Paper Reading Summary Design

Goal

Strengthen Qiongli B2 /paper-read so it produces both trustworthy single-paper reading notes and project-level paper reading summaries that can be reused by gap analysis, related-work writing, synthesis, and manuscript drafting.

User-Approved Scope

Use the combined approach:

1. Enhance the existing /paper-read workflow rather than introducing a new top-level task ID.
2. Keep B2 as the canonical task for targeted paper reading.
3. Add explicit project-level summary artifacts for organizing multiple paper notes.
4. Treat truthfulness and boundary control as hard requirements: summaries must be evidence-grounded, uncertainty must be visible, and missing evidence must never be filled in by invention.

Artifact Contract

B2 should continue to produce:

• notes/
• bibliography.bib
• retrieval_manifest.csv

B2 should additionally produce or update:

• literature/paper_reading_summary.md
• literature/paper_reading_matrix.md

The single-paper note remains RESEARCH/[topic]/notes/[citekey].md. The project-level summary and matrix live under RESEARCH/[topic]/literature/ because they organize the reading corpus rather than a single paper.

Workflow Behavior

/paper-read should run two layers:

1. Single-paper deep reading:

   • Retrieve or identify full text where possible.
   • Normalize metadata and citekey.
   • Write a structured paper note.
   • Record retrieval status and bibliography state.

2. Project-level summary integration:

   • Update or create literature/paper_reading_matrix.md.
   • Update or create literature/paper_reading_summary.md.
   • Integrate the current paper into themes, methods, datasets, findings, limitations, contradictions, gaps, and reusable writing points.
   • Preserve existing human-written content. If an update cannot be merged safely, append a clearly marked dated entry instead of replacing prior prose.

Truthfulness And Boundary Controls

All B2 summary outputs must follow these rules:

• Do not invent citations, page numbers, sample sizes, methods, results, effect sizes, datasets, author claims, or implications.
• Each central summary claim must be traceable to a note, source section, quote, table, abstract, or metadata field.
• If only metadata or abstract is available, label the note and summary entry with evidence_limit: abstract_only.
• If a full-text claim cannot be verified, write it as a gap note or uncertainty, not as fact.
• Separate author claims, agent interpretation, and project relevance.
• Mark inference strength explicitly: direct_evidence, reasonable_inference, or unsupported_gap.
• Keep strong synthesis claims out of B2 unless multiple notes support them; B2 may organize evidence, but it should not overstate review-grade coverage.

Template Changes

templates/paper-note.md should add explicit fields for:

• evidence limit and retrieval status
• source anchors for key claims
• author-stated contribution
• agent interpretation
• project relevance
• reusable citation points
• uncertainty and missing-information notes

Create templates/paper-reading-summary.md for project-level narrative organization:

• corpus overview
• theme clusters
• method and data patterns
• stable findings
• contradictions and contested claims
• research gaps
• implications for the current project
• writing-ready citation points
• uncertainty register

Create templates/paper-reading-matrix.md for compact cross-paper comparison:

• citation/citekey
• evidence limit
• theory/framework
• method/identification
• dataset/source
• main finding
• limitations
• project relevance
• claim anchors

Contract And Reference Updates

Update the canonical contract and generated references so B2 declares the two new literature artifacts.

Primary source files:

• standards/research-workflow-contract.yaml
• qiongli-workflow/workflows/paper-read.md
• qiongli-workflow/references/stage-B-literature.md
• qiongli-workflow/references/workflow-contract.md generated from the YAML contract
• templates/paper-note.md
• templates/paper-reading-summary.md
• templates/paper-reading-matrix.md

Generated or mirrored payloads should be refreshed by existing sync scripts after source edits.

Testing And Verification

Verification should include:

• A failing-first check that B2 contract outputs include the new summary and matrix artifacts.
• A failing-first check that the paper note template exposes evidence-limit and anti-hallucination fields.
• A failing-first check that the new summary template requires source anchors and uncertainty handling.
• python3 scripts/validate_research_standard.py --strict
• sync/package checks needed to keep payload copies consistent.

Out Of Scope

• No new top-level task ID.
• No new provider implementation.
• No automatic PDF parsing engine.
• No claim that B2 produces systematic-review-grade synthesis.
• No fabricated examples using real paper details.