What happens in the few minutes a report is being built

The nine stages of a Startup Intelligence run, from your idea to the finished report: what each stage does, why runs go one at a time, and what you can watch.
From a description to a report
You write your idea in plain words and set the outcome it has to reach. A few minutes later you have a report. In between, the engine works through nine stages, and the run's own page shows each one as it completes. Here is what happens, in order.
The nine stages
1. Decompose
The idea is broken into nine claim links: the customer, their job and pain, what they do today, the mechanism, the value created, the value captured, the path to the customer, the cost to serve and what persists. If fewer than four links are stated, or the customer or the mechanism is missing, the run stops here and says which sentences would fill the gaps. Stopping at this stage means the expensive work never starts, and the refused run is not counted against you.
2. Extract
For each link, the assumptions it depends on are written down one by one.
3. Type
Each assumption is placed by two things: how much the outcome leans on it and how well it is evidenced. The heavy, weakly evidenced ones form the "kill zone" that gets attention first. The placement is provisional: after the simulation has measured what actually moves the outcome, every assumption is placed again.
4. Price
Each assumption gets a range and a rung on the seven-rung evidence ladder, from a measured figure to an unsourced belief. A weakly evidenced figure gets a wider range. Today no outside source is consulted during a run, so the base figures come from the AI model's own knowledge and are filed on the lowest rungs. Where no figure can be given, the gap is marked for you to fill rather than invented.
5. Assemble
The assumptions are joined into one model of the outcome you asked about, working back from your threshold to the quantities that would have to hold.
6. Propagate
The model is simulated across thousands of possible futures, each drawing its assumptions from their ranges. This produces the spread of outcomes and the share of futures that clear your bar.
7. Kill lines
For each assumption, the engine finds the value at which the plan stops clearing your threshold, and how likely it is that the truth is already on the wrong side.
8. Structural reading
A second reading asks four questions that a simulation cannot answer: why now, what compounds, what is melting, and whether the downside is bounded while the upside is open. It is asked to judge the structure on its own terms, without regard to whether the numbers clear your bar, so it is allowed to disagree with the arithmetic. When the two disagree, the report says so and treats the disagreement as a finding. We explain this stage in why now: the question a simulation cannot answer.
9. Emit
The last stage designs the experiments that could end the idea fastest and keeps the top three, each with the result that would kill it stated as a number. It also writes a few adjacent versions of the idea, each aimed at a specific weak assumption, and records the run's stated ranges so they can be checked later. The report is then put together page by page, with every figure carrying its rung, on the web and as a PDF.
How long it takes, and why one at a time
A full run usually takes a few minutes. Runs are handled one at a time on our server, so if several people start runs together, a queue can make yours take longer. The run page shows progress while you wait, and the same address opens the finished report.
What you can watch
While a run works, its page shows the stages completing and fills in the findings established so far. When it finishes, the report opens at the same address. If the run was refused, the page shows the refusal with the missing links instead. Email delivery is coded, but no mail service is connected on our server yet, so the page and your runs list are the reliable places to find a report today.
Takeaway
A run is a sequence of small, checkable steps: decompose, extract, type, price, assemble, propagate, find the kill lines, read the structure, write the report. The first step can refuse; every later step keeps its evidence visible. See the walkthrough on the how-it-works page or read when the engine refuses.
Reports support planning. They are not investment advice and not a guarantee of results.
Key takeaways
- A run has nine stages, from decomposing the idea to writing the report.
- It can stop at the first stage if the idea is too thin, before the expensive work.
- You can watch each stage complete on the run's own page.
Questions
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.
Can I close the page while it runs?
The run continues on our server, and the same page shows the report when it is done. No mail service is connected on our server yet, so come back to the page or your runs list.
Startup Intelligence: what it does



