01 · Origin
Why WedgeScout exists
A while back I built an AI reading app.
I did what many first-time founders do. I described it to people and asked what they thought. Friends said it sounded useful. A few hundred beta signups came in by the time we shipped. Almost every conversation was encouraging.
Then the paid version launched, and four or five people bought it.
In hindsight, nothing about that was mysterious. I had been asking whether people liked an idea, which is almost guaranteed to produce a polite answer. I had not studied what the same kinds of people said in public, unprompted, on days when nobody was selling them anything: the threads where they complain about tools they already pay for, explain a spreadsheet workaround, and describe what the problem costs them.
That evidence existed while I was building. It was spread across Reddit, Hacker News, forums, reviews, and comment sections, across more pages than I could read by hand.
WedgeScout is the tool I wanted that year: a bounded way to collect those signals, keep the source trail attached, look deliberately for evidence that weakens the idea, and turn the result into a decision you can inspect.
What people say when you ask them can be polite fiction. What they complain about in public when nobody is selling to them can be evidence.
03 · Method
How a report is made
A completed WedgeScout report follows a bounded research contract. The scope is decided before collection so the result can be audited instead of expanding until it finds a comforting answer.
- 01
Narrow the direction
A broad idea such as “AI for accountants” has no single community or vocabulary. The first step turns it into focused directions with a specific user, a concrete workflow pain, and a risky assumption. You choose which direction deserves the research budget.
- 02
Write the source plan first
Platforms, source categories, query themes, and limits are planned before collection. The plan remains visible so you can see where the report looked and where it did not.
- 03
Collect and grade the evidence
Public source material is normalized into evidence units with provenance. First-hand detail, specificity, independence, context, and corroboration matter more than raw mention counts. A vague complaint and a concrete workflow failure do not carry the same weight.
- 04
Run the counter-signal pass
The query plan includes searches designed to weaken the direction. Relevant counter-evidence is attached to the claims it challenges. If the run does not find a reliable counter-signal, that absence is treated as a limitation, not as proof that none exists.
- 05
Deliver a judgment and trail
Evidence-backed factual claims cite collected sources. Unsupported statements are downgraded to hypotheses or open questions, or removed. The report shows a qualitative signal judgment and keeps the plan, evidence, citations, and limitations inspectable.
04 · Boundaries
Where the evidence comes from, and where it does not
What the source plan prioritizes: public, human-written discussion. Reddit threads, Hacker News comments, and the open web are useful when people describe problems, workarounds, switching resistance, or buying behavior in their own words.
What it does not treat as demand evidence: scraped private data, a proprietary panel you cannot inspect, or a vendor saying its own category is important. Vendor pages can provide competitor context, but they are not independent proof of customer pain. Paid provider fallbacks require explicit consent.
Source-backed findings keep a citation destination and stored provenance. Public pages can still be edited, deleted, or moved after a report is produced. If a source later stops opening, that changes the link, not the evidence boundary the report recorded at collection time.
A snapshot
A report reflects what was publicly visible during a bounded run. Markets move, so an older report is historical evidence.
A posting bias
Public discussion over-represents people who post. Some professions leave rich trails; others barely discuss buying decisions in public.
An incomplete absence
No visible complaints can mean a quiet market, a weak source plan, or a problem people have stopped discussing. It is not proof that no problem exists.
05 · Disconfirmation
How we look for evidence against you
A research plan should not only ask where the pain exists. It should also ask whether the same target user already has a solution they are satisfied with, whether switching is too costly, whether the problem is not worth paying to solve, or whether the platform makes the proposed workflow impractical.
WedgeScout treats that as a distinct query intent. Evidence only counts as a true counter-signal when it concerns the same target user, the same workflow need, and a credible reason the proposed opportunity may not hold. A founder promoting a product or a vendor praising itself is kept as context, not mistaken for independent evidence.
Counter-evidence that survives validation stays attached to the claims it weakens. When none survives, the report records the limitation instead of turning “not found” into “does not exist.”
A research tool that only returns encouragement is not a research tool. It is a mirror.
06 · Decision language
How we judge the signal: qualitative, not theatrical
Reports use one of four signal judgments: Strong, Moderate, Weak, or Inconclusive. The judgment is accompanied by reasons and controls how far the report is allowed to recommend action.
A strong, direct evidence set may support a bounded first wedge. Narrow or indirect evidence should produce an early signal, a conditional product shape, or a clearer next research move. It should not be dressed up as a validated market.
There is no 78/100. A precise-looking number can hand the decision back to the founder while pretending to have answered it. The useful output is a stated judgment, the evidence that earned it, and the condition that would change it.
07 · Limits
What a report cannot do
It cannot predict the future. It can describe what was visible during the run, which is narrower than saying a product will work.
It cannot replace talking to people. Use a report to enter your first conversations with better questions and a clearer view of the communities worth approaching. Our guide to The Mom Test explains how to conduct those conversations without inviting polite answers.
It cannot tell you whether you should build it. Market evidence says nothing about your unfair advantage, your appetite for the problem, or whether you will care about it in year three.
It is not legal, financial, or investment advice.
08 · Contact
Tell me where the evidence breaks
Email about a quote you could not find at its source, a market where the research came back thin and you know why, or a judgment you think we got wrong. The last one is often the most useful message.