Five real recipe files — the same bare Recipe JSON recipe files
already documents, nothing app-specific about the format — each demonstrating a different real
capability. Open one from the desktop app (Open… in the toolbar, or drag it onto the app
icon) or run it from the command line:
sensym generate samples/student-roster.json --out /tmp/students --format csv
Every one of these was validated with sensym validate and actually generated for real before
being committed — see each recipe's own row below for what that run produced.
| Recipe | What it demonstrates | Rows |
|---|---|---|
student-roster.json |
A first, ordinary tabular dataset — identifiers, a composed full name, categorical and continuous columns. The same shape Getting started builds by hand. | 200 |
retail-orders.json |
A business/e-commerce table — money, categories, dates, order status. | 300 |
ad-spend-causal.json |
A declared causal structure: market_size confounds ad_spend (treatment) and revenue (outcome). Open it and switch to the Causal view to see the true ATE next to the naive regression estimate a real analyst would get wrong — see The causal graph view. |
500 |
student-roster.json + class-roster.csv |
The same student recipe run once per roster row via Cohort — one independently-seeded dataset per student, no roster identity entering the seed or the output. See Cohort generation. | 8 members |
support-tickets-narrated.json |
Multi-modal generation: a text column read aloud by the real neural TTS pipeline into real WAV files. Needs the Speech runtime installed first (Settings → Components in the desktop app; uv sync --extra speech for CLI/dev use). |
5 |
The cohort example#
# from the desktop app: click Cohort, pick student-roster.json and class-roster.csv
# from the command line, use sensym_df.io.roster / sensym_df.run directly -- see
# docs/cohort-generation.md's own "From the command line" section.
class-roster.csv uses opaque ids (student-01, student-02, …) rather than names, matching
this project's own privacy guarantee: no roster entry's real identity ever enters the seed, the
per-member output directory names, or any provenance.json.
Seeds#
Every recipe uses seed: 20260807 — deterministic, so re-running any of them on the same
machine and engine version reproduces byte-identical output, per this project's own
reproducibility guarantee.