Methodology
How we analyze a niche.
Nine specialized agents run in sequence. Each pass narrows the candidate set with a different lens, so what comes out the other end has been argued for and against by independent passes before a verdict is rendered.
Broad scan
A specialized agent surveys 20-30 candidate niches across your declared geographies. Filters for the rules you set: compliance exclusions, audience match, capital fit. Output is a candidate set, not a ranking yet.
Skeptic pass
A second agent argues against each niche. Pulls counter-evidence: incumbents, funding moves, regulatory shifts, market saturation. Anything that survives moves forward; the rest get killed and logged so you can see what we ruled out.
Reframe
Looks for hidden angles in the survivors. A niche that fails as 'translation software' might survive as 'compliance workflow for sworn translators.' The pass that finds wedges other generic research misses.
Buyer demand
Measures actual market urgency. Search trends, forum activity, regulatory deadlines, recent funding in adjacent categories. Distinguishes 'interesting topic' from 'people are actively looking to spend money on this right now.'
Adjacent threats
Scans the competitive surface. Who is already shipping into this space, who could enter from an adjacent category, who has the customer relationship that you would need to disrupt. Pre-mortems the obvious failure modes.
Founder precedent
Matches the survivors against your profile. Skills, languages, budget, timeline, what you refuse to build. A niche that requires regulated-industry experience or $200k seed when you have neither gets quietly downranked.
Sub-niche segmentation
Isolates the wedge. Inside each surviving parent niche, the agent finds the specific sub-segment with the strongest combination of urgency, fit, and weak incumbents. The wedge is where you actually start.
Final synthesis
A scoring pass with weighted rubric: demand, fit, timing, moat, threats, competition. Each survivor gets a numeric score and a one-paragraph rationale. The full breakdown is in your report so you can argue with it.
Decision gate
A separate Opus model renders the verdict on each survivor: GO, GO_WITH_CAVEATS, or NO_GO. Carries a kill condition for every GO and an 'if no go, what would reopen' for every kill. Built so the customer can sanity-check the agent's confidence.
What you get
One report per tier cadence. Every claim cites its source. Every niche the system killed is logged so you can see what was ruled out and why. The verdict is the agent's, not ours; you read the rationale and decide whether to build.