agentleFS
scholayscholay/skills/evidence-review/SKILL.md

evidence-review

Skill141 starsChanged 5 months agoMITView on GitHub

Assess and promote corpus slices toward citation readiness. Use when deciding whether a piece of evidence can be cited, upgrading machine-translated layers to reviewed translations, building evidence matrices for writing, or prioritizing which sources to review first.

8 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 7 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Only level 3 slices should appear as direct citations in writing.

    line 23
  2. Distinctive:

    Don't review everything. Identify which slices are thesis-critical first:

    line 31
  3. Distinctive:

    Verify that the source text (original excerpt or OCR text) accurately represents the page.

    line 43
  4. Distinctive:

    If this slice is suitable for direct quotation or citation:

    line 47
  5. Distinctive:

    Before each writing session, verify:

    line 65
  6. Very specific:

    Every slice you plan to cite has at least verification_status: checked

    line 66
  7. Very specific:

    Every slice you plan to quote directly has quote_ready: true

    line 67
  8. Distinctive:

    Every citekey in your evidence matrix resolves in your bibliography

    line 68
  9. Very specific:

    Do not promote a slice to quote_ready based on quality scores alone.

    line 72
  10. Very specific:

    Do not cite from machine_draft or agent_draft translation layers directly.

    line 73
  11. Distinctive:

    When you first use a slice in writing, add it to the evidence matrix immediately.

    line 76
  12. Very specific:

    Read references/citation-criteria.md for the full checklist of what each ladder level requires for text and image slices.

    line 81

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Evidence Review

The core problem

Most corpus slices are created by automated pipelines: OCR, machine translation, auto-generated descriptions. They are useful for browsing, but not safe to cite. The gap between "we have 10,000 slices" and "we can write the paper" is evidence review — the deliberate process of deciding which slices are trustworthy enough to quote and cite.

Without a clear ladder between "raw" and "citation-ready," researchers either over-trust machine output or spend time reviewing everything equally. Neither is efficient.

The citation readiness ladder

LevelStatusWhat it means
0draftAuto-generated, not yet looked at by a human
1checkedSource text verified against the original document
2reviewedWorking translation confirmed accurate; can be paraphrased safely
3quote_readyCan be quoted or cited directly; citekey and page locator confirmed

Only level 3 slices should appear as direct citations in writing. Level 2 is sufficient for paraphrase and synthesis.

The key insight: levels 0 and 1 are about accuracy of the source text. Levels 2 and 3 are about the translation/working layer. They are separate judgments.

Workflow

1. Prioritize what to review

Don't review everything. Identify which slices are thesis-critical first:

  • Slices already referenced in your writing or notes
  • Slices from your highest-value sources
  • Slices that answer your core research questions directly

For each priority cluster, look at quality scores from corpus building (P1 pages need review before any other use; P2 should be sampled).

2. Per-slice review procedure

For each slice you're reviewing:

  1. Open the original source page alongside the slice card.
  2. Verify that the source text (original excerpt or OCR text) accurately represents the page. Correct OCR errors if found.
  3. Set verification_status: checked when source text is confirmed.
  4. Review the working translation: does it accurately convey the source meaning? Correct or rewrite as needed.
  5. Set translation_status: reviewed when the working layer is trusted.
  6. If this slice is suitable for direct quotation or citation:
    • Confirm the citekey exists in your bibliography
    • Confirm the page locator is correct
    • Set quote_ready: true

3. Maintain an evidence matrix

For each chapter or section you're writing, keep a simple matrix:

Slice IDChapterSectionRoleStatusNotes
SLC-0001-T003Ch22.1supports claim about Xreviewedparaphrase only
SLC-0007-T012Ch33.2direct quotationquote_readycite p.47

This makes the link between your argument and your evidence explicit and checkable.

4. Run the integrity check before writing

Before each writing session, verify:

  • Every slice you plan to cite has at least verification_status: checked
  • Every slice you plan to quote directly has quote_ready: true
  • Every citekey in your evidence matrix resolves in your bibliography

Rules

  • Do not promote a slice to quote_ready based on quality scores alone. Human review of both source text and translation is required.
  • Do not cite from machine_draft or agent_draft translation layers directly. These are working materials, not citation materials.
  • Separate source text accuracy from translation accuracy — they are different judgments, made at different stages.
  • A slice at reviewed level is safe to paraphrase. A slice at quote_ready level is safe to quote.
  • When you first use a slice in writing, add it to the evidence matrix immediately.
  • OCR-assisted image descriptions (ocr_assisted_draft) are not citation-ready visual evidence.

References

  • Read references/citation-criteria.md for the full checklist of what each ladder level requires for text and image slices.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.