A warehouse before the product has users
Dashboards, metrics layers and the first dbt models get built months before there is real data to point them at. The usual stand-in is uniform noise, which renders every chart as a flat line and hides every bug that depends on growth, mix or seasonality. Declare the business you expect instead.
Who: founders, first data hires, analytics leads. Time: about 2 seconds for 291,000 rows across four tables.
The declaration#
Revenue grows from 180k to 344k a month, starter, growth and enterprise plans bring exactly 20%, 45% and 35% of it, and refunds fall from 4% to 2% over the year:
# misata.yaml
name: saas_warehouse
seed: 2025
tables:
accounts:
rows: 2000
columns:
account_id: {type: int, unique: true, min: 1, max: 2000}
company: {type: text, text_type: company}
region: {type: categorical, choices: [amer, emea, apac], probabilities: [0.5, 0.32, 0.18]}
signed_up_at: {type: date, start: "2024-01-01", end: "2025-12-31"}
users:
rows: 9000
columns:
user_id: {type: int, unique: true, min: 1, max: 9000}
account_id: {type: foreign_key}
email: {type: email, unique: true}
role: {type: categorical, choices: [owner, admin, member], probabilities: [0.2, 0.2, 0.6]}
invoices:
rows: 30000
columns:
invoice_id: {type: int, unique: true, min: 1, max: 10000000}
account_id: {type: foreign_key}
plan: {type: categorical, choices: [starter, growth, enterprise]}
amount: {type: float, min: 20, max: 4000}
is_refunded: {type: boolean}
issued_on: {type: date, start: "2025-01-01", end: "2025-12-31"}
events:
rows: 250000
columns:
event_id: {type: int, unique: true, min: 1, max: 100000000}
user_id: {type: foreign_key}
event_type: {type: categorical, choices: [page_view, search, export, invite], probabilities: [0.7, 0.18, 0.08, 0.04]}
occurred_at: {type: datetime, start: "2025-01-01", end: "2025-12-31"}
relationships:
- accounts.account_id -> users.account_id
- accounts.account_id -> invoices.account_id
- users.user_id -> events.user_id
outcome_curves:
- table: invoices
column: amount
time_column: issued_on
value_mode: absolute
start_date: "2025-01-01"
curve_points:
- {month: 1, target_value: 180000}
- {month: 2, target_value: 192000}
- {month: 3, target_value: 207000}
- {month: 4, target_value: 221000}
- {month: 5, target_value: 238000}
- {month: 6, target_value: 251000}
- {month: 7, target_value: 262000}
- {month: 8, target_value: 270000}
- {month: 9, target_value: 289000}
- {month: 10, target_value: 305000}
- {month: 11, target_value: 322000}
- {month: 12, target_value: 344000}
group_shares:
- table: invoices
measure: amount
group_column: plan
shares: {starter: 0.2, growth: 0.45, enterprise: 0.35}
rate_curves:
- table: invoices
column: is_refunded
time_column: issued_on
rate_points:
- {period: "2025-01", rate: 0.04}
- {period: "2025-06", rate: 0.03}
- {period: "2025-12", rate: 0.02}Show the numbers
| Month | Revenue |
|---|---|
| Jan | 180,000 |
| Feb | 192,000 |
| Mar | 207,000 |
| Apr | 221,000 |
| May | 238,000 |
| Jun | 251,000 |
| Jul | 262,000 |
| Aug | 270,000 |
| Sep | 289,000 |
| Oct | 305,000 |
| Nov | 322,000 |
| Dec | 344,000 |
Source: outcome_curves in the misata.yaml above; every month holds exactly
Run it#
pip install "misata[parquet,duckdb]"
misata explain misata.yaml
misata generate -c misata.yaml -o lake --format parquet --partition-by plan --attest
misata generate -c misata.yaml -o warehouse --format duckdblake/ holds one Parquet file per table, with invoices split Hive-style
into invoices/plan=starter/, plan=growth/ and plan=enterprise/.
warehouse/data.duckdb holds the same four tables in one file that any
DuckDB client, Evidence, Rill or a dbt-duckdb profile can open.
What a dashboard sees#
Querying lake/invoices with DuckDB:
plan revenue share
enterprise 1078350.0 0.35
growth 1386450.0 0.45
starter 616200.0 0.20
month revenue invoices refund_rate
2025-01 180000.0 1753 0.0399
2025-06 251000.0 2444 0.0299
2025-12 344000.0 3350 0.0200Row counts follow the revenue, because the curve plans how many invoices each month needs at a plausible average value. The refund rate is interpolated between the declared anchors and holds to the nearest invoice.
The check#
misata attest lake -s misata.yaml
python lake/verify.pyok 6 file(s) match their attested sha256
ok 80 of 80 invariant(s) holdThe 80 invariants are every monthly total and the grand total, the planned invoice count per month, every plan share in every month and overall, the refund rate at each anchor, three foreign keys, every unique key, and every declared category domain.
Related#
- Outcome curves: the monthly revenue declaration
- Rate curves: the refund rate between anchors
- Streaming large datasets: the bounded-memory path behind
--format - CLI reference:
--format,--partition-byand--attest - Publish to Hugging Face: share the attested folder
Files#
The recipe folder runs as it is. Its files on GitHub:
Prefer generating data visually?
Misata Studio lets you turn natural language prompts into relational schemas with exact outcome invariants, live preview tables, and one-click database seeding without writing code.