How one idea gets pressure-tested

Discovery ends with a shortlist of things you could build. Deep Research is what happens after you pick one. Instead of spreading attention across every idea, it takes the single one you selected and tries hard to find the reasons it might not work. The point isn't to cheerlead. It's to surface the objections a careful investor or a skeptical co-founder would raise, while they're still cheap to act on.

Why it's a separate phase

Discovery is broad and shallow by design: it looks across a whole niche to find problems worth solving. (See How the research works.) That breadth is the wrong tool for judging one specific product. A real go/no-go call needs depth on a narrow target — the exact buyer, the real competitors, the channel you'd actually use to reach people. So Deep Research re-runs the research, but pointed at your chosen idea rather than the niche as a whole.

What it checks

It works through the questions that decide whether an idea is worth your next few months:

• Who, exactly. It pins down the specific buyer the idea serves — the tools they already use and what frustrates them about those tools — and feeds that into every later check, so competitors and keywords are judged against your real audience, not a generic version of the niche.
• The competition. It looks for the products already solving this problem, what they do well, and where the gap is that your idea would live in.
• Whether you can get found. For ideas that lean on search to grow, good keyword scores aren't enough. It also asks the harder question: is there a set of pages you could realistically rank for, or is the front page already owned by entrenched sites? An idea can have strong demand and still have nowhere to win — and the verdict reflects that.
• What people will pay. It reads willingness-to-pay signals from real discussion and any competitor pricing, rather than guessing a number.
• How big the slice is. A focused tool wins a small beachhead first — the slice of the niche it actually serves — before it can grow into the wider category, so it shouldn't be credited with the whole market on day one. The serviceable market (SAM) is anchored on your idea's own validated demand — the slice it addresses — not on the whole-niche number. The niche-wide search volume is shown separately as a reach ceiling (the total addressable market): the size of the whole category the idea could eventually expand into, not what it serves now. Outsized top-down figures ("1% of a huge market") are the most over-trusted number in any pitch, so the headline is the beachhead and the big number is kept as the ceiling above it.

The verdict

All of that rolls up into a Go, Conditional, or No-Go call, with a short explanation of why it landed where it did. The explanation uses plain terms — strong demand, weak distribution, crowded field — rather than raw internal numbers, because the reasoning matters more than the score. A verdict can only be lowered by a problem the checks surface, never inflated; a clean Go has to earn it on every axis.

What it can't do

It's still working from public discussion and snapshot data, so it reads the present, not the future. Growth rates and market sizes are estimates with wide ranges, not measured figures, and it says so when confidence is low. And it judges the idea you selected — if a different idea from the shortlist would have fared better, Deep Research won't tell you that. Picking the idea is still your call; this phase tells you how hard the one you picked will be.