Research
Deep research
Fan out, collect sources, and make the machine cite. A conclusion without a trail is a guess with confidence.
Deep research with a machine is a fan-out you direct: multiple search angles, sources collected and read, claims cross-checked, a synthesis with citations at the end. The failure mode is the fluent survey, paragraphs that sound like knowledge and trace to nothing. The fix is structural: demand citations as you go, spot-check the ones that matter, and treat any conclusion that cannot produce its trail as a guess wearing a bibliography.
Vagueness is the tell. When the summaries stay general, the sources are usually thin, stale, or imagined; specificity is what real reading produces. Push until the answers name papers, dates, and numbers you can open, and remember that the machine grades its own sources generously. The seventh law applies to bibliographies too: spot-check before you build on one.
How to practice it
Fan out on purpose.
Name the angles: the survey, the skeptics, the primary data, the recent work. One search path is one bias.
Citations or it didn't research.
Every load-bearing claim carries an openable source. Demand the trail during, not after.
Spot-check the keystones.
Open the two or three sources the conclusion stands on. A fabricated reference reads identically to a real one.
Push past vague.
Generalities signal thin sources. Ask for the specific paper, the specific number, the specific year, until it appears.