Testing Reference

MockContext is an in-memory pipeline simulator that mirrors the engine's aggregation semantics. It lets you unit test datasets, pipelines, and extractors without running Postgres, Kafka, or any services.

Quick start

from thyme.testing import MockContext

ctx = MockContext()

# Ingest events
violations = ctx.add_events(Review, [
    {"restaurant_id": "r1", "rating": 4.8, "timestamp": "2026-03-15T10:00:00Z"},
    {"restaurant_id": "r1", "rating": 4.2, "timestamp": "2026-03-15T11:00:00Z"},
])

# Check raw aggregates
result = ctx.get_aggregates(RestaurantRatingStats, "r1")
assert result["avg_rating_7d"] == 4.5

# Query features through extractors
features = ctx.query(RestaurantFeatures, "r1")
assert features["avg_rating_7d"] == 4.5

API

`MockContext()`

Creates a new in-memory context. No arguments.

ctx = MockContext()

`ctx.add_events(dataset_class, events) -> list[ExpectationViolation]`

Ingests events for a dataset and processes them through all registered pipelines.

Parameter	Type	Description
`dataset_class`	`type`	The `@dataset`-decorated class
`events`	`list[dict]`	List of event dicts matching the dataset schema

Returns: A list of ExpectationViolation objects (empty if no expectations are defined or all pass). Events are still processed even when violations occur - expectations are observational.

Event format: Each event is a dict with keys matching the dataset field names. Timestamps can be datetime objects, ISO 8601 strings, or numeric epoch seconds.

violations = ctx.add_events(Order, [
    {"user_id": "u1", "amount": 42.50, "timestamp": "2026-03-15T10:00:00Z"},
    {"user_id": "u1", "amount": 15.00, "timestamp": "2026-03-15T11:00:00Z"},
])

`ctx.get_aggregates(dataset_class, entity_id) -> dict[str, Any]`

Returns raw aggregated values for an entity from all pipelines on the dataset.

Parameter	Type	Description
`dataset_class`	`type`	The output `@dataset` class (where the pipeline is defined)
`entity_id`	`str`	The entity key value

Returns: A {output_field: value} dict. Missing entities return zeros.

result = ctx.get_aggregates(UserOrderStats, "u1")
# {"order_count_1h": 2.0, "total_spend_24h": 57.5, ...}

`ctx.query(featureset_class, entity_id) -> dict[str, Any]`

Queries features for an entity, running extractors in dependency order.

Parameter	Type	Description
`featureset_class`	`type`	The `@featureset`-decorated class
`entity_id`	`str`	The entity key value

Returns: A {feature_name: value} dict with all computed features.

features = ctx.query(FraudSignals, "u_fraud")
# {"user_id": "u_fraud", "order_count_1h": 6, "is_suspicious": True, ...}

The query method:

Seeds features from pipeline aggregates (for extractors with deps)
Runs pure-transform extractors (no deps) with the seeded values

Full example

from thyme.testing import MockContext

def test_fraud_detection():
    # Given: a mock context with events
    ctx = MockContext()
    ctx.add_events(Order, [
        {"user_id": "u1", "order_id": "o1", "amount": 100.0,
         "item_count": 2, "timestamp": "2026-03-15T10:00:00Z"},
        {"user_id": "u1", "order_id": "o2", "amount": 200.0,
         "item_count": 1, "timestamp": "2026-03-15T10:05:00Z"},
    ])

    # When: querying aggregates
    stats = ctx.get_aggregates(UserOrderStats, "u1")

    # Then: aggregations are correct
    assert stats["order_count_1h"] == 2.0
    assert stats["total_spend_24h"] == 300.0

    # And: features are computed through extractors
    features = ctx.query(FraudSignals, "u1")
    assert features["is_suspicious"] == False

Supported aggregations

MockContext supports all 6 aggregation operators:

Operator	MockContext behavior
`Count`	Counts events in window
`Sum`	Sums field values in window
`Avg`	Mean of field values in window (0.0 if empty)
`Min`	Minimum field value in window
`Max`	Maximum field value in window
`ApproxPercentile`	Exact percentile rank (not approximate - good enough for tests)

The ApproxPercentile operator uses exact percentile rank computation in tests, which is deterministic and sufficient for correctness verification.

Testing Reference

On this page