Main idea

In research steered by the state of the corpus, a new iteration isn't given a subject — "find some more relevant articles" — but a goal that comes out of the diagnosis. That goal names the type of evidence that is missing right now:

• broaden: find an important dimension that is still absent;
• reinforce: add independent evidence to an important piece of knowledge;
• fill a gap: document a missing mechanism, step, metric or condition;
• look for contradiction: deliberately find what would invalidate or limit what you believe is solid;
• pin down the limits: establish where and when a claim stops being true;
• separate the competing meanings of a term across authors, products or disciplines;
• move from description to mechanism: understand why and how, not just what;
• diversify the evidence: theory, benchmarks, empirical studies, technical documentation, field feedback;
• test the transferability of an idea from one field to another;
• identify the dependencies of the mechanism;
• look for side effects: costs, latency, bias, regressions, governance risks;
• check for obsolescence in part of the corpus after a technical change;
• connect separate areas that the same mechanism runs through;
• rebalance a conclusion that only looks strong because one type of source is over-represented;
• look for what would break the model;
• measure saturation.

The search engine, or the agent querying it, is no longer there to find good sources, but the type of evidence the corpus needs.

Why it matters

It changes the query itself: looking for a contradiction, a limit or a side effect isn't phrased like looking for a subject, and doesn't bring back the same results.

And it gives you a vocabulary for saying why a source was added — useful to whoever picks the research back up later.

Nuances and limits

The list isn't a procedure to run through: an iteration usually pursues only one or two of these goals, the ones the diagnosis points to.

Some goals pull against each other: reinforcing seeks confirmation, looking for contradiction seeks the opposite. Pursuing both at once on the same claim muddies the result.

Open questions

• Are some types of evidence structurally harder than others to get from a general-purpose search engine — contradiction, for instance?