Data · Intermediate
Clean data and run repeatable tasks
Convert messy records, check a CSV analysis and test a small batch before running the same task over more inputs.
3 lessons · About 35 minutes including practice · Free to read
What you will learn
- Specify a schema without making up missing values.
- Check charts and insights against a small known dataset.
- Review batch inputs, failures and downloaded results.
Lesson 1 of 3
Define the output schema
Data Formatter separates “Validate only” from “Clean and convert”. Validate when you want a report about the input; convert when you want a rewritten result. Choose a target format and state the field names, types and missing-value rules in Extra instructions.
Keep a copy of the original records. Cleaning should not silently turn an unknown number into zero or drop a row because it is incomplete. Check the resulting structure in the system that will consume it before using it in a larger workflow.
Worked example
Input: name,quantity Notebook,3 Pencil, Target: JSON Extra instructions: Return an array of objects with name (string) and quantity (number or null). Use null for a missing quantity. Preserve both records.
Try it yourself
Convert the input to JSON and compare every field with the original. Then run Validate only on the source to see how the missing quantity is reported.
Check your result
- There are two records in the output.
- Notebook has the number 3, not a quoted string.
- Pencil has null, not an invented zero.
Lesson 2 of 3
Check a small analysis
In Data Analysis, upload a CSV or paste rows with a header. Add a question that names the measure you want to compare. Read the row count, column profile and missing values before interpreting a chart.
Nevatal computes chart values from the uploaded data, while the model proposes insights from a profile and sample. Check the narrative against the numbers: an observed difference does not by itself explain why it happened, and a missing value can change which rows support a comparison.
Worked example
day,orders,revenue Monday,2,40 Tuesday,3,75 Wednesday,1,15 Question: Compare revenue by day and report total revenue. Do not infer the reason for any difference.
Try it yourself
Analyze the sample and calculate the total yourself. Check which day has the highest revenue and whether the response invents a cause.
Check your result
- The file contains three data rows.
- Total revenue is 130 and Tuesday has the highest revenue at 75.
- No claim about advertising, demand or seasonality is invented.
Lesson 3 of 3
Test before batching
Batch Runner applies one selected tool to multiple inputs. Test a few representative items first and inspect the output shape before scaling up. Each item makes a provider request, so choose the size of the run deliberately.
Choose “One item per line” for short standalone inputs, or “Blank line between items” for paragraphs that contain line breaks. Review the item list before running it. Afterward, inspect failed items and download the results; a completed request still needs a content check.
Worked example
Three proofreading inputs, one per line: The notebooks is ready. Please send the draft by Friday. We may extend the review if the team agrees.
Try it yourself
Select Proofread, load the three lines and run a small batch. Compare the corrected sentence, the already correct sentence and the conditional statement before trying more inputs.
Check your result
- The preview contains three separate items.
- “The notebooks is ready” is corrected without changing the subject.
- The conditional extension remains conditional in the downloaded result.
Put it into practice
Repeat one exercise with your own material. Keep the original beside the result, use the lesson’s checklist and revise the input if something is missing.