Look for one that can say no and shows you when it did, not one that says no to everything. WhittleOS publishes how often its checks turned candidates down, keeps the reason for each, and can hand back an empty shortlist. Its Kill My Idea mode is built to argue the case against, so treat it as a list of reasons, not a verdict.
A tool that always says no is no better
The question has a trap in it. A validator that tells everyone their idea is promising is useless, and the fix is not one that tells everyone to quit. Both answers come out the same whatever you type, so neither tells you anything about your idea.
Our own adversarial mode is the example. Kill My Idea is built to argue why you should not build something: ranked reasons, each tied to a named check or a scored area, and what would have to be true for the verdict to change. It can return another verdict for a strong idea, but it leans hard against: in our benchmark, on an earlier version of its prompt, it said "Drop it" on all 8 ideas. That makes it a good way to find the reasons and a poor way to decide.
So the useful question is not how harsh a tool sounds. It is whether its no depends on the idea.
Four tests you can run on any validator
First, give it an idea that already failed, described as it looked at the start, and one that worked. If both come back promising, its yes carries no information; if both come back dead, neither does its no. We ran that test on ourselves on 2026-06-23, with 8 companies whose endings are known, under two founder profiles. For a solo founder our idea check dropped 4 of 5 failures and 1 of 3 successes; for a funded team, a deliberately softer lens, it dropped 2 and 0. A verdict here is about the founder too, so a success that needed a funded team can still be the wrong idea for one person. These are hand-picked cases with hindsight in them, so they prove no accuracy, and we claim none.
Second, ask how often it says no, and whether it publishes the number. Ours is a census: across 33 real Discovery runs between 2026-06-02 and 2026-09-08, the checks turned down 895 of 1,728 candidates and kept the reason for each. It is a record of refusals, not proof the refusals were right, and most of those runs are our own.
Third, find out who sets the total. A model that writes its own score can talk itself into a pass. In WhittleOS a model rates each area, and the Whittle Score is computed from those ratings in code, with fixed weights.
Fourth, ask whether it can return nothing. When a market we search is genuinely weak, a Discovery run hands back an empty shortlist with the reasons, instead of dressing up the least bad candidate.
What WhittleOS does with a no
An idea you bring passes 12 deal-breaker checks and then a scored rubric. A hard failure on one is enough to drop it, and the report names the check, so you can argue with the reason rather than with a mood.
A sensible order is the check first, for the call, and Kill My Idea after it, when you want every reason against an idea you are tempted to keep. The check costs 1 credit and Kill My Idea 1 credit; the 2 free credits a new account gets cover both.
What a no cannot tell you
Whether the idea would have sold. Nobody has ground truth on unbuilt ideas, so no tool in this category can be measured for accuracy, ours included. A no here means the idea failed a check for your situation or, for an idea we sourced, that we found no recorded problem behind it. It does not predict the future, and a founder with a different profile can get a different call on the same idea.
Common questions
Is a harsher validator a better one?
No. A validator is useful when its answer changes with the idea. Harshness that applies to everything tells you as little as praise that applies to everything. Look for published refusals, named reasons and score weights that do not move between runs, not for tone.
Does Kill My Idea ever recommend building?
It can, when an idea is genuinely strong, but it is built to lean against the idea: in our benchmark, on an earlier version of its prompt, it said "Drop it" on all 8 ideas. Use it for the list of reasons and for what would change them, and use the idea check when you want the call itself.
Can I test this without paying?
Yes. A full idea report and a complete Discovery run are published on this site with no signup, and the benchmark lists every idea we tested with its verdict. A new account gets 2 free credits, enough for an idea check and Kill My Idea on one idea.
See it rather than take our word
Everything this page describes is published on this site, with no signup — a complete run, a full report, and the checks that produced them.