Try the time-validation demo¶
This tutorial walks you through the time-validation demo — a synthetic energy-consumption dataset with intentional time-series errors planted in it — so you can see what datasight's quality checks catch. Allow about three minutes.
What you'll load¶
The demo generates roughly 70 MB of synthetic hourly data with a mix of realistic values and planted issues:
- missing hours (gaps in the hourly series)
- duplicate timestamps
- DST spring-forward and fall-back artifacts
- leap-year inconsistencies across groups
A time_series.yaml file ships with the demo declaring the expected
temporal structure, so datasight quality knows what "correct" looks
like.
1. Install datasight¶
Don't have uv yet? See Install datasight.
2. Generate the dataset¶
Run datasight demo time-validation --help for generator options
(dataset size, planted-error rate, seed).
3. Detect the planted errors¶
The output includes Time Series and Temporal Completeness sections listing each gap and duplicate the generator planted.
4. Explore interactively¶
Add an API key to .env (see Set up a project)
and launch the web UI:
Try questions like:
- Are there any gaps in the load data?
- Which groups have the most duplicate timestamps?
- Show the hourly load profile for January 1
What's next¶
- Declare time series — write a
time_series.yamlfor your own data sodatasight qualitycan audit completeness. - Audit data quality — the full set of deterministic quality commands.