When is a literature search finished?

The number of relevant papers is a property of the question, not of the effort. A narrow question about a specific assay in a specific organism may have nine relevant papers in existence. A question about an established intervention may have six hundred. Collecting forty tells you nothing about which situation you are in.

Worse, a count is satisfiable by searching badly. Forty papers from one database using one set of terms is easier to reach than fifteen from a search that actually spanned the literature, and the first is far less complete.

Saturation is measured, not felt. The test is simple and takes about twenty minutes.

1. Choose a path you have not used. A different database, vocabulary harvested from your most recent finds, a forward citation sweep from a paper you added late, or a scan of one relevant journal.
2. Run it properly — not a token query, a real one.
3. Count only genuinely new items that pass your inclusion rule. Duplicates and near-misses do not count.
4. Record the path, the date and the count.
5. Repeat with a different path. Two or three consecutive zeros is saturation. A single zero is a weak query.

Record the zeros. A log showing three distinct paths each returning nothing new is the evidence that supports the completeness claim. Without it, saturation is an assertion.

All three share a shape: they are observations about one path being exhausted, treated as evidence about the whole literature. Saturation requires the new path to be genuinely different from the ones already run.

Some questions do not saturate on a realistic budget. A broad question in a large field will keep producing relevant work indefinitely, and continuing to search is the wrong response.

The right response is to narrow the question until the boundary is defensible, then saturate within it and state the boundary explicitly. A search reported as complete within a stated scope is credible. A search reported as complete over a scope it could never have covered is not.