Frequently asked questions

What is WhittleOS?

A decision OS for solo and indie founders. It answers the one question that saves you months: is this idea worth building, worth testing first, or worth dropping? It either checks an idea you already have, or sources real candidates for you and hands back a short list ranked against your founder profile — each with a clear Build it / Test it first / Drop it call and the reasons behind it.

Why not just ask ChatGPT or Claude to do this?

For a second opinion on one idea you already have, a chat is fine — and free, and with search on it will cite pages too. The difference is what happens when you don't have the idea yet, and what happens after the answer. One Discovery run plans the market into sub-markets, runs live searches, reads the pages it is allowed to read, considers up to 150 candidates and keeps at most 5 — and the problems it keeps carry the link they came from, so you can check them, or a not-verified label when we could not find one. Here the checklist, the weights and the threshold are fixed in code rather than decided per answer — the model rates each area and never sets the total — and each run checks against the problems every earlier run found. A chat hands you prose, and whatever its memory keeps is context rather than a record; a run hands you a short list, a gate that said no to the rest, and a decision you can hold yourself to. You are paying for the looking and the throwing away, not for the writing. Here is a complete real run, no signup: /sample-discovery.

How is it different from other idea validators?

Most validators grade an idea you paste in and skew toward encouraging you. WhittleOS does two things they don't: it can source the ideas itself, across real markets, and it's deliberately kill-biased — built to find the reason an idea should die early. The Whittle Score is computed in code from a fixed weighted rubric and a fixed decision matrix; the model rates each area and supplies the evidence, but the weighting, the total and the gate are code — it never hands us its own overall number. That's why the call is auditable — you can see which check produced it — rather than a vibe.

What is the Whittle Score?

The Whittle Score is the 0–100 every idea receives: a language model rates 7 areas of the idea, each rating is multiplied by a fixed weight and added up in code, and the total is banded into a letter grade from A to D. The model supplies the ratings and the evidence behind them; it never sets the total. The arithmetic is fixed; the ratings underneath it are a model's and do move between runs, so the Whittle Score is an auditable number — you can see which area and which weight produced it — not a promise that two runs return the same one.

Does it just tell me what I want to hear?

No — that's the whole point. Most ideas should die early, and a tool that rarely says no isn't protecting you from the three-month mistake. When a market we search is genuinely weak, WhittleOS hands back nothing rather than manufacture a promising idea. 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 — a record of refusals, not proof they were right, and most of those runs are our own. We'd rather lose your trust by being blunt than keep it by flattering you into a build.

Do I bring an idea, or does it find one for me?

Either. If you already have an idea, run a check and get a Build it / Test it first / Drop it call with the reasons, plus focused deep-dives (Kill My Idea, Pricing Power, Support Burden, Platform Risk, Pivot Wedge, and more). If you don't, point Discovery at a market — or at nothing at all — and it sources real candidates, drops the ones it cannot tie to a problem it actually found, and ranks what's left against your profile.

Will it make me get on customer calls?

Never. WhittleOS is no-call / async by design. When it's time to validate, the plan is a landing page with a real CTA — a waitlist, pre-order, or paid pilot — plus async outreach that fits your stated time budget. It never tells you to go book 20 discovery calls.

How does it decide: Build it / Test it first / Drop it?

An idea you bring passes through 12 deal-breaker checks — money, repeat revenue, pricing power, reach, support load, platform dependence, whether a real product survives without the AI, founder fit, compliance, async selling, payback inside a year, and whether it could ever be sold on — and then a scored rubric. All 12 are written out on the How it works page, and you can watch them applied one at a time on our sample report. A hard failure on one is enough to drop the idea. Discovery, where we source ideas for you, uses a related list of 10 published beside them: the lists differ because a sourced candidate has to prove a documented problem exists first, and an idea you hand us doesn't have one to prove yet. Ratings and checks are combined by a fixed decision matrix: given the same check results and the same area ratings, the call is always the same. The ratings themselves come from a language model and do vary between runs, so re-running an idea can move it — what does not move is the standard it is judged against, or who does the judging. The AI supplies evidence and reasoning; the math and the gate are code. When there isn't enough evidence to decide, it says so ("On hold") instead of guessing.

Where do the ideas and evidence come from?

From live sources — real search results and pages fetched at run time — not model memory. Every problem we cite says where it came from: a page we opened and read, a search result we saw but couldn't open, or text you pasted. Where we have a source we show a link you can click. When we can't find one, we mark it "not verified" rather than invent it.

Is it accurate? Can I trust the verdict?

We don't claim accuracy, because no one can — there's no ground truth on whether an unbuilt idea would have sold. What we do claim is that we show our work: the score is computed by a fixed rubric, the decision by a fixed matrix, the evidence is sourced and labelled with where it came from, and every step is auditable. The model's ratings vary run to run, as any model's do — the rubric and the gate they feed do not. Treat it as a rigorous, kill-biased second opinion that shows its work — not an oracle.

How does pricing work?

Prepaid credits — you pay per decision, not per month. No subscription and no sales calls. Signup includes a couple of free credits, meant for a founder read and a first idea check before paying anything, and credits never expire. A Discovery run is the one thing that comes out of a paid pack: a run searches live sources for tens of minutes, so the price follows the work (2 for the first market, +1 per extra). If you want to see a run before paying, we publish a full real one at /sample-discovery. See the pricing page for the current packs.

What if a run fails, or the result isn't useful?

If a run fails on our side, its credit is refunded automatically — you're never charged for a broken run. Unused credit packs are refundable within 14 days. Payments are handled by Polar as our Merchant of Record; full details are on the Refunds page.

Is my idea kept private?

Your ideas and founder profile are yours. We don't sell them, and we don't publish them. Running an analysis does send your inputs to the providers that do the work — our AI provider, and for a Discovery run the market you named goes into the search queries we issue. Nothing else, and nobody trains on it. See the Privacy page for the full list of who receives what.

Who is it for — and is it production-ready?

It's built for solo and indie founders who ship alone and can't afford to build the wrong thing. It's live today — you can get a free founder read right now — and we're honest that we're early-stage: no logo wall or testimonials yet. We'd rather show you a real, dated run of our own product getting killed by our own gate than a wall of unverified numbers.

Have a question about validating an idea rather than about us? The answers library takes them one at a time, each grounded in what our own runs produced.

Comparing tools rather than reading objections? WhittleOS vs ChatGPT, WhittleOS vs IdeaProof and WhittleOS vs FounderPal, WhittleOS vs IdeaBrowser and WhittleOS vs DimeADozen go feature by feature, our gate run against eight ideas with known outcomes shows the decision layer working on history, and the guide to startup idea validation tools covers the wider field.