Most ideas are not killed by a bad market. They are killed by a belief nobody checked: that people will pay, that they can be reached, that the problem is frequent enough to matter. Validation is the work of finding those beliefs and testing them before they cost you a year.
1. Write down the beliefs your idea depends on
Every idea rests on a handful of load-bearing beliefs. Write them as plain, checkable sentences. Not “people need this”, but “mobile bike mechanics lose time every week rescheduling customers”. A good list has between three and seven beliefs, and at least one of them should make you uncomfortable.
Sort them by what breaks if they are wrong. A belief about who has the problem usually matters more than a belief about which feature to build first.
2. Find a source for each belief
For each belief, ask what kind of source would settle it, then go and get it. People describing the problem in their own words, on forums or in reviews, are strong evidence of pain. Published prices of alternatives are strong evidence of what the market already pays. Public datasets can bound a market, when a category matches; often it will not, and that is worth knowing too.
Keep a ledger as you go: an ID, the source, a link, and what it supports. It feels slow. It is the only way to come back in a month and know which conclusions still stand.
3. Argue against yourself
Once you have a draft conclusion, read it as a sceptical investor would. Which claims do the sources not support? Which citation says something narrower than your sentence? Which risks did you leave out because they were inconvenient?
This is the step people skip, because it hurts. In our own pipeline a separate critic model does it, and in our benchmark of 30 ideas it sent 27 drafts back for revision. Expect the same from yourself.
4. Treat every number as a claim
A number in a plan is persuasive out of proportion to its quality. Before you keep one, check that it appears in a source you can open. If it does not, cut it or replace it with “unknown”.
A sentence that states a number stays only if the number appears in a source it cites. Lodestake’s verifier applies this rule to every sentence; across our benchmark it removed 79 that failed it.
Market-size figures deserve extra suspicion. Different reports often scope the same market very differently; when sources disagree by a factor of five, the right answer is a range with its sources, not the biggest number.
5. List the unknowns
Write down what you could not establish. Typical entries: what customers would pay, how many of them there are, whether a direct competitor already exists, how you would reach the first hundred. This list is not a failure. It is your test plan.
6. Decide, then test the cheapest unknown
With a ledger and a list of unknowns, the decision gets simpler. If a load-bearing belief has no support and no cheap way to get it, walk away or reshape the idea. If the beliefs hold and the unknowns are testable, stake a small claim: pick the cheapest unknown to resolve, often willingness to pay, and test it directly with a landing page, a pre-order or ten conversations.
Where Lodestake fits
Lodestake does steps 2 to 5 as a report: planned questions, fetched evidence with IDs, a critic pass, a revision, a verifier for numbers, and an unknowns section. It is in pilot today. The app already covers ideas, plans, names and domains. See the Demo.