Long-form research runs

Some questions genuinely live across dozens of sources — a market's full picture, a regulation's practical impact, the honest state of a technology. For those, the researcher desk has its heavy gear: deep research runs that read widely for minutes and return a structured, cited report. As of September 2026, this capability sits behind its own mode; this lesson is when to fire it and how to get reports worth the wait.

When it earns the minutes. The test is genuine breadth: would a good answer have to draw on many independent sources? Market overviews, regulatory landscapes, state-of-the-topic reads, literature scans before a big move. Not: questions one good source settles (quick sourced answer), not facts (the standard check), not decisions (that's your frame — deep research feeds it, never replaces it).

Brief it like a contractor. A deep run amplifies its briefing — vague in, sprawling out. Give it the full researcher's brief plus structure: scope (time/place/segment), evidence preferences, the three sub-questions that matter, and the report shape you want ("sections per sub-question, each claim sourced and dated, a final section for open disagreements between sources"). The disagreements section is the power move: where sources conflict is where the real information lives.

Read the report like a sourced answer, scaled. Everything from the citations lesson applies, multiplied: scan the source list for who's actually behind the report, open the load-bearing citations, watch for one-origin funnels dressed as breadth. Long reports add one hazard — the smoothing: contradictions between sources get averaged into confident middle claims. That's what your requested disagreements section counteracts; if it came back empty on a contested topic, treat the whole report more skeptically, not less.

Deep runs feed desks. The report is rarely the final artifact — it's the evidence layer for work that happens elsewhere: the careful desk turns it into the client brief, the decision frame consumes its facts, the market scan absorbs it as the motion section. Pipe it explicitly: save the report, then hand it over with its receipts intact ("using only this research report, draft…"). Breadth gathered once, spent many times.

Budget the heavy gear. Deep runs cost the most time and, on paid plans, the most allowance — the aiming rule one last time: a few runs a month on questions with genuine breadth, quick sourced answers for the daily flow. If you're firing deep research daily, either your work is unusually research-shaped or your triage slipped; check which before the allowance does.

One mistake to skip: outsourcing synthesis entirely. A deep report organizes what sources say; it cannot know which finding changes your plan. Read with your stake in hand — mark what surprised you, what contradicts your assumption, what you'd verify before betting. The report is the reading; the noticing is still the researcher's job, and at this desk, the researcher is you.