What you're actually buying,
answered plainly.
No jargon, no hedging. If something here turns out to be wrong, that's a bug and we want to hear about it — this is the page people quote back at us.
| An answer costs | $2 – $25 |
| You see the price | before it runs |
| Failed builds | never billed |
| Asking again later | half price |
| The code that ran | yours to download |
Cost
What does an analysis cost?
Three prices, and you see the price before anything runs:
| Snapshot — $2 | one chart, one verified insight · about 2 minutes |
| Brief — $5 | the answer presented on one page · about 7 minutes |
| Deck — $25 | a full commissioned report · 30–45 minutes |
Larger volumes, teams, or running inside your own network are priced by contract — that's here.
Do I need a card to start?
No. Signing up gives you 500 welcome credits, which covers your first Brief. Credits never expire, and there's no subscription to cancel — you can pay per answer for as long as you like. How credits work →
What happens if the build fails?
You are never billed for a failed build. If an analysis can't be made correct, it isn't delivered and it isn't charged.
Do I pay full price every time I ask again?
No. Re-running an analysis you already own on new data costs half the build price, every time. The expensive part was building it once — after that you're paying for the compute, not the thinking.
Trust
How do I know the answer is right?
The numbers are computed, not generated. Your data goes through real statistical code running in an isolated container, and the words in the report are written afterwards, from the computed values — never instead of them.
Every analysis passes automated checks before delivery. A Deck is additionally reviewed by an independent verification pass before it reaches you. How the work gets built →
Can I see the code that produced my numbers?
Yes. The R source that produced the result is printed inside the report, and there's a Download analysis.R button on it. The file you download is the code that ran — not a reconstruction of it. Reproducibility →
Can I cite it?
Yes — one click for APA, MLA, Chicago or BibTeX, on every report, with a permanent link that opens without a login and a PDF export. Citations →
How is this different from asking ChatGPT?
A language model predicts a plausible answer. It doesn't run the statistics on your data, so there's nothing to check and no reason the number will be the same tomorrow.
A peer-reviewed evaluation put ChatGPT at roughly 33% accuracy on inferential statistics with everyday prompts.* Microsoft tells Copilot users not to rely on it for work “requiring high accuracy or reproducibility.” That's the work we exist for. The full comparison →
What if the report is wrong?
Flag it on the report itself. We diagnose it, fix it, and resend at no cost. If it still isn't right, you can request a refund from the report page — reviewed case by case.
What you get, and keep
Do I own the analysis afterwards?
Yes. It's a re-runnable analysis in your account, not a one-off answer in a chat thread. Point it at next quarter's file and it runs the identical method — fixed seeds, pinned versions, isolated container — so a change in the answer means a change in your business, not a change in our pipeline.
How long are my reports kept?
It depends on the plan: 7 days on free, a year on pay-as-you-go, and permanently on a subscription. A report you've cited is kept permanently regardless of plan — we don't delete research someone has committed to.
Can I share a report without giving someone a login?
Yes. Reports have a permanent link that opens without an account, plus PDF export. The method travels with the report, so whoever you send it to can check the working rather than taking your word for it.
How big a dataset can I use?
Tens of thousands of rows on the standard plans, and substantially more on business plans. If your file is larger than your plan allows, you're told before anything runs — not after you've paid.
Getting started
Do I have to leave my AI assistant to use it?
No. It connects over MCP, so you can ask from Claude, ChatGPT, Cursor or any MCP-capable assistant and the analysis comes back in the same conversation. There's also a browser account for your datasets, billing and your report library. Connect it →
What do I actually need to bring?
A dataset and a question in plain English. No modelling, no setup, and no choosing a statistical test — working out the right method is the part we do. How it works →
Can I try it before paying anything?
Yes — the free tools run a real analysis on your CSV with no account at all, and signing up adds 500 welcome credits on top.
*Accuracy figure: peer-reviewed evaluation of ChatGPT's statistical analysis, JMIR 2025.