
C5_quant: "..."
C6_baseline: "..."
C7_ablation: "..."
C8_generalization: "..."
D1_disc_open: "..."
D2_limitation: "..."
D3_outlook: "..."
D4_closing: "..."
E1_methods_subheads: ["..."]
F1_caption: "..."
G1_hedges_used: ["..."]
G2_strong_verbs_used: ["..."]
G3_connectives: ["...", "..."]
G4_taken_together: "..."
notable: "anything striking / unusual / template-worthy"

Keep quotes exact and ≤25 words; truncate with "…" if needed. Skip a field rather than guess.