Writing and editing
Edits what is there rather than rewriting it into its own voice.
Numbers that were counted.
A spreadsheet, two exports, a month of usage. The difference between this and asking a model to look at your data is that here it is counted - and what was counted is shown, so you can check it.
Three steps.
- 1
Attach the file
xlsx, csv or tsv. Formulas come along; it reads what they compute.
- 2
Ask for a number
A count, a median, a difference - not 'what do you think of this'.
- 3
Check the working
The tool result is in the answer, so the figure can be traced back to rows.
Computed, not estimated.
Attach a spreadsheet and it runs Python on it in your browser: pivots, statistics, regressions. The analysis runs on your computer, and the result goes into the answer.
What it does.
- Reads xlsx, csv and tsv including formulas, and says what a cell computes rather than what it displays
- Median, quartiles, standard deviation, outliers and correlation, worked out and not estimated
- Matches two lists on the columns you name and reports every row that does not line up
- Runs Python in your browser, with pandas, numpy and statsmodels, for pivots and regressions the built-in tools do not cover
- Business accounts can run read-only SQL over their own account data
- Hands back a bar, line, pie or scatter chart, or a new file, with the figures behind it
Where it stops
It does not decide what the numbers mean. A correlation is a number; whether it is a cause is a question about the world, and nothing in a spreadsheet answers it. Where the model has to interpret rather than count, it says which part was which.
Asked like this
Here is last quarter's export. Which cost centres are more than two standard deviations from their own twelve-month average, and by how much?
Which size to use
Medium is enough for reading and counting. Large earns its keep when the question spans many sheets at once.
Compare the sizesQuestions.
Does it actually calculate, or does it guess?
For anything statistical it calls a tool or runs Python in your browser, and that computes it. Asked for a median in prose, a model sorts in its head and lands near it - near the median is a different number.
How large a file can I attach?
Depends on the plan; the limit is shown before the upload. Very large tables are better queried than attached.
Is my data used for anything else?
No training, no analysis, no profiling. It is processed to answer your question and for nothing else.
Other kinds of work.
Contracts and documents
Finds the passages. Does not tell you they are fine.
Developer plumbing
The small exact things, done exactly.