The analysis layer
The numbers behind the verdict.
This is not a dashboard. There is nothing to configure, no metrics to arrange, no charts to interpret at the end of a long day. It is the working underneath one decision — the receipts, the coverage, the margin arithmetic — shown so you can check it rather than trust it.
Why show the working at all
An AI answer is easy to produce and hard to trust. Confident prose costs nothing to generate, which is exactly why it should not be enough to move your money.
So the analysis is built to be checkable. Structured source receipts are produced by the engine itself, independently of anything the model writes, and they record what was actually consulted. If the prose and the receipts ever disagree, the receipts are the record.
What a source receipt records
One receipt per source, on every completed research, private to your account.
- Source
- Which source was consulted — TikTok organic search, AliExpress, Google Shopping, Google Trends, Amazon search, or the public web.
- Access
- Whether it came from a dedicated live integration or from ordinary public-web research. These are never blurred together.
- Status
- Succeeded, failed, skipped, or unavailable. A source that did not run is shown as a source that did not run.
- Checked at
- The timestamp of the call, so you can tell a fresh read from one you ran last month.
- Result count
- How many records the sample actually returned — the denominator behind any average you are shown.
- Links
- The pages the run drew on, so you can open the underlying listing or article yourself.
- Limitation
- The written ceiling on that source. For example: a sample of listings, not a complete count of suppliers.
Coverage and confidence
Coverage is the plain ratio of checks that completed against checks that were planned, alongside counts of what failed, what was skipped as irrelevant, and what was unavailable. That ratio sets the run's confidence, and the run is marked complete or incomplete on its face.
Most of the planned checks ran and returned usable results. The verdict is standing on real coverage.
Enough ran to say something useful, but there are gaps. Read the trail before you treat a number as settled.
Too much is missing to be confident. The run says the evidence is incomplete, avoids false precision, and leans towards Caution.
A failed source is never quietly swapped for a weaker one. If public-web research stands in for a direct check, the substitution is labelled and the missing direct evidence is named.
The margin reality check
Most products that look good die here. A source price of four dollars against a retail price of twenty-five reads like a winner right up until shipping and ad spend are put back in. So the analysis prices them in before it calls anything worth testing.
What a unit costs from the sampled listings, in USD.
The delivery cost to get it to a buyer — the line most first-time margins forget entirely.
A 15–20% assumption, stated as an assumption. It is a planning figure, never observed performance.
Named as assumptions you should set for your own store rather than numbers we can observe for you.
The 15–20% ad-spend figure is an assumption and is always described as one. A rough margin quoted without stating its assumptions is not a margin, it is a wish. If you want to push your own numbers around instead, the free profit margin calculator runs the same arithmetic in your browser, with no account and nothing sent anywhere.
How a verdict is reached
- 01
Numbers are bound to their sample. A price is the price across the listings analysed. A view count covers the videos sampled, not a platform. Every figure keeps its denominator attached so it cannot quietly become a bigger claim than it is.
- 02
Unverifiable stays unverifiable. Any figure the sources could not confirm returns as “couldn’t verify”. That is a real answer, and it is a better one than a confident number nobody can trace.
- 03
Thin evidence lowers the call. When a core source fails, the run states that the evidence is incomplete, drops its confidence and lands on Caution — instead of promoting a guess to a recommendation.
- 04
Nothing is sold as a guarantee. A margin estimate, a market read and generated ad copy are inputs to your decision. None of them is presented as a prediction of sales.
Where this sits
Check the working yourself.
Every research you run keeps its receipts. Plans start at $29 a month for 25 credits, unused credits roll over, and there is no free trial — every plan is paid from day one.