A real cold stranger must pay before you build. The verdict is recomputed from your raw receipts. An honest NO is a first-class outcome — confidence on what to build, and the cheaper, rarer confidence on what not to build, before you sink weeks into the wrong app.
/launchthesis:discover <seed>
Frame the concept and sharpen the ICP and candidate wedge. D0–D1 triage can return a NO-GO in minutes.
Fan out need, demand-intensity, WTP, and red-team research — then name the wedge as a versioned, first-class object.
A GTM + conversion plan that arms the sprint — weight families, tier the plays, flag the ones the brand should decline. Not a build plan.
A 48-hour guerrilla sprint gated on real cold-stranger pay-proof. You build the page; the verdict is recomputed from the raw rows.
A real cold stranger must pay before you build. LIVE-mode Stripe only — sandbox pay-proofs are rejected as not measurable. Demand you can deposit, not applause.
The verdict is recomputed from your raw receipts with
gate-eval. You can't grade your own
homework, and the tool won't soften the signal to make you feel good.
A NO-GO is a first-class outcome, recorded cheap with the adjacent wedge to try next. Killing a dead idea in a day is the product working.
AI writes the code now and a working prototype is an afternoon. The expensive mistake is no longer a bad build — it's weeks poured into an app nobody wanted. LaunchThesis does the part you skip.
This is for the vibe coders who ship fast with Claude Code, Cursor, Lovable, v0, Bolt, Replit — and have already launched something that got no users. The trigger isn't optimism. It's a past loss.
You're strong at building and rarely run disciplined validation. LaunchThesis does the part you skip — the research and the un-gameable scoring — and asks you for only the parts you're good at: the idea, the audience, and a quick landing page.
Sit in enough doomed launches and you stop seeing bad engineering and start seeing one pattern, fatal and universal: products shipped on enthusiasm, defended by sunk cost, kept on life support long after the market had already voted no.
That is not a skills problem. It is a nerve problem — and it has buried more brilliant people than incompetence ever will. So the honesty floor and “the builder is not the scorer” stay non-negotiable. The tool will not soften the signal to make you feel good.
A GO emits a Validated Launch Thesis handoff — the artifact you paste straight into your AI builder, carrying the verdict, the validated wedge, exactly what the market paid for, and what to not build. LaunchThesis does not own your build, your tests, or your tickets.
Anyone can ship now. Almost no one can decide. LaunchThesis is for the few who'd rather win.
:setup
Run once per project. Detects your stack and writes the config. No app required — an idea is enough.
:discover <seed>
Frame → triage (can NO-GO fast) → demand pulse → red-team. Emits a versioned Launch Thesis + named wedge.
:strategy
Weight families, map and tier the plays, flag the declined ones, wire metrics. A GTM + conversion plan, not a build planner.
:validate
You build the page; it polls and recomputes the verdict from the raw rows. On a GO, emits the AI-builder handoff.