--- title: "How Preuve AI works and what the viability score means" slug: how-it-works description: "How Preuve AI works: 10 agents pull live evidence from 50+ sources, score your idea 0-100 on six factors, and cite every claim. What the score does and does not tell you." canonical: https://preuve.ai/how-it-works --- # How Preuve works, and what the score means > **TL;DR:** Preuve AI turns a one-paragraph idea into a cited 0-100 viability score in about 60 seconds. Ten agents query 50+ live data sources in parallel, findings are cross-validated across models, and every claim links to a source you can check. About 18.3% of scored ideas reach launch-ready (70+). Preuve AI doesn't rate your idea from training data. Ten agents pull live evidence from 50+ sources, score it 0-100 on six factors, and link every claim to a source you can click. ## How does Preuve AI work? Validation runs as a multi-agent pipeline. You give it one paragraph; it returns a cited verdict in about a minute. No single LLM's opinion at any step. 1. **You describe the idea.** One paragraph: what you are building, who it is for, how it makes money. No deck, no spreadsheet. 2. **10 agents query in parallel.** Each agent owns one domain (competitors, market sizing, demand signals, community sentiment, regulation) and pulls live evidence from 50+ data sources. 3. **Findings are cross-validated.** When two model runs disagree, the verdict reruns until they converge. The score is the convergence point, not a guess from training data. 4. **You get a cited 0-100 score.** Six factors, one number, a source URL behind every claim, the risk flags dragging the score down, and 2-3 pivots that could lift it. The full source breakdown (10 families, audited across 246 paid reports) lives at https://preuve.ai/sources. ## What does the score measure? The Preuve AI Score is a single 0-100 viability rating built from six factors, each checked against the same 50+ live sources, then cross-checked across models: 1. **Market size & growth** - bottom-up TAM/SAM/SOM and whether the market is expanding or shrinking. 2. **Real demand signals** - whether people search for, complain about, or pay to solve this. 3. **Competitive density** - how crowded the space is and whether incumbents already solve it well. 4. **Moat durability** - how easily the idea is copied, and what makes it defensible. 5. **Business model viability** - pricing power, acquisition cost reality, and a path to margin. 6. **Regulatory & execution risk** - legal, compliance, and operational blockers that quietly kill ideas. ## What does your viability score actually tell you? The number is a calibrated read on how the idea looks against live market evidence today. - **70-100 - Launch-ready.** Strong signals across the board. Only about 18.3% of scored ideas reach this band, where most AI validators hand out a 75-85 to nearly everything. - **40-69 - Worth reshaping.** Where most ideas land. The core can work, but positioning, target, timing, or crowding is holding it back. A sharper angle often moves the number 10-20 points. - **0-39 - Strong headwinds.** The market is signaling hard against the idea as framed. Rethink the core, not just the wording, before you write code. ### What the score does not tell you - **It does not score you.** The number rates the idea and its market, not your ability to execute. - **It is not a guarantee.** It is a probability read on public evidence available today. - **It is a snapshot, not a verdict.** Change the target, positioning, or timing and the evidence changes with it. Re-score and the number moves. ## What a low score is really telling you A low score doesn't mean the idea is worthless. It usually means one of these, and each is fixable before you build: - **The market is crowded** - 15+ competitors already doing this. You need a sharper angle than the incumbents. - **The timing is off** - the market may not be ready, or it peaked two years ago. - **The positioning is wrong** - same product, different framing, different score. One idea tested across four angles ranged from 42 to 66. - **The target is too broad** - narrow to a specific buyer and the demand signals sharpen. Every report shows the risk flags pulling your score down, then suggests 2-3 pivots that could add points. ## What is open, and what stays under the hood The exact factor weights, thresholds, and agent prompts are proprietary and stay unpublished. What is not proprietary, and what most tools skip: every factor is scored against live, cited evidence you can open and check yourself. The verdict is auditable even though the recipe is not. ## Common questions ### How does Preuve AI work? You describe your idea in a paragraph. Ten AI agents then query 50+ live data sources in parallel, each covering one domain. Findings are cross-validated across models, and the result is a 0-100 viability score with a source URL behind every claim, plus risk flags and 2-3 suggested pivots. ### What does the viability score actually tell you? It is a single 0-100 viability rating built from six factors. It tells you how the idea looks against live market evidence today. About 18.3% of ideas reach launch-ready (70+), so the number stays honest rather than flattering. ### What does a low viability score mean? A low score is a direction, not a verdict. It usually means the market is crowded, the timing is off, the positioning is wrong, or the target is too broad. The report shows which factor is dragging the score down and suggests pivots that can lift it. ### Where does the data come from? Every scan pulls from 50+ live data sources across 10 categories: regulatory filings, hiring data, market research, funding databases, community signals, search demand, pricing pages, and more. Every claim links back to a source you can check. Full breakdown at https://preuve.ai/sources. ### Is the score just ChatGPT's opinion? No. A raw LLM rates an idea from memory and tends to agree with you. Preuve scores each factor against live, cited evidence and cross-checks the verdict across models until they converge. ## Canonical - HTML: https://preuve.ai/how-it-works - Markdown: https://preuve.ai/how-it-works.md - Sample report: https://preuve.ai/examples - Sources: https://preuve.ai/sources