Startup idea assumption testing
Startup Intelligence: test the assumptions behind your startup idea
It reads your idea and writes a de-risking report: the few assumptions it depends on, how strong the evidence for each one is, where you should stop, and the cheapest experiments to run first.
Check whether an idea has enough to test, right on this page.

The problem it solves
Founders often spend months building before they test the assumptions that decide success. Many idea reports give a score or a confident forecast without showing what it rests on.
What you get instead
Not a verdict on your idea. An ordered queue of the things that could end it, and the cheapest way to test each one first.
The real product
See a report before you ask for one
Real app screens, rendered from a made-up test idea. The numbers on them are test data.



Try it before you sign in
Would your idea have enough to test?
The engine starts by turning an idea into nine claim links. Tick the ones your idea already states in plain words. This runs only in your browser: nothing you tick is stored or sent.
The rule it applies: at least four of the nine links must be stated, and two are always needed, the customer and the mechanism. With fewer, it refuses and names the sentence that would fill each gap.
Tick the links your idea states, and the result appears here.
The report, part by part
What Startup Intelligence does
6 built, 1 partly built, 1 planned. Each line says which.
- Built
Load-bearing assumptions
The report names the 5 to 8 assumptions your idea depends on, each with a grade for its evidence.
- Built
No number without a source grade
Every figure carries a grade and a note on where it came from. The report gives no single score.
- Built
Honest refusal
If the idea has too little detail to test, the engine refuses instead of guessing.
- Built
Outcome range and kill thresholds
A simulation shows the range of outcomes against a threshold you set, and where you should stop.
- Built
Cheapest experiments first
The report lists the three cheapest experiments that would test your riskiest assumptions.
- Built
Web report and PDF
Watch the run progress live, then read the report on the web or download it as a PDF.
- Partly built
Report by email
Email for verification and report delivery is coded but not yet confirmed working on our server.
- Planned
Evidence from real sources
Linking each assumption to published data you can open and check. Today the base figures come from the AI model's own knowledge.
Four steps
How it works
Describe your idea
Write your idea in plain words and set the outcome you need.
Break it down
The engine turns the idea into a chain of claims, or refuses if there is too little to test.
Grade and simulate
It grades each key assumption and simulates the range of outcomes.
Act on the report
You get kill thresholds and the three cheapest experiments to run next.
No number without a rung
Every figure says where it came from
Each number in a report sits on a seven-rung evidence ladder, from a measured figure at the top to an unsourced belief at the bottom. The rung is printed beside the number, not hidden in a footnote, and a weakly evidenced figure is given a wider range: an estimate can never look more precise than a sourced figure.
Where it stands today. No outside source is consulted during a run yet, so the base figures come from the AI model's own knowledge and the report files them on the lowest rungs. Linking each assumption to published data you can open and check is planned.

- P1MeasurementmeasuredA figure someone measured directly.
- P2AnalogmodelledA comparable venture's figure, with the difference stated.
- P3Reference classmodelledA base rate taken from a group of similar cases.
- P4Structural derivationmodelledWorked out from other figures by arithmetic.
- P5Constraint boundmodelledA limit the figure cannot cross.
- P6ElicitationassertedAn estimate someone gave when asked.
- P7Founder beliefassertedYour own estimate, or a figure nobody has sourced yet.
A range, not a score
Set the bar, then see how many futures clear it
A report never gives your idea a score. It simulates the outcome you asked about across thousands of possible futures, shows the spread, and tells you what share of them clears the threshold you set yourself. Move the bar to see how that reading works.
Illustration with made-up numbers. It is not a report, not a forecast and not advice.
298 of 1,000 illustrative futures clear this bar (30%). The shaded band is where nine in ten of them land.
Who it is for
Founders
Before you spend months building, find out which few assumptions your idea depends on and how to test them cheaply.
Startup Intelligence for foundersEarly-stage teams
Before a small team commits months to building, the report shows which few assumptions to test first.
Startup Intelligence for foundersIncubators
Every idea read on the same frame: its claim links, what it rests on, how well that is evidenced and where it would stop.
Startup Intelligence for incubatorsConsultants
A structured first look at a client's idea, with the evidence behind every figure on the page.
Startup Intelligence for consultants
Where you can use it
- WebAvailable
- AndroidIn development
Honest status
Running on our server. New accounts are by invitation only for now.
Today: 6 built, 1 partly built, 1 planned. The status below is read from the status service.
Pricing
Pricing will be announced at launch. Nothing is charged for now.
Questions people ask
Does the report include market research or a five-year financial model?
Not today. The report covers assumptions, outcome ranges, kill thresholds and experiments. Whether to add those deliverables is still being decided.
How long does a report take?
A full run usually takes a few minutes. Runs are handled one at a time, so a queue can make it longer, and you can watch each stage while it works.
Is this investment advice?
No. The report says it is not advice. It gives no single score: it shows ranges, thresholds, the assumptions behind them and the experiments that would test them.
Where do the numbers in a report come from today?
Today the base figures come from the AI model's own knowledge. No outside source is consulted during a run, and the report files those figures as the weakest kind of evidence. Linking each assumption to published data you can open and check is planned.
What happens if my idea is too thin to test?
The engine refuses instead of guessing. It names the parts of the idea that are missing and the sentence that would fill each one, and a refused run is not counted against you.
How do I get access?
New accounts are by invitation only for now. Call us to ask for access. Please do not describe your idea in a first message.
Talk to us
Ask for access, a walkthrough or anything else. New accounts are by invitation for now.
Talk to a person
Online forms are not open yet. Call and a person will answer; nothing is sent from this page.
Please do not describe your startup idea in a first message: the idea itself goes only into the signed-in app.
