Built on your data, checked, and run once more before you see it.
When no analysis in the library fits your question, one is built for it. Your data is profiled, the analysis is written in R and run on your data, code checks the results, a separate pass reads them against your question, a repair step fixes what fails, and the finished analysis is run once more after it is deployed.
Walk the steps ↓ The product overview →
Prebuilt analyses live in the library.
The steps, in order
These are the steps a build goes through today, named as the build tracker names them while you watch.
Exploring your data
Your dataset’s real values are profiled before any analysis is written, so the work starts from the data you actually have.
Your estimate becomes the brief
The estimate you reviewed sets the brief: each place in the report, the question it must answer, and the method. The estimate’s numbers are not carried over; they are computed again from your data.
Computing the core analysis
An analyst agent maps your columns and writes R that computes a result for every place: a set of named numbers, a table, or per-row data. This is the code that runs again when you run the analysis on new data.
Checking the analysis
Code confirms the R ran, produced an answer, and holds a result for every place, or records why a place could not be computed.
Evaluating the analysis
A separate model pass reads a profile of the results against your question and each place’s question, and checks that the numbers agree with each other and with your data.
Fixing the analysis
A repair step reads what failed and changes the R, and the result goes through the checks again.
Laying out the report
Each result is laid into its card and every card is checked against the card format. A plain-English sentence is written under each card from the numbers it holds; a sentence that states a number its card does not hold is removed when the check finds it.
Running it once for real
The analysis is saved to your account and run end to end once. If that run fails, it goes back to the repair step.
What you keep
A built analysis is yours. It stays in your account as a report and as an analysis you can run again.
The report
Each place from your estimate as a card, with its numbers, its chart or table, and the sentence written from those numbers.
The method
The method each card used, named, so the report can be cited and checked.
The code
The code behind the report is shown on the report itself.
Run it again
Run it again on new data, from your account or from your AI agent.
When a check fails
A failed check is not handed to you. It goes to the repair step, and the repaired analysis goes through the checks again.
The check says what failed
A code check records the place and the problem; the evaluation pass writes down its reasoning.
The repair step changes the R
It reads what failed and why, and changes the analysis to answer it.
Checked again
The repaired analysis runs again and goes back through the same checks.
Repairs have a limit
Repairs are limited to two rounds. A build that still fails ends as a failure rather than being delivered, and failed builds are never billed.
Bring a question
Bring a CSV and one question. Review the estimate, watch the build live, and get back an analysis you can cite, share, and run again, charged only if the build succeeds.
Create your analysis →Where to start
One email with your data and your question. We take it from there.
Cymple
Data Scientist
Send me your data and question, I’ll send you the analytics.
Try it on your own data
14 days free.
Answers, analytics and custom analytics. No card, no call, nothing to install.
14 days free. →