Lay every study you're reviewing on a grid of the method and the topic that define your field. The clusters show where the papers pile up; the empty cells are the method-question pairs no one has run - the ones a reviewer will ask you about.
Embed in an emergency department for a quarter and record every time a clinician accepts, edits, or ignores the model's ranked differential. Your interviews capture what clinicians say they do with the tool and your trials measure accuracy in the abstract - this is the naturalistic reliance data that sits between them, the study a reviewer will notice is missing.
Pulls Clinicians who override triage advice · Explanation format and diagnostic accuracy · Copilot rollout across 200 engineers
Field Deployment × Clinical Support is empty - ringed by your clinic interviews and your controlled diagnostic trials, yet no one has watched the tool actually used on a ward.
Illustrative sample - names are invented. Your grid is built from your own ideas.
01The problem
A literature review lives in a spreadsheet with forty rows and no shape. You can feel that the field over-relies on lab experiments, but you can't point to the one question everyone has surveyed and no one has ever deployed. So "this is understudied" stays a hunch - and hunches don't survive peer review.
02How it works
1
Paste your corpus, one study per line - AI proposes the method and topic axes.
2
The grid clusters your papers and exposes the thin columns at a glance.
3
Click an empty cell for a defensible study direction, built from the tension of its neighbours.