The Shelf / Strategy / Demand Validation Engine
Demand Validation Engine
A straight verdict on your idea — build it, cut it down, change it, or drop it — before you write code.
The job: get a straight answer on whether an idea is worth building — build it, cut it down, change it, or drop it — from evidence that already exists in public. One sitting. No code written yet.
Idea checks usually fail one of two ways. Either the agent cheers you on ("strong signal, users would love this") and you build something nobody wanted, or it holds you to a bar meant for a company chasing investors — must be unbeatable, must own the market — and kills every idea, including the ones that would have quietly paid your rent. This is the middle. A repeatable gate, tuned by real calls that went both ways.
How it runs
It starts with the cheapest question: is anyone already paying for this? If rivals are visibly earning money, that settles demand in minutes and the rest is skipped. If not, it goes and reads what people have already written in public — what competitors earn, named forums, Hacker News, trade boards, review sites. The instructions tell the agent to knock your idea down, not prop it up, and to come back with real quotes and links, including the strongest thing said against you.
Then the call. A crowded market is not a reason to quit. There are exactly four reasons that are, and the file names them. Free alternatives, easy to copy, a small market — none of those on their own. And once demand passes, one more gate most checks skip: can anyone actually find you? Name the single place your users will come from, and show it already brings people to something like yours. The file carries the real case that made this gate mandatory. An idea passed the demand check cleanly, then died here. That category already had an app with famous founders behind it. They pushed it to their own audience of millions. It had a few hundred App Store ratings to show for it.
What's inside
SKILL.md— every step, the four verdicts, the four reasons to drop an idea, the can-they-find-you gate, and the honesty rulesreferences/demand-dossier-template.md— the fill-in report, with a scoring guide for filling it in honestlyQUICKSTART.md— how to install it in Claude Code, claude.ai and Codex, plus a first prompt to run
Who it's for
Solo builders and small teams picking what to build next. Also anyone who wants the same yes-or-no step every time their agents start on a new idea. If you've ever shipped something nobody used, this is the file that would have told you first.
Why not a free directory download
A free "validate my idea" prompt has never killed anything it should have, or saved anything it shouldn't. Every rule in this one traces back to a real call — including the ones it got wrong and then corrected in the file: ideas killed for being in a crowded market, forum complaints trusted over the money competitors were making. Field-tested. Issued as-is.
FIELD REPORT real output, not a promise
Setup: two real validation runs from our own pipeline (products anonymized), showing the skill's two decisive moves — the Phase 0 shortcut and the discovery-path-gate reversal.
Run 1 — revealed WTP short-circuits the mine (verdict: BUILD)
Idea: a photo-library cleanup app (swipe to keep/delete, reclaim storage).
Phase 0 check: willingness-to-pay was already revealed before any forum was opened. Public app-intelligence revenue estimates at the time of the run: the leading swipe-cleaner incumbent around $1M/month, the category leader around $5M/month, and the top-10 cleaner apps' combined annual revenue estimated near $200M.
The call: demand and WTP are PROVEN — money at that scale settles it; no corpus mine needed. The residual risk named honestly: funnel conversion and store-review differentiation in a crowded category, which are build/GTM problems, not demand problems. Verdict: BUILD, with the wedge and discovery path defined next. The app was built and shipped.
What the skill prevented here: a week of forum mining that would have surfaced hundreds of "just do it manually / there are free ones" comments — the anti-buyer's voice, already outranked by revealed revenue.
Run 2 — demand passed, discovery gate killed it (verdict: KILL)
Idea: a bucket-list travel app (keep and share a life list; no booking).
Phase 3 read: PIVOT — demand looked real. People demonstrably pay for pure list-keeping apps in this space at prices from $4.99 up to $239.99/yr, with no booking function. Thesis adjusted, wedge identified.
Phase 3.5 discovery-path gate: killed it anyway. The category's own influencer-backed entrant — founded by a family-travel brand with millions of YouTube/Instagram followers, promoting their own app to their own audience — had topped out at 679 App Store ratings at the time of the check. A massive built-in audience converting that thin is strong evidence the niche itself is small, not that a new entrant merely needs a better channel. No named channel could show proof it carries demand at viable scale.
The call: KILL, explicitly superseding the Phase 3 PIVOT. Recorded verbatim in the dossier: gate result FAIL, verdict flipped, reasoning attached.
What the skill prevented here: the more dangerous failure — an idea that passes the demand check, feels validated, and ships into a niche where even the best-distributed player can't find an audience.
SERVICE RECORD living gear — updated as the factory learns
v1.1.0 — 2026-08-25
Phase 0 upgraded from a revealed-WTP check to a general cheapest-disconfirming-check-first rule, with two new prior-work checks (sweep your own recent kills for recurring failure modes; read the code of any "head start" before believing it). Discovery-path gate hardened with two rules learned running validations since the first issue: a community channel doesn't count until you've read its own promotion rules (demand threads are not access), and audience-fit is not conversion-proof (the bar is paid conversion via the named channel, not reach). Dossier template updated to match.
v1.0.0 — 2026-07-17
First issue. Ported from the factory's internal skill: sanitized for general use, methodology intact, field report captured from a real run.
Every update ships free to owners — your locker always serves the latest version.
QUESTIONS
Does this replace talking to customers?›
It replaces waiting on interviews to make the call. The verdict comes from what's already public — what rivals earn, what real users say, whether the channel works. Interview questions are included, but you have to ask for them.
Won't the agent just tell me my idea is great?›
It isn't allowed to. The research instructions are written to knock your idea down, not prop it up. The report has to include the strongest evidence against you, with a link. And soft non-answers like 'the signal is encouraging' are banned.
What's the second gate — the one about getting found?›
After the demand check passes, one more: name the single place your users will come from, and show it already brings people to products like yours. It exists because a good product nobody can find earns the same nothing. The worked example in the file is a real one.
Is this only for iPhone apps?›
No. The examples lean toward phone apps because that's where it was tested, but the steps, the four verdicts and the gates work on any idea — software, tools for businesses, content.