Evidence-Led Paper Reading

Read research papers with AI while keeping important claims tied to source locations, limitations and the boundary between findings and inference.

Who this is for

For graduate students, research assistants, analysts, researchers and knowledge workers who need to decide whether academic evidence is useful enough to cite or apply.

Result

A source-located reading memo that maps the study, separates evidence identities, records limitations and ends with a practical next action.

Start here

  1. State the decision you need to make from the paper.
  2. Confirm whether the source is a full PDF, full text, excerpts or an abstract.
  3. Map the research question, gap, design, data, findings and stated limitations.
  4. Attach pages, sections, figures or tables to consequential claims.
  5. Stress-test the conclusion with alternative explanations, null results and scope limits.
  6. Decide whether to cite, compare, reproduce, apply cautiously or hold.

Find out what the paper supports before turning it into a summary

The Skill starts with the decision the reader needs to make. It maps the study, traces consequential claims back to the source and keeps conclusions inside the design and source coverage actually examined.

Paper map
Research question, claimed gap, design, data, primary outcome, main finding and boundary.
Evidence table
Claim, evidence identity, source locator, what it establishes and the caveat that changes interpretation.
Next action
Cite, compare, reproduce, apply cautiously or do not use yet.

Begin with one paper and one real decision

Provide a readable PDF or full text and explain whether you need to understand the argument, verify a claim, inspect the method, compare studies or decide whether the paper is useful enough to cite.

  1. Set the reading decision
  2. Declare source coverage
  3. Map the paper
  4. Locate consequential claims
  5. Stress-test the conclusion
  6. Choose the next action

Keep four evidence identities separate

A paper-reported finding is not the same as the authors’ interpretation. A reader inference is recorded as an inference, and missing or unreadable information stays unknown. Important claims include a page, section, figure, table or appendix locator whenever the source permits it.

Paper-reported finding
A result directly supported by the paper's data or analysis.
Author interpretation
The paper's explanation of what that result means.
Reader inference
A reasoned conclusion added during the reading.
Unknown
Information that the available source does not establish.

Traceability is not a guarantee of correctness

The method can expose where a conclusion comes from and where it overreaches. It does not replace peer review, specialist judgement or independent replication, and one paper is not enough for a high-stakes medical, legal, safety or financial decision.

Download the fixed v1.0.0 Skill

The public repository and versioned ZIP are available under the MIT License. The ZIP SHA-256 is ff98769953bab0a6a1c39247c513ae34ff384878cfe7979f61f94615a3df9b00. The core Skill requires no external service; PDF access depends on the reading tools available in the host agent environment.

Limits

The Skill makes a reading traceable. It cannot establish that a paper is correct, replace peer review or domain expertise, or claim replication without independent evidence. Conclusions remain limited to the source coverage available.