Methodology
How we analyze a niche.
Five stages, sequenced. Each stage narrows the candidate set with a different lens, and the verdict on the surviving niche has been argued for and against by independent passes before it ever reaches your inbox.
Ideate and rank.
A specialized agent generates candidate niches against your declared profile: skills, budget, languages, geographies, hard refusals. The list is then hard filtered against regulated industries we do not touch, and ranked on a multi axis rubric before any further work begins.
Groundability gate.
Before spending on the deep dive, a cheap public web check verifies that the top ranked niche can actually be grounded in primary source material. If the evidence is thin, we refuse the spend, surface three salvage angles you could try instead, and credit the original run.
Grounded research.
A research agent surveys the public web for the surviving niche, collects primary source material, and writes every claim back to a real source URL. Hallucinated citations are structurally impossible. The output is a structured evidence ledger, not a finished report.
Synthesis and verdict.
A synthesis agent reads the evidence ledger and renders the verdict: GO, GO with caveats, or NO GO. Carries a rubric score, a kill condition for every GO, and an honest read of what would have to change for a NO GO to flip.
Adversarial critic.
A separate critic pass rechecks every claim in the synthesis against the evidence ledger. Anything that drifted from its source is downgraded or dropped before the report ever reaches you. The critic is grounded against the same ledger so it cannot invent new evidence to confirm a verdict.
Common questions
How long does an analysis take?
Typically 30 to 60 minutes end to end. The pipeline runs five sequenced stages and surfaces a single grounded verdict; we do not stream half-done reports. You get an email when your analysis is ready.
What sources does the pipeline read?
Live web research — primary sources first (industry studies, government data, founder accounts), then secondary trade media, then forum and community signals. Every claim in the final report cites the URL it came from; sources are tiered by confidence (primary / industry / media / low) and the viewer renders low-confidence sources with a visible downweight.
How is this different from a ChatGPT or Perplexity session?
Single-shot chat tools are great for exploration but not for decisions. Our pipeline runs adversarially: an ideate stage proposes candidates, a kill stage tries to refute them, a research stage anchors the surviving niche to live primary sources, and a critic stage independently verifies the load-bearing claims before delivery. The verdict has been argued against, not just argued for.
What does the groundability gate do?
Before any expensive research run, a cheap preflight check vets whether there is enough public evidence to anchor a real verdict on the niche you submitted. If the gate refuses, no monthly cycle is consumed and you get an honest explanation instead of a confident but ungrounded report.
Can I see the data behind the verdict?
Yes. Every report renders the sources next to the claims they support, the niches the pipeline killed and why, the kill criteria that survived versus failed, and the critic pass's verification notes. Nothing is summarized in a way that hides what was actually read.
What happens if a run fails on your side?
We automatically add a free re-run credit to your account (no monthly cycle consumed) and email you. If the credited re-run also falls short, we escalate to a human-approved cash refund, typically within one business day. Full policy on the refunds page.
What you get
One report per 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.