Literature Discovery

The core problem

Most researchers start by downloading everything that looks relevant. This creates an unmanageable pile with no structure and no sense of what's missing. The alternative: build the corpus in deliberate waves, starting from a small seed set and expanding outward in three directions.

Workflow

1. Define scope before searching

Write one sentence: what is this research about, and what is it not about? This sentence is your filter for every candidate you encounter.

Then build a search matrix:

2. Start from a seed set, not a broad search

Pick 3–5 items you already know are strong:

• One overview or survey paper
• One foundational monograph or book
• One case-study cluster
• One methods paper

Do not start by downloading 200 results from a keyword search.

3. Expand in three directions from each seed

Backward — who does this paper cite? Follow references to earlier foundational work.

Forward — who cites this paper? Use Google Scholar, Semantic Scholar, or OpenAlex to find later work building on it.

Lateral — what related terms or adjacent topics does this paper not cover? Search those gaps explicitly.

4. Record every candidate before annotating

Keep a running candidate list. Add a row for every item you encounter, even if you don't have the PDF yet. Minimum fields per row:

• title, authors, year
• source (where you found it)
• language
• status: candidate / keep / exclude / needs_pdf / needs_translation
• reason (one line)

Do not annotate or read deeply at this stage.

5. Check for coverage bias

Before promoting items to intake, ask:

• Am I over-representing one language, one region, one time period?
• Are there practitioner sources, grey literature, or non-English materials I haven't touched?
• What would a critic say is missing from this list?

6. Promote only strong items

Only keep-status items move to the next stage (source intake). Everything else stays in the candidate list with a reason recorded.

Output expectations

• A reusable search matrix
• A candidate list with statuses and one-line reasons
• A gap list: what's under-covered and why
• A keep queue ready for source intake

Rules

• Search in waves. One wave = one seed's full expansion before starting the next seed.
• Never download everything from a keyword search in one pass.
• De-duplicate before promoting: DOI first, then title + year, then manual check.
• Keep translated titles and original titles together.
• Record the search URL and date for every batch — you will need to verify later.
• A candidate list with 200 items and no statuses is not useful. Triage as you go.

References

• Read references/search-matrix.md for a reusable matrix template and project-specific axis examples.