Understanding the verdict
The verdict runs on Opus 4.7 with an extended-thinking budget, reading every completed section before producing a final score, a 4-way category, and a one-paragraph blurb. Of all the sections, this is the one founders bet months of their life on — so it gets the strongest model and the deepest reasoning pass.
The math works, the timing is right, the founder has an edge, and the first experiment is obvious.
BUILD_NOW means the report found a clear problem with willing buyers at the price your unit economics need, a defensible angle on the competitive landscape, and tailwinds you can ride. It does not mean the idea is risk-free — it means the known risks have known mitigations and the next move is to ship, not to keep researching.
The core insight is real, but one specific element needs to change before the build is justified.
Most early-stage ideas land here. The market is there, the founder fit is there, but one slice — the wedge, the buyer, the channel, the pricing — is wrong as currently framed. The roast section and the validation experiments will name the slice. Iterate on that slice, then re-run validation.
Multiple foundational pieces are weak. More research before any code, or a meaningful rethink.
When the report flags weak demand signals, fuzzy personas, or unit economics that only work at scale you have no path to — this is the verdict. Founders who push past RECONSIDER often spend 6-12 months building something nobody buys. The fastest move from here is the customer-interview-plan section: 10 buyer conversations beats 10 more weeks of building.
Structural problems the founder cannot fix alone — saturated market, missing capability, illegal in target geography.
DO_NOT_BUILD is rare and the prompt is calibrated to use it sparingly. When it appears, read the risk register and the legal section first — those are usually the deciding signals. A DO_NOT_BUILD verdict is not the end of you; it is the end of this specific framing. The founder-market-fit section often points to an adjacent problem worth pursuing instead.
Why the 30 / 30 / 25 / 15 target
A validation tool whose verdicts cluster too heavily on BUILD_NOW is broken — that is what feel-good consultants do, not what real validators do. The target distribution across a healthy sample of 50+ ideas is roughly 30% BUILD_NOW, 30% BUILD_WITH_PIVOT, 25% RECONSIDER, 15% DO_NOT_BUILD. When the live distribution drifts more than 8 percentage points off target on any verdict, the verdict prompt is showing rot and gets re-calibrated.
Admins can see the live distribution at /admin/idea-validation with a drift indicator per verdict. The score is not the only thing that determines the category — the verdict prompt also reads the risk register and the validation experiments to decide between adjacent categories near a boundary.
The override rule
Founders can build any idea regardless of the verdict — even DO_NOT_BUILD. The build button stays live on every report. What the verdict does is anchor the conversation: if you push past a DO_NOT_BUILD, you should be able to point at the specific section that the verdict misread and why. The verdict is a forcing function for honesty, not a vetoer.
When to re-run
Validation reports go stale fast. Re-run when: the idea framing changes meaningfully, the market moves (a major competitor exits, a regulation shifts), or it has been more than 90 days since the last run on a still- unbuilt idea. Re-running creates a new Idea linked via supersededByIdeaId so the history is preserved.
Cost of running: see credits & per-section pricing.